Worldviews,

Science and Us
Philosophy and Complexity
This page intentionally left blank
 editors

Carlos Gershenson
Brussels Free University, Belgium

Diederik Aerts
Brussels Free University, Belgium

Bruce Edmotads
Manchester Metropolitan University Business School, UK

University o f Liverpool, Ul< 11- 14 September 2005

Worldviews,I 4 d

Science and Us 5

Philosophy and Complexity

World Scientific
NEW JERSEY * LONDON SINGAPORE * BElJlNG - SHANGHAI * HONG KONG TAIPEI * CHENNAI
Published by
World Scientific Publishing Co. Re. Ltd.
5 Toh Tuck Link, Singapore 596224
USA ofice: 27 Warren Street, Suite 401-402, Hackensack, NJ 07601
UK ofice: 57 Shelton Street, Covent Garden, London WC2H 9HE

British Library Cataloguing-in-PublicationData

A catalogue record for this book is available from the British Library.

WORLDVIEWS, SCIENCE AND US

Philosophy and Complexity
Copyright 0 2007 by World Scientific Publishing Co. Re. Ltd.
All rights reserved. This book, or parts thereoj may not be reproduced in any form or by any means,
electronic or mechanical, including photocopying, recording or any information storage and retrieval
system now known or to be invented, without written permissionfrom the Publisher.

For photocopying of material in this volume, please pay a copying fee through the Copyright
Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA. In this case permission to
photocopy is not required from the publisher.

ISBN-13 978-981-270-548-8
ISBN-10 981-270-548-1

Printed in Singapore by B & JO Enterprise

 CONTENTS

Introduction
Carlos Gershenson, Diederik Aerts and Bruce Edmonds 1

Restricted Complexi$y, General Complexity

Edgar Morin 5

Complexity Science as an Aspect of the Complexity of Science

Don C. Mikulecky 30

On the Importance of a Certain Slowness

Paul Cilliers 53

Simplicity is Not Truth-Indicative

Bruce Edmonds 65

Can the Whole be More than the Computation of the Parts? A

Reflection on Emergence
Camilo Olaya 81

Why Diachronically Emergent Properties Must Also Be Salient

Cyrille Imbert 99

On the Relativity of Recognising the Products of Emergence and

the Nature of the Hierarchy of Physical Matter
Kurt A . Richardson 117

Truth in Complex Adaptive Systems Models should be Based on

Proof by Constructive Verification
David Shipworth 141

Complexity as an Epistemic Revolution: Considerations on the

New Science in the Context of Western Intellectual History
Damian Popolo 156

V
 vi

Metaphors and Method: Epistemological Considerations on

Complexity Science
Rodrigo Zeidan and Maria Fonseca 173

Some Problems for an Ontology of Complexity

Michael McGuire 181

How to Love the Bomb - Trying to Solve the Prisoner’s

Dilemma with Evolutionary Game Theory
Vasco Castela 203

Physical Complexity and Cognitive Evolution

Peter Jedlicka 221

Informational Dynamic Systems: Autonomy, Information,

Function
Walter Riofrio 232

Grasping the Complexity of Living Systems Through Integrative

Levels and Hierarchies
J. M. Siqueiros and Jon Umerez 250

Simulation as Formal and Generative Social Science: The Very

Idea
Nuno David, Jaime S i m i o Sachman and Helder Coelho 266

A Compromise Between Reductionism and Non-Reductionism

Eray Ozkural 285

The Complexity of Information-Processing Tasks in Vision

John Symons 300

On the Possible Computational Power of the Human Mind

Hector Zenil and Francisco Hernandez- Quiroz 315

How Does Complex Mathemical Theory Arise? Phylogenetic and

Cultural Origins of Algebra?
Helen De Cruz 338
 INTRODUCTION

CARLOS GERSHENSON~,DIEDERIK AERTS~,AND BRUCE EDMONDS~

Centrum Leo Apostel, Vrije Universiteit Brussel
Centre for Policy Modelling, Manchester Metropolitan University

Scientific, technological, and cultural changes have always had an im-

pact upon philosophy. They can force a change in the way we perceive
the world, reveal new kinds of phenomena to be understood and provide
new ways of understanding phenomena. Complexity science, immersed in
a culture of information, is having a diverse but significant impact upon
philosophy’. It has highlighted classes of phenomena composed of mul-
tiply interacting parts where there may be no simple adequate analytic
model, so that new ways of modelling and understanding these phenom-
ena become necessary. In turn these new methods, based upon complex
simulation, have brought new phenomena into the domain of science by
making possible their formal representation. The result has been a rapid
co-development of new domains and new methods that suggests one of the
“paradigm” shifts described by Kuhn2. Previous ideas do not necessarily
sit comfortably with the new paradigm, resulting in new ideas or new in-
terpretations of old ideas, that is to say the “knock-on” effect is a larger
evolution of our worldviews.
In this book, researchers from different backgrounds join efforts to up-
date thinking upon philosophical questions with developments in the sci-
entific study of complex systems. The contributions are quite diverse, in
the sense that they focus on a wide range of topics. However, they all
share the common goal of increasing our understandings and improving
our descriptions of our complex world.
Of course, the background to this is that there are many that hold world-
views that are essentially untouched by complexity science. To take an
example there is the view that our world only appears complex but would,
if we could only understand it in the right way, be revealed as essentially
simple. Whether the world is simple or complex the fact is that, as humans,
there are severe limits to the complexity which we can comprehend. Thus,

1
2

regardless of how complex the world isa, there is an inescapable tension be-
tween the relative simplicity of human cognition and the apparent (or real)
complexity of the phenomena they are trying to understand. The different
philosophies and approaches can, to a large extent, be seen as responses
to this tension. Thus some seek to show that this tension is not feasibly
soluble, which may have consequences for how we think about or manage
the world we inhabit. Some seek to explain productive ways forward-how
one might explain and understand the complexity we observe. Others take
a step back to examine what consequences the tension itself has upon the
scientific method.
This volume is the fourth in a series entitled ‘Worldviews, Science and
Us’ published by World Scientific. The series collects interdisciplinary ar-
ticles with the global aim of stimulating new insights about the world and
the place of human in this world, and also about the role science plays in
the construction of new aspects of worldviews.
Many chapters in this volume are derived from presentations given at
the Philosophy and Complexity session of the Complexity, Science and So-
ciety conference, held in Liverpool, UK, between September llthand 14th
2005. We appreciate the work of the program committee for that session,
which was able to select relevant works from all the submissions received.
The program committee members were: William Bechtel, Mark Bedau,
Jacques Dubucs, Bruce Edmonds, Carlos Gershenson, Francis Heylighen,
Alicia Juarrero, Michael Lissack, Chris Lucas, Edgar Morin, Robert Pen-
nock, Kurt Richardson, John Symons, Pedro Sotolongo, Jean Paul Van
Bendegem, Franz Wuketits, and Roger Young. We would also like to thank
Robert Geyer, Jan Bogg, and Abbie Badcock for their organizational sup-
port at the conference. Clem6nt Vidal helped in polishing the translation of
Edgar Morinls chapter. Finally the speakers and participants of the session
and the quality of their discussions motivated us to compile this volume.
After the conference, the participants were invited to prepare a
manuscript and more people were invited to contribute to the volume. An
internal review process followed, where contributors reviewed other submis-
sions. From these, the actual chapters were selected to be included in the
volume.
In what follows, summaries of the chapters are given. They are diverse
enough to make it difficult to classify in different topic areas, so the order
is simply dictated by similarity.

aThis is obviously not a question that can itself be settled in a scientific way.
 3

From a philosophical perspective, Edgar Morin reviews the historical

development of complexity, noting that the current accepted notion of com-
plexity is restricted. Morin aims for a generalized complexity, which implies
an integrative change in our way of thinking and knowing, in education and
science. Don Mikulecky discusses similar issues from a scientific perspective,
putting an accent on scientific modeling.
Paul Cilliers notes the importance of a certain slowness in modern life,
since current practices seem to promote a cult of speed, where faster is bet-
ter. However, this is far from being the case, as proper adaptation requires
certain slowness to maintain the integrity of a system. In a similar spirit,
Bruce Edmonds refutes Occam’s razor, arguing that there is no reason to
suppose that a simpler theory is more likely to be true.
Emergence is discussed by several authors. Camilo Olaya reflects on
the use of rule-based modeling and process thought to understand complex
systems. Cyrille Imbert develops the notion of salience as a requirement
for diachronically emergent properties. Kurt Richardson explores the onto-
logical status of emergent products. David Shipworth proposes the use of
proof by constructive verification to determine truth in complex adaptive
systems, since a realist truth model of emergent properties is not feasible,
thus favoring epistemology over ontology.
Damian Popolo discusses the impact of complexity in science as an epis-
temic revolution, while Rodrigo Zeidan and Maria Fonseca present episte-
mological considerations on complexity science. Michael McGuire discusses
ontological problems of complexity.
Vasco Castela uses evolutionary game theory to show the importance
of the role of emotions in the evolution of cooperation and social modeling.
Peter Jedlicka studies the evolutionary increase of physical complexity and
its relationship with cognition.
Walter Riofrio proposes ‘Informational Dynamic Systems’ to study the
pre-biotic world, explaining naturalistically the physical emergence of func-
tions and information. Jeszis Sigueiros and Jon Umerez present a concep-
tual overview of different perspectives that were developed during the 20th
century to understand biological complexity.
Nuno David, Jaime S i m i o Sichman and Helder Coelho propose the use
of intentional knowledge to successfully simulate social systems, as opposed
to formal and empirical perspectives.
Eray Ozkural uses algorithmic information theory to propose a ‘mild’
non-reductionism, with implications for philosophy of mind. John Symons
questions the use of the information-processingmodel of the mind, focussing
4

on vision.
Hkctor Zenil and Francisco Herna’ndez Quiroz discuss the possibility
of using artificial neural network models to characterize the computational
power of the human mind.
Helen De Cruz argues for an active externalism as a requirement for the
emergence of algebra, studying its emergence in different cultural settings.

References
1. Heylighen, F., P. Cilliers, and C. Gershenson (forthcoming 2007). “Complex-
ity and Philosophy”. In Bogg, J. and R. Geyer (eds.) Complexity, Science
and Society. Radcliffe Publishing, Oxford.
2. Kuhn, Thomas S. The Structure of Scientific Revolutions. University of
Chicago Press, 2nd Ed.
 RESTRICTED COMPLEXITY, GENERAL COMPLEXITY*

EDGAR MORIN
CNRS Emeritus Director
Centre d’Etudes Transdisciplinaires. Sociologie, Anthropologie, Histoire
Ecole des Hautes Etudes en Sciences Sociales

Why has the problematic of complexity appeared so late? And why

would it be justified?

1. The three principles of the rejection of complexity by

‘classical science’
Classical science rejected complexity in virtue of three fundamental ex-
planatory principles:

The principle of universal determinism, illustrated by Laplace’s

Daemon, capable, thanks to his intelligence and extremely devel-
oped senses, of not only knowing all past events, but also of pre-
dicting all events in the future.
The principle of reduction, that consists in knowing any composite
from only the knowledge of its basic constituting elements.
The principle of disjunction, that consists in isolating and separating
cognitive difficulties from one another, leading to the separation
between disciplines, which have become hermetic from each other.

These principles led to extremely brilliant, important, and positive de-

velopments of scientific knowledge up to the point where the limits of in-
telligibility which they constituted became more important than their elu-
cidations.
In this scientific conception, the notion of “complexity” is absolutely
rejected. On the one hand, it usually means confusion and uncertainty;

*Presented at the Colloquium “Intelligence de la complexit6 : Qpist6mologieet pragma-

tique”, Cerisy-La-Salle, France, June 26th, 2005”. Translated from French by Carlos
Gershenson

5
6

the expression “it is complex” in fact expresses the difficulty of giving a

definition or explanation. On the other hand, since the truth criterion of
classical science is expressed by simple laws and concepts, complexity relates
only to appearances that are superficial or illusory. Apparently, phenomena
arise in a confused and dubious manner, but the mission of science is to
search, behind those appearances, the hidden order that is the authentic
reality of the universe.
Certainly, western science is not alone in the search of the “true” reality
behind appearances; for Hinduism, the world of appearances, the m&y& is
illusory; and for Buddhism the samsara, the world of phenomena, is not
the ultimate reality. But the true reality, in the Hindu or Buddhist worlds,
is inexpressible and in extreme cases unknowable. Whereas, in classical
science, behind appearances, there is the impeccable and implacable order
of nature.
Finally, complexity is invisible in the disciplinary division of the real. In
fact, the first meaning of the word comes from the Latin complexus, which
means what is woven together. The peculiarity, not of the discipline in
itself, but of the discipline as it is conceived, non-communicating with the
other disciplines, closed to itself, naturally disintegrates complexity.
For all these reasons, it is understood why complexity was invisible or
illusory, and why the term was rejected deliberately.

2. Complexity: A first breach: irreversibility

However, a first breach is made within the scientific universe during the
nineteenth century; complexity would appear from it de facto before start-
ing to be recognized de jure.
Complexity would make its appearance de facto with the second law of
thermodynamics, which indicates that energy degrades into caloric form:
this principle lies within the scope of the irreversibility of time, while until
then physical laws were in principle reversible and that even in the concep-
tion of life, the fixism of species did not need time.
The important point here is not only the irruption of irreversibility, thus
time, but it is also the apparition of a disorder since heat is conceived as
the agitation of molecules; the disordered movement of each molecule is
unpredictable, except at a statistical scale where distribution laws can be
determined effectively.
The law of the irreversible growth of entropy has given place to multiple
speculations, and beyond the study of closed systems, a first reflection about
 7

the universe, where the second law leads toward dispersion, uniformity, and
thus towards death. This conception of the death of the universe, long ago
rejected, has appeared recently in cosmology, with the discovery of black
energy. This will lead to the dispersion of galaxies and would seem to
announce us that the universe tends to a generalized dispersion. As the
poet Eliot said: “the universe will die in a whisper” ...
Thus, the arrival of disorder, dispersion, disintegration, constituted a
fatal attack to the perfect, ordered, and determinist vision.
And many efforts will be needed-we are not there precisely because
it is against the reigning paradigm-to understand that the principle of
dispersion, which appears since the birth of the universe with this incred-
ible deflagration improperly named big bang, is combined with a contrary
principle of bonding and organization which is manifested in the creation
of nuclei, atoms, galaxies, stars, molecules, and life.

3. Interaction Order/Disorder/Organization
How is it that both phenomena are related?
This is what I tried to show in the first volume of La Mkthode (The
Method). We will need to associate the antagonist principles of order and
disorder, and associate them making another principle emerge that is the
one of organization.
Here is in fact a complex vision, which one has refused to consider during
a very long time, for one cannot conceive that disorder can be compatible
with order, and that organization can be related to disorder at all, being
antagonist to it.
At the same time than that of the universe, the implacable order of life
is altered. Lamarck introduces the idea of evolution, Darwin introduces
variation and competition as motors of evolution. Post-darwinism, if it
has, in certain cases, attenuated the radical character of the conflict, has
brought this other antinomy of order: chance, I would say even a vice of
chance. Within the neodarwinian conception, to avoid calling “creation”
or “invention” the new forms of living organization such as wings, eyes-
one is very afraid of the word ‘Linventionlland of the word ‘kreation”---one
has put chance at the prow. One can understand the rest of the fear of
creation because science rejects creationism, i.e. the idea that God is a
creator of living forms. But the reject of creationism finished in masking
the creativity that manifests itself in the history of life and in the history of
humanity. And, from the philosophical point of view, it is rather recently
8

that Bergson, and then in another way, Castoriadis, put at the centre of
their conception the idea of creation.
In addition, in the beginning of the twentieth century, microphysics in-
troduced a fundamental uncertainty in the universe of particles that ceases
to obey the conceptions of space and time characteristic of our universe
called macro-physic. How thus these two universes, that are the same, but
at a different scale, are compatible? One begins today to conceive that one
can pass, from the micro-physical universe to ours, since between them a
certain number of quantum elements are connected, in virtue of a process
called decoherence. But there remains this formidable logical and concep-
tual hiatus between the two physics.
Finally, at a very large scale-mega-physical-Einstein’s theory discov-
ers that space and time are related to one another, with the result that our
lived and perceived reality becomes only meso-physical, situated between
micro-physic reality and mega-physical reality

4. Chaos
All this made that the dogmas of classical science are reached, but de fucto:
although increasingly mummified, they remain.
Yet a certain number of strange terms would appear. For example, the
term “catastrophe”, suggested by RenC Thom to try to make intelligible
the discontinuous changes of form; then the fractalism of Mandelbrot; then
the physical theories of chaos, which distinguishes itself from the rest, since
today it is thought that the solar system, which seems to obey an abso-
lutely impeccable and measurable order with the most extreme precision,
considering its evolution in millions of years, is a chaotic system comprising
a dynamic instability modifying for example Earth’s rotation around itself
or around the Sun. A chaotic process may obey to deterministic initial
states, but these cannot be know exhaustively, and the interactions de-
veloped within this process alter any prevision. Negligible variations have
considerable consequences over large time scales. The word chaos, in these
physics, has a very limited meaning: that of apparent disorder and unpre-
dictability. Determinism is saved in principle, but it is inoperative since
one cannot know exhaustively the initial states. We are in fact, since the
original deflagration and forever, plunged in a chaotic universe.
 9

5. The emergence of the notion of complexity

However, complexity remained always unknown in physics, in biology, in
social sciences. Admittedly, after more than half a century, the word com-
plexity irrupted, but in a domain that also remained impermeable to the
human and social sciences, as well as to the natural sciences themselves.
It is at the bosom of a sort of nebulous spiral of mathematicians and engi-
neers where it emerged at about the same time, and became connected at
once, in the forties and fifties, with Information Theory, Cybernetics, and
General Systems Theory. Within this nebula, complexity will appear with
Ashby to define the degree of variety in a given system. The word appears,
but does not contaminate, since the new thinking remains pretty confined:
the contributions of Von Neumann, of Von Foerster will remain completely
ignored, and still remain in the disciplinary sciences closed on themselves.
One can also say that Chaitin’s definition of randomness as algorithmic in-
compressibility becomes applicable to complexity. consequently, the terms
chance, disorder, complexity tend to overlap one another and sometimes to
be confused.
There were breaches, but still not an opening.
This would come from the Santa Fe Institute (1984) where the word
will be essential to define dynamical systems with a very large number
of interactions and feedbacks, inside of which processes very difficult to
predict and control take pla,ce, as “complex systems”, where the classical
conception was unable to be considered.
Thus, the dogmas or paradigms of classical science began to be disputed.
The notion of emergence appeared. In “Chance and Necessity”, Jacques
Monod makes a great state of emergence, i.e. qualities and properties that
appear once the organization of a living system is constituted, qualities that
evidently do not exist when they are presented in isolation. This notion is
taken, here and there, more and more, but as a simple constatation without
being really questioned (whereas it is a conceptual bomb).
It is like this that it was arrived to the complexity I call “restricted”:
the word complexity is introduced in “complex systems theory”; in addi-
tion, here and there the idea of “sciences of complexity” was introduced,
encompassing the fractalist conception and chaos theory.
Restricted complexity spread rather recently, and after a decade in
France, many barriers have been jumped. Why? Because more and more a
theoretical vacuum was faced, because the ideas of chaos, fractals, disorder,
and uncertainty appeared, and it was necessary at this moment that the
10

word complexity would encompass them all. Only that this complexity is
restricted to systems which can be considered complex because empirically
they are presented in a multiplicity of interrelated processes, interdepen-
dent and retroactively associated. In fact, complexity is never questioned
nor thought epistemologically.
Here the epistemological cut between restricted and generalized com-
plexities appears because I think that any system, whatever it might be, is
complex by its own nature.
Restricted complexity made it possible important advances in formal-
ization, in the possibilities of modeling, which themselves favor interdis-
ciplinary potentialities. But one still remains within the epistemology of
classical science. When one searches for the “laws of complexity”, one still
attaches complexity as a kind of wagon behind the truth locomotive, that
which produces laws. A hybrid was formed between the principles of tradi-
tional science and the advances towards its hereafter. Actually, one avoids
the fundamental problem of complexity which is epistemological, cognitive,
paradigmatic. To some extent, one recognizes complexity, but by decom-
plexifying it. In this way, the breach is opened, then one tries to clog it:
the paradigm of classical science remains, only fissured.

6. Generalized complexity
But then, what is “generalized” complexity? It requires, I repeat, an epis-
temological rethinking, that is to say, bearing on the organization of knowl-
edge itself.
And it is a paradigmatic problem in the sense that I have defined
“paradigm”a. Since a paradigm of simplification controls classical science,
by imposing a principle of reduction and a principle of disjunction to any
knowledge, there should be a paradigm of complexity that would impose a
principle of distinction and a principle of conjunction.
In opposition to reduction, complexity requires that one tries to com-
prehend the relations between the whole and the parts. The knowledge
of the parts is not enough, the knowledge of the whole as a whole is not
enough, if one ignores its parts; one is thus brought to make a come and go
in loop to gather the knowledge of the whole and its parts. Thus, the prin-
ciple of reduction is substituted by a principle that conceives the relation
of whole-part mutual implication.

=Cf La Me‘thode 4 , Les ide‘es, p.211-238, Le Seuil, 1990

 11

The principle of disjunction, of separation (between objects, between

disciplines, between notions, between subject and object of knowledge),
should be substituted by a principle that maintains the distinction, but
that tries to establish the relation.
The principle of generalized determinism should be substituted by a
principle that conceives a relation between order, disorder, and organiza-
tion. Being of course that order does not mean only laws, but also stabili-
ties, regularities, organizing cycles, and that disorder is not only dispersion,
disintegration, it can also be blockage, collisions, irregularities.
Let us now take again the word of Weaver, from a text of 1948, to
which we often referred, who said: the XIXth century was the century of
disorganized complexity and the X X t h century must be that of organized
complexity.
When he said “disorganized complexity”, he thought of the irruption
of the second law of thermodynamics and its consequences. Organized
complexity means to our eyes that systems are themselves complex because
their organization supposes, comprises, or produces complexity.
Consequently, a major problem is the relation, inseparable (shown in La
Me‘thode 1) , between disorganized complexity and organized complexity.
Let us speak now about the three notions that are present, but to my
opinion not really thought of, in restricted complexity: the notions of sys-
tem, emergence, and organization.

7. System: It should be conceived that “any system is

complex”
What is a system? It is a relation between parts that can be very different
from one another and that constitute a whole at the same time organized,
organizing, and organizer.
Concerning this, the old formula is known that the whole is more than
the sum of its parts, because the addition of qualities or properties of the
parts is not enough to know those of the whole: new qualities or properties
appear, due to the organization of these parts in a whole, they are emergent.
But there is also a substractivity which I want to highlight, noticing
that the whole is not only more than the sum of its parts, but it is also less
that the sum of it parts.
Why?
Because a certain number of qualities and properties present in the parts
can be inhibited by the organization of the whole. Thus, even when each of
12

our cells contains the totality of our genetic inheritance, only a small part
of it is active, the rest being inhibited. In the human relation individual-
society, the possibilities of liberties (delinquent or criminal in the extreme)
inherent to each individual, will be inhibited by the organization of the
police, the laws, and the social order.
Consequently, as Pascal said, we should conceive the circular relation:
‘one cannot know the parts if the whole is not known, but one cannot know
the whole if the parts are not known’.
Thus, the notion of organization becomes capital, since it is through
organization of the parts in a whole that emergent qualities appear and
inhibited qualities disappearb.

8. Emergence of the notion of emergence

What is important in emergence is the fact that it is indeductible from
the qualities of the parts, and thus irreducible; it appears only parting
from the organization of the whole. This complexity is present in any
system, starting with H 2 0 , the water molecule which has a certain number
of qualities or properties that the hydrogen or oxygen separated do not
have, which have qualities that the water molecule does not have.
There is a recent number of the Science et Awenir journal devoted to
emergence; to relate emergence and organization, one wonders wether it is
a hidden force in nature, an intrinsic virtue.
From the discovery of the structure of the genetic inheritance in DNA,
where it appeared that life was constituted from physicochemical ingredi-
ents present in the material world, therefore from the moment that it is clear
that there is not a specifically living matter, a specifically living substance,
that there is no d a n vital in Bergson’s sense, but only the physicochem-
ical matter that with a certain degree of organizing complexity produces
qualities of the living-f which self-reproduction, self-reparation, as well
as a certain number of cognitive or informational aptitudes, as from this
moment, the vitalism is rejected, the reductionism should be rejected, and
it is the notion of emergence that takes a cardinal importance, since a
certain type of organizing complexity produces qualities specific of self-
organization.
The spirit (mens, mente) is an emergence. It is the relation brain-

bI develop the idea that organization consists of complexity in La Me‘thode 1, La nature

de la nature,p.94-151, Le Seuil, 1977.
‘Science and Future, a popular French journal (Translator’s Note)
 13

culture that produces as emergent psychic, mental qualities, with all that
involves language, consciousness, etc.
Reductionists are unable to conceive the reality of the spirit and want to
explain everything starting from the neurons. The spiritualists, incapable
of conceiving the emergence of the spirit starting from the relation brain-
culture, make from the brain at most a kind of television.

9. The complexity of organization

The notion of emergence is a capital notion, but it redirects to the problem
of organization, and it is organization which gives consistence to our uni-
verse. Why is there organization in the universe? We cannot answer this
question, but we can examine the nature of organization.
If we think already that there are problems of irreducibility, of inde-
ductibility, of complex relations between parts and whole, and if we think
moreover that a system is a unit composed of different parts, one is obliged
to unite the notion of unity and that of plurality or at least diversity. Then
we realize that it is necessary to arrive at a logical complexity, because we
should link concepts which normally repel each other logically, like unity
and diversity. And even chance and necessity, disorder and order, need to
be combined to conceive the genesis of physical organizations, as on the
plausible assumption where the carbon atom necessary to the creation of
life was constituted in a star former to our sun, by the meeting exactly
a t the same t i m e a b s o l u t e coincident-f three helium nuclei. Thus, in
stars where there are billions of interactions and meetings, chance made
these nuclei to meet, but when this chance occurs, it is necessary that a
carbon atom will be constituted.
You are obliged to connect all these disjoined notions in the under-
standing that was inculcated to us, unfortunately, since childhood, order,
disorder, organization.
We then manage to conceive what I have called the self-ecc-organization,
i.e. the living organization.

10. The self-eco-organization

The word self-organization had emerged and had been used as of the end
of the 50’s by mathematicians, engineers, cyberneticians, neurologists.
Three important conferences had been held on the topic of “self-
organizing systems”, but a paradoxical thing, the word had not bored in
biology, and was a marginal biologist, Henri Atlan, who retook this idea, in
14

a great intellectual isolation within his corporation, in the 70’s. Finally the
word emerged in the 8O’s-9O’s in Santa Fe as a new idea, whereas it existed
already for nearly half a century. But it is still not imposed in biology.
I call self-eco-organization to the living organization, according to the
idea that self-organization depends on its environment to draw energy and
information: indeed, as it constitutes an organization that works to main-
tain itself, it degrades energy by its work, therefore it must draw energy
from its environment. Moreover, it must seek its food and defend against
threats, thus it must comprise a minimum of cognitive capacities.
One arrives to what I call logically the complex of autonomy-
dependence. For a living being to be autonomous, it is necessary that
it depends on its environment on matter and energy, and also in knowl-
edge and information. The more autonomy will develop, the more multiple
dependencies will develop. The more my computer will allow me to have
an autonomous thought, the more it will depend on electricity, networks,
sociological and material constraints. One arrives then to a new complexity
to conceive living organization: the autonomy cannot be conceived without
its ecology. Moreover, it is necessary for us to see a self-generating and
self-producing process, that is to say, the idea of a recursive loop which
obliges us to break our classical ideas of product --f producer, and of cause
+ effect.

In a self-generating or self-producing or self-poetic or self-organizing

process, the products are necessary for their own production. We are the
products of a process of reproduction, but this process can continue only
if we, individuals, couple to continue the process. Society is the product
of interactions between human individuals, but society is constituted with
its emergencies, its culture, its language, which retroacts to the individuals
and thus produces them as individuals supplying them with language and
culture. We are products and producers. Causes produce effects that are
necessary for their own causation.
Already the loop idea had been released by Norbert Wiener in the idea of
feedback, negative as well as positive, finally mainly negative; then it was
generalized without really reflecting on the epistemological consequences
which it comprised. Even in the most banal example which is that of
a thermal system supplied with a boiler which provides the heating of a
building, we have this idea of inseparability of the cause and effect: thanks
to the thermostat, when 20” is reached, the heating stops; when the tem-
perature is too low, the heating is started. It is a circular system, where the
effect itself intervenes in the cause which allows the thermal autonomy of
 15

the whole compared to a cold environment. That is to say that the feedback
is a process which complexifies causality. But the consequences of this had
not been drawn to the epistemological level.
Thus feedback is already a complex concept, even in non-living systems.
Negative feedback is what makes it possible to cancel the deviations that
unceasingly tend to be formed like the fall in temperature compared to the
standard. Positive feedback develops when a regulation system is not able
anymore to cancel the deviations; those can then be amplified and go to-
wards a runaway, kind of generalized disintegration, which is often the case
in our physical world. But we could see, following an idea advanced more
than fifty years ago by Magoroh Maruyama, that the positive feedback, i.e.
increasing deviation, is an element that allows transformation in human his-
tory. All the great transformation processes started with deviations, such
as the monotheist deviation in a polytheist world, the religious deviation
of the message of Jesus within the Jewish world, then, deviation in the
deviation, its transformation by Paul within the Roman empire; deviation,
the message of Mohammed driven out of Mecca, taking refuge in Medina.
The birth of capitalism is itself deviating in a feudal world. The birth of
modern science is a deviating process from the XVIIth century. Socialism
is a deviating idea in the XIXth century. In other words, all the processes
start by deviations that, when they are not suffocated, exterminated, are
then able to make chain transformations.

11. The relationship between local and global

In logical complexity, you have the relation between the local and the global.
One believed to be able to assume the two truths of the global and
of the local with axioms of the style: “think globally and act locally”. In
reality, one is, I believe, constrained in our global age to think jointly locally
and globally and to try to act at the same time locally and globally. Also,
which is also complex, local truths can become global errors. For example,
when our immune system rejects with the greatest energy the heart that
one grafts to him, like a nasty foreigner, this local truth becomes a global
error, because the organism dies. But one can also say that global truths
can lead to local errors. The truth of the need to fight against terrorism
can lead to interventions, which will favor even more the development of
terrorism, just look at Irak.
16

12. Heraclitus: “live of death, die of life”

In this union of notions logically complex, there is a relationship between
life and death.
I often quoted the illuminating phrase of Heraclitus, from the VIth cen-
tury b.C.: “live of death, die of life”. It became recently intelligible, from
the moment when we learned that our organism degrades its energy, not
only to reconstitute its molecules, but that our cells themselves are de-
graded and that we produce new cells. We live from the death of our cells.
And this process of permanent regeneration, almost of permanent rejuve-
nilization, is the process of life. What makes it possible to add to the very
right formula of Bichat, saying: “life is the ensemble of the functions that
fight against death”, this strange complement that presents us a logical
complexity: “Integrating death to fight better against death”. What one
again knows about this process is extremely interesting: it has been learned
rather recently that cells that die are not only old cells; in fact apparently
healthy cells receiving different messages from neighboring cells, “decide”,
at a given moment, to commit suicide. They commit suicide and phago-
cytes devour their remains. Like this, the organism determines which cells
must die before they have reached senescence. That is to say that the
death of cells and their postmortem liquidation are included in the living
organization.
There is a kind of phenomenon of self-destruction, of apoptosis, since
this term has been taken from the vegetal world, indicating the split of the
stems operated by trees in autumn so that dead leafs fall.
On the one hand, when there is an insufficiency of cellular deaths fol-
lowing different accidents and perturbations, there are a certain number of
diseases that are deadly in the long run, like osteoporosis, various types of
sclerosis, and certain cancers, where cells refuse to die, becoming immortal,
forming tumors and go for a stroll in the form of metastases (It can seem
that it is a revolt of cells against their individual death that lead to these
forms of death of the organism). On the other hand, the excess of cellular
deaths determine AIDS, Parkinson’s, and Alzheimer’s disease.
You see at which point this relationship between life and death is com-
plex: it is necessary for cells to die, but not too much! One lives between
two catastrophes, the excess or insufficiency of mortality. One finds again
the fundamentally epistemological problem of generalized complexity.
 17

13. On non-trivial machines

Living beings are certainly machines, but unlike artificial machines that
are trivial deterministic machines (where one knows the outputs when one
knows the inputs), these are non-trivial machines (von Foerster) where one
can predict innovative behaviors.
We are machines, this truth was already in L'homme-machine of La
Mettrie. We are physical machines, thermal machines, we function at the
temperature of 37'. But we are complex machines.
Von Neumann established the difference between living machines and
artificial machines produced by technology: the components of the technical
machines, having the good quality of being extremely reliable, go towards
their degradation, towards their wear, from the very start of their opera-
tion. Whereas the living machine, made up mainly by components far from
reliable, degrading proteins-and one understands very well that this lack
of reliability of proteins makes it possible to reconstitute them non stop-is
able to be regenerated and repaired; it also goes towards death, but after
a process of development. The key of this difference lies in the capacity of
self-repair and self-regeneration. The word regeneration is capital here.
One can say that the characteristic of innovations that emerge in the
evolution of life (which are determined by environmental changes, or by the
irruption of multiple hazards), such as the appearance of the skeleton in
vertebrates, wings in insects, birds, or bats, all these creations, are char-
acteristic non-trivial machines. That is to say, it gives a new solution to
insurmountable challenges without this solution.
All the important figures of human history, on the intellectual, religious,
messianic, or politic levels, were non-trivial machines. One can advance that
all the History of Humankind, which begins ten thousand years ago-is a
non-trivial history, i.e. a history made of unforeseen, of unexpected events,
of destructions and creations. The history of life that precedes it is a non-
trivial history, and the history of the universe, where the birth of life and
then of humankind are included, is a non-trivial history.
We are obliged to detrivialize knowledge and our worldview.

14. To complexify the notion of chaos

We have seen how the notion of system brings us to complexities of orga-
nization which themselves lead us to logical complexities. Let us look now
at the notion of chaos, as it appears within chaos theory, and which com-
prises disorder and impredictibility. The beat of the wings of a butterfly
18

in Melbourne can cause by a succession of chain processes a hurricane in

Jamaica, for example.
Actually, I believe that the word chaos must be considered in its deep
sense, its Greek sense. We know that in the Greek worldview, Chaos is at
the origin of Cosmos. Chaos is not pure disorder, it carries within itself
the indistinctness between the potentialities of order, of disorder, and of
organization from which a cosmos will be born, which is an ordered universe.
The Greeks saw a bit too much order in the cosmos, which is effectively
ordered because the immediate spectacle, the impeccable order of the sky
that we see each night with the stars, is always in the same place. And if
the planets are mobile they also come to the same place with an impeccable
order. However, we know today with the widened conceptions of cosmic
time that all this order is at the same time temporary and partial in a
universe of movement, collision, transformation.
Chaos and Cosmos are associated-I have employed the word
Chaosmos-there is also a circular relation between both terms. It is neces-
sary to take the word chaos in a much deeper and more intense sense than
that of physical chaos theory.

15. T h e need of contextualization

Let us take again the “complexus” term in the sense of “what is woven
together”.
It is a very important word, which indicates that the breaking up of
knowledge prevents from linking.and contextualizing.
The knowledge mode characteristic of disciplinary science isolates ob-
jects, one from another, and isolates them compared to their environment.
One can even say that the principle of scientific experimentation allows
to take a physical body in Nature, to isolate it in an artificial and con-
trolled laboratory environment, and then study this object in function of
perturbations and variations that one makes it undergo. This indeed makes
it possible to know a certain number of its qualities and properties. But
one can also say that this principle of decontextualization was ill-fated, as
soon as it was ported to the living. The observation since 1960 by Jane
Goodall of a tribe of chimpanzees in their natural environment could show
the supremacy of observation (in a natural environment) over experimen-
tation (in a laboratory) for knowledged. A lot of patience was necessary

dSee “ L e Paradigme Perdu”, pp. 51-54

 19

so that Jane Goodall could perceive that chimpanzees had different per-
sonalities, with rather complex relations of friendship, of rivalry; a whole
psychology, a sociology of chimpanzees, invisible to the studies in a labora-
tory or in a cage, appeared in their complexity.
The idea of knowing the living in their environment became capital in
animal ethology. Let us repeat it, the autonomy of the living needs to be
known in its environment.
From now on, becoming aware of the degradations that our techno-
economic development makes to the biosphere, we realize the vital link
with this same biosphere that we believe to have reduced to the rank of
manipulable object. If we degrade it, we degrade ourselves, and if we
destroy it, we destroy ourselves.
The need for contextualization is extremely important. I would even say
that it is a principle of knowledge: Anybody who has made a translation
in a foreign language will seek an unknown word in the dictionary; but
with words being polysemous, it is not immediately known which is the
good translation; the sense of the word will be sought in the sense of the
sentence in the light of the global sense of the text. Though this play
from text to word, and from text to context, and from context to word, a
sense will crystalize. In other words, the insertion in the text and in the
context is an evident cognitive necessity. Take for example the economy,
the most advanced social science from a mathematical point of view, but
which is isolated from human, social, historic, and sociologic contexts: its
prediction power is extremely weak because the economy does not function
in isolation: its forecasts need to be unceasingly revised, which indicates us
the disability of a science that is very advanced but too closed.
More generally, mutual contextualization is lacking in the whole of social
sciences.
I have often quoted the case of the Aswan dam because it is revealing and
significant: it was built in Nasser’s Egypt because it would make it possible
to regulate the course of a capricious river, the Nile, and to produce electric
power for a country which had a great need for it. However, after some
time, what happened? This dam retained a part of the silts that fertilized
the Nile valley, which obliged the farming population to desert the fields
and overpopulate large metropolises like Cairo; it retained a part of the fish
that the residents ate; moreover today, the accumulation of silts weakens
the dam and causes new technical problems. That does not mean that the
Aswan dam should not have been built, but that all the decisions taken in a
techno-economic context are likely to be disastrous by their consequences.
20

It is like the deviation of rivers in Siberia that the Soviet government

made and where the perverse consequences are more important than the
positive ones.
It is thus necessary to recognize the inseparability of the separable,
at the historical and social levels, as it has been recognized at the mi-
crophysical level. According to quantum physics, confirmed by Aspect’s
experiments, two microphysical entities are immediately connected one to
the other although they are separated by space and time. Even more, one
arrives to the idea that everything that is separated is at the same time
inseparable.

16. The hologrammatic and dialogical principles

The hologrammic or hologrammatic principle should also be advanced, ac-
cording to which not only a part is inside a whole, but also the whole is
inside the part; just as the totality of the genetic inheritance is found in
each cell of our organism, the society with its culture is inside the spirit of
an individual.
We return again to the logical core of complexity which we will see,
is dialogical: separability-inseparability, whole-parts, effect-cause, product-
producer, life-death, homo sapiens-homo demens, etc.
It is here that the principle of the excluded middle reveals its limit.
The excluded middle states “A cannot be A and not A”, whereas it can
be one and the other. For example, Spinoza is Jewish and non-Jewish, he
is neither Jewish, nor non-Jewish. It is here that the dialogic is not the
response to these paradoxes, but the means of facing them, by considering
the complementarity of antagonisms and the productive play, sometimes
vital, of complementary antagonisms.

17. For the sciences, a certain number of consequences

Regarding sciences, we can see a certain number of consequences.
First of all, classical science is somehow complex, even when it produces
simplifying knowledge. Why?
Because science is a quadruped which walks on the following four legs:
the leg of empiricism made of data, experimentation or observation; the
leg of rationality, made of logically constituted theories; the leg of verifica-
tion, always necessary; and the leg of imagination, because great theories
are products of a powerful creative imagination. Thus science is complex,
produced by a quadruped movement, which prevents it from solidifying.
 21

The objective knowledge which is its idea, resulted in the need of elim-
inating subjectivity, i.e. the emotional part inherent to each observer, to
each scientist, but it also comprised the elimination of the subject, i.e. the
being which conceives and knows. However, any knowledge, including ob-
jective, is at the same time a cerebral translation starting from data of
the external world and a mental reconstruction, starting from certain or-
ganizing potentialities of the spirit. It is certain that the idea of a pure
objectivity is utopian. Scientific objectivity is produced by beings who are
subjects, within given historical conditions, starting from the rules of the
scientific game. The great contribution of Kant was to show that the ob-
ject of knowledge is co-constructed by our spirit. He indicated us that it
is necessary to know knowledge to know its possibilities and limits. The
knowledge of knowledge is a requirement of the complex thinking.
As Husserl indicated in the ~ O ’ S , in particular in his conference on
the crisis of European science, sciences developed extremely sophisticated
means to know external objects, but no means to know themselves. There
is no science of science, and even the science of science would be insufficient
if it does not include epistemological problems. Science is a tumultuous
building site, science is a process that could not be programmed in advance,
because one can never program what one will find, since the characteristic
of a discovery is its unexpectedness. This uncontrolled process has lead
today to the development of potentialities of destruction and of manipula-
tion, which must bring the introduction into science of a double conscience:
a conscience of itself, and an ethical conscience.
Also, I believe that it will be necessary to arrive more and more to
a scientific knowledge integrating the knowledge of the human spirit to
the knowledge of the object which this spirit seizes and recognizing the
inseparability between object and subject.

18. Two scientific revolutions introduced complexity de

facto
I already indicated how the concept of complexity emerged in a marginal
fashion in a sphere of mathematicians/engineers. It should be indicated
now that the X X t h century knew two scientific revolutions which de facto
introduced complexity without, however, recognizing this notion that re-
mains implicit.
The first revolution, after the thermodynamics of the XIXth century, is
that of the microphysics and cosmophysics that introduced indeterminism,
22

risk-where determinism reigned-and elaborated suitable methods to deal

with the uncertainties met.
The second revolution is that which gathers disciplines and restores be-
tween them a common fabric. It begins in the second half of the XXth
century. Thus in the ~ O ’ S , Earth sciences designed Earth as a complex
physical system, which makes it possible today to articulate geology, seis-
mology, vulcanology, meteorology, ecology, etc. At the same time, ecology
develops as a scientific knowledge bringing together data and information
coming from different physical and biological disciplines in the conception
of ecosystems. It makes it possible to conceive how an ecosystem either
degrades, develops, or maintains its homeostasis. From the ~ O ’ S , the eco-
logical conception extends to the whole biosphere, necessarily introducing
knowledge from the social sciences.
Although ecology, at the biosphere level, cannot make rigorous pre-
dictions, it can give us vital hypothesis, concerning, for example, global
warming, which manifests itself by the melting of glaciers in the Antarctic
or the Arctic. Thus ecology, cosmology, and Earth sciences have become
poly-disciplinary sciences, even transdisciplinary. Sooner or later, this will
arrive in biology, from the moment when the idea of self-organization will
be established; this will arrive in the social sciencese, although they are
extremely resistant.
Finally, the observer, chased by the objectivity postulate, was intrc-
duced into certain sciences, such as microphysics where the observer per-
turbs what it observes. In the case of cosmology, even if one does not adhere
to what Brandon Carter called the anthropic principle, which holds account
of the place of humans in the universe, one is obliged to conceive that this
universe, among its perhaps negligible possibilities, had the possibility of
human life, perhaps only on this planet Earth, but perhaps also elsewhere.
Thus, the common fabric between the human, the living, and the Uni-
verse can be restored, which implies a complex conception capable at the
same time to distinguish the human from the natural and to integrate it.

19. The insertion of science in History

In addition, there is the problem of the insertion of the Sciences in human
History.
You know that there are two conceptions of history of sciences, the in-
ternalist conception and the externalist conception. The internalist mode

‘Cf my Humanite‘ de l’humanite‘, La Me‘thode 5, Le Seuil.

 23

sees the development of sciences in isolation, only in function of their in-

ternal logic and their own discoveries. The externalist mode sees them in
function of historical and social developments which determine the scientific
developments.
I think that it is necessary to link both, and this is appropriate for other
developments than those of sciences. Thus, some wanted to understand the
perversion of the Soviet Union starting from internal factors, such as in-
sufficiencies of the Marxist doctrine, limitations of that of Lenin. Others
wanted to impute it to external elements such as the surrounding and hos-
tility of the capitalist powers with regard to the Soviet Union or former
elements such as the backwardness of tsarist Russia. Whereas the true
cognitive game is to link these two aspects in a dialogical fashion.
If one continues to place oneself from the viewpoint of modern West-
ern history of science, one sees how from its marginal and quasi-deviating
birth in the XVIIth century, it is developed in the XVIIIth, introduced
in universities in the XIXth, then in states and companies in the XXth,
and how it becomes central and driving within human history in the form
of techno-science, and produces not only all the major elements for a re-
newed knowledge of the world and beneficial effects for humanity, but also
formidable and uncontrolled powers which threaten it.
I don’t know if I am right or wrong in retaking an expression of Vico,
but it is necessary for us to arrive to the “Scienza Nuova”. Very precisely,
Vico inscribed the historical perspective at the heart of the scienza nuova.
It is necessary to amplify the idea of scienza nuova by introducing the
interaction between the simple and the complex, by conceiving a science
that does not suppress disciplines but connects them, and consequently
makes them fertile, a science which can at the same time distinguish and
connect and where the transdisciplinarity is inseparable from complexity.
I repeat it, as much as the compartmentalization of disciplines disinte-
grates the natural fabric of complexity, as much a transdisciplinary vision
is capable of restoring it.

20. The link between science and philosophy

The link between science and philosophy has been broken. Still in the
XVIIth century, the great scientists were at the same time great philoso-
phers. Certainly, they did not identify Science and Philosophy. When
Pascal made his experiments in Puy de DGme, he did not think about the
bet problem. But in the times of Pascal, Gassendi, Leibniz, there was not
24

this cut. This became a frightening ditch. The ditch of ignorance separates
the scientific culture from the culture of the humanities.
But the current has started to be reversed: the most advanced sciences
arrive to fundamental philosophical problems: Why is there a universe out
of nothing? How was this universe born from a vacuum which was not at
the same time the vacuum? What is reality? Is the essence of the universe
veiled or totally cognizable?
The problem of life is posed from now on in a complexity that exceeds
biology: the singular conditions of its origin, the conditions of emergences
of its creative powers. Bergson was mistaken by thinking that there was an
e'lan vital, but was right while speaking about creative evolution. He could
even have spoken about evolutionary creativity.
Today we can foresee the possibility of creating life. From the moment
when it is believed that life is a process developed starting only from physic-
ochemical matter under certain conditions, in underwater thermal vents or
elsewhere, one can very well consider creating the physical, chemical, ther-
modynamic conditions which give birth to organisms gifted with qualities
that one calls life. We can also foresee the possibility to modify the human
being in its biological nature. Therefore, we have to meditate about life,
as we never did it. And at the same time we must meditate about our
relationship with the biosphere.
Thus all the most advanced sciences arrive to fundamental philosophical
problems that they thought to have eliminated. They do not only find them,
they renew them.
If one defines philosophy by the will and capacity of reflection, it is
necessary that the reflectivity is also introduced into the sciences, which
does not eliminate the relative autonomy of philosophy nor the relative
autonomy of scientific procedures compared to philosophical procedures.
Finally and especially, any knowledge, including the scientific one, must
comprise in itself an epistemological reflection on its foundations, principles,
and limits.
Still today there is the illusion that complexity is a philosophical prob-
lem and not a scientific one. In a certain way, it is true, in a certain way,
it is false. It is true when you place yourselves from the point of view of an
isolated and separated object: the fact that you isolate and separate the
object made the complexity to disappear: thus it is not a scientific problem
from the point of view of a closed discipline and a decontextualized object.
But, as soon as you start to connect these isolated objects, you are in front
of the problem of complexity.
 25

21. Second epistemological rupture with restricted

complexity
It is here that a second epistemological rupture with restricted complexity
appears.
Restricted complexity is interested essentially in dynamical systems
called complex. That is to say, it constitutes its own field, within the
field of sciences.
But generalized complexity not only concerns all fields, but also relates
to our knowledge as human beings, individuals, persons, and citizens. Since
we have been domesticated by our education which taught us much more
to separate than to connect, our aptitude for connecting is underdeveloped
and our aptitude for separating is overdeveloped; I repeat that knowing,
is at the same time separating and connecting, it is to make analysis and
synthesis. Both are inseparable, and our atrophy of the capacity to connect
is increasingly serious in a globalized, complexified mode, where it is a
matter of generalized interdependence of everything and everyone.
The International Ethical, Political and Scientific Collegium has formu-
lated a declaration of interdependence which it would wish to see promul-
gated by the United Nations. We must think the interdependence in all
fields, including the complex relation between the parts and the whole. We
need to be able to face uncertainties of life whereas nothing prepares us for
it. We need to face complexity, including for action, whereas one opposes
the cautionary principle to the risk principle, while Pericles had truly ex-
pressed the union of the two antagonistic principles when he said during
a speech to the Athenians during the Peloponnesian war: “we Athenians,
we are capable of combining prudence and audacity, whereas the others
are either timorous or bold”. It is the combination which we need. Also,
precaution needs today sometimes much invention.
We need to deeply reform all our way of knowing and thinking.

22. The principle of ecology of action

The principle of ecology of action is, in my opinion, central: from the
moment an action enters a given environment, it escapes from the will
and intention of that which created it, it enters a set of interactions and
multiple feedbacks and then it will find itself derived from its finalities, and
sometimes to even go in the opposite sense. The ecology of action has a
universal value, including for the development of sciences, whose destructive
nuclear consequences were absolutely unexpected.
26

Think that when Fermi elucidated the structure of the atom in the ~ O ’ S ,
it was a purely speculative discovery and he had by no means thought that
this could allow the fabrication of an atomic bomb. However, a few years
later, the same Fermi went to the United States to contribute to the fabri-
cation of the atomic bomb that would be used in Hiroshima and Nagasaki.
When Watson and Crick determined the structure of the genetic inheritance
in DNA, they thought that it was a great conquest of knowledge without
any practical consequences. And hardly ten years after their discovery, the
problem of genetic manipulations was posed in the biology community.
The ecology of action has a universal value. One can think of examples
in our recent French history: a dissolution of the Parliament by President
Chirac to have a governmental majority led to a socialist majority; a refer-
endum made to win general support led to its rejection. Gorbachev tried a
reform to save the Soviet Union but this one contributed to its disintegra-
tion. When one sees that a revolution was made in 1917 to suppress the
exploitation of man by his fellow man, to create a new society, founded on
the principles of community and liberty, and that this revolution, not only
caused immense losses of blood, destruction, and repression by a police sys-
tem, but, after seventy years, it led to its contrary, i.e. to a capitalism even
more fierce and savage than that of the tsarist times, and with a return
of religion! Everything that this revolution wanted to destroy resurrected.
How not to think about the ecology of action!

23. Creating “Institutes of fundamental culture”

The reform of the spirit seems to me absolutely necessary.
Once that I had understood that the reform of thought, deep work
that I carried out in La Me‘thode, is a necessity, I accepted the offer of a
Minister of Education when he called me for the reform of the content of
secondary education. I tried to introduce my ideas of reform of thought
into an educational project. I saw its total failure-finally it did not failed,
it was not applied!-That pushed me to reflect even more. I wrote a book
called La TCte bien fuite (The head well made), then on the initiative of
UNESCO I made a book called Les Sept savoirs ne‘cessaires a l’e‘ducation
du f u t u r (The seven knowledges necessary in the education of the future).
Following a University which will be created on these principles in Mex-
ico, I had the more restricted but maybe more necessary idea of creating
“Institutes of fundamental culture”, which would be sheltered in a Univer-
sity or independent, addressing everybody, i.e. before University or during
 27

University or after University, students, citizens, members of trade unions,

entrepreneurs, everybody.
Why the word “fundamental culture”? Because it is that what is miss-
ing. In fact, it is the most vital matter to be taught, the most important
to face life, and which is ignored by education.

(1) Knowledge as a source of error or illusion; nowhere the traps of

knowledge are taught, which come owing to the fact that all knowl-
edge is translation and reconstruction.
(2) Rationality, as if it were an obvious thing, whereas we know that
rationality knows its perversion, its infantile or senile diseases.
(3) Scientificity. What is science, its frontiers, its limits, its possibilities,
its rules. Moreover, there is an abundant literature, but which has
never been consulted by the scientists who are recruited at CNRS
for example. Most of the time, they do not know anything about
the polemic between Niels Bohr and Einstein, the works of Popper,
Lakatos, Kuhn, etc.
(4) What is complexity.
And also:

A teaching on “what is the human identity and condition”, which

is not found anywhere.
A teaching on the global age, not only today’s globalization, but all
its antecedents starting from the conquest of America, the coloniza-
tion of the world, its current phase, and its future prospects.
A teaching on human understanding.
A teaching concerning the confrontation of uncertainties in all the
fields: sciences, everyday life, history (we have lost the certainty of
progress, and the future is completely uncertain and obscure).
A teaching on the problems of our civilization.
That is for me the fundamental teaching that can aid the reform of the
spirit, of thought, of knowledge, of action, of life.

24. I conclude: generalized complexity integrates restricted

complexity
Unfortunately, restricted complexity rejects generalized complexity, which
seems to the former as pure chattering, pure philosophy. It rejects it because
restricted complexity did not make the epistemological and paradigmatic
28

revolution which complexity obliges. That will be done without a doubt.

But in the meantime, we see that the problematic of complexity have in-
vaded all our horizons, and I repeat “problematic”, because it is an error
to think that one will find in complexity a method that can be applied
automatically to the world and anything.
Complexity is a certain number of principles which help the autonomous
spirit to know. Whereas a program destroys the autonomy of the one who
seeks, the problematic of complexity stimulates an autonomous strategy,
obliges in the field of action-once that one knows that the ecology of
action can pervert the best intentions-to reconsider our decisions like bets
and incites us to develop an adequate strategy to more or less control the
action.
In other words, in all the fields, I would say “help yourself and the
complexity will help you”, which has nothing to do with the mechanical
application of a program or a rule. It is a deep reform of our mental
functioning, of our being.
These ideas now marginal, deviating, begin to constitute a tendency
still in minority, or rather tendencies since there are several paths to go
towards complexity. These ideas, these deviations, can be developed and
become cultural, political, and social forces.
The probabilities of a global future are extremely alarming: our space-
ship is pulled by four engines without any control: science, technology,
economy, and the search for profit-all this under conditions of cham since
the techno-civilizational unification of the planet, under the Western push,
causes singular cultural resistances and cultural and religious re-closings.
The planet is in crisis with all the possibilities, ones regressive and
destructive, others stimulant and fertile, such as invention, creation, new
solutions.

25. We should even apprehend the possibilities of

metamorphosis
We should even apprehend the possibilities of metamorphosis because we
have completely astonishing examples of it from the past. The change in
certain places where there have been demographic concentrations in the
Middle East, in the Indus basin, in China, in Mexico, in Peru, from pre-
historic societies of hundreds of men, without cities, without state, without
agriculture, without army, without social class, to enormous historical soci-
eties with cities, agriculture, army, civilization, religion, philosophy, works
of art ... that constituted a sociological metamorphosis.
Perhaps we are going towards a meta-historical metamorphosis suitable
for the birth of a society-world at a global scale.
I would say that complexity does not put us only in the distress of the
uncertain, it allows us to see besides the probable, the possibilities of the
improbable, because of those which have been in the past and those that
can be found again in the future.
We are in an epoch of doubtful and uncertain combat.
That makes one think of the Pacific war, after the Japanese had broken
into the Pacific Islands and had begun to threaten California, there was
a gigantic naval fight over 200 kilometers along the Midways between the
Japanese and American fleets: battleships, aircraft carriers, submarines,
planes. The global vision was impossible for both of them: there were
sunken Japanese ships, sunken American ships, planes that did not find
the enemy fleet; in short, total confusion, the battle divided in several
fragments. At a given moment, the Japanese Admiral realizing his losses
in battleships and planes, thought that they were defeated, thus called
for retreat. But the Americans, who had lost as much, were not the first
to think that they were defeated; after the Japanese retreat, they were
victorious.
Well, the outcome of what will happen, we cannot conceive it yet! We
can always hope and act in the direction of this hope.
The intelligence of complexity, isn’t it to explore the field of possibilities,
without restricting it with what is formally probable? Doesn’t it invite us
to reform, even to revolutionize?
 COMPLEXITY SCIENCE AS AN ASPECT OF THE
COMPLEXITY OF SCIENCE

DONALD C. MIKULECKY
Senior Fellow, Virginia Commonwealth University’s Center for the Study of
Biological Complexity, Richmond, VA

1. INTRODUCTION
How can we treat science as an object of scientific inquiry? The central
problem arises with that question. Science has tried to rid itself of circu-
larity and in so doing has become a very limited method for examining the
complex world it tries to have us understand. Self reference is at the center
of so many of the interesting thing we want to understand including and
especially life itself. The existence of this self referential character is the
essence of what we have come to call “complexity”. The works of Robert
Rosen [l, 2, 31 spell this out in great detail. This series of investigations
began over a half century ago yet still remains virtually unrecognized by the
vast majority of those who call themselves “scientists”. That fact alone can
be a springboard to launch a study of science as an object, which is what
this study is all about. I have reviewed the technical aspects of Rosen’s
work elsewhere [4]and will consider the broader philosophical implications
here.
Using the ideas Rosen developed, we can begin with the following work-
ing definition of complexity:

Complexity is the property of a real world s y s t e m that is manifest

in the inability of a n y one formalism being adequate to capture all
its properties. I t requires that we find distinctly different ways of
interacting with systems. Distinctly diflerent in the sense that when
we make successful models, the formal systems needed t o describe
each distinct aspect are NOT derivable from each other.

Rosen created a dichotomy between complex system and simple systems,

or mechanisms. The essence of that dichotomy is summarized in Table 1.

30
 31

Table 1. Complex systems vs. simple mechanisms.

Complex Simple
No largest model Largest model
Whole more than sum of parts Whole is sum of parts
Causal relations rich and intertwined Causal relations distinct
Generic Non-generic
Analytic Synthetic Analytic = Synthetic
Non-fragment able Fragmentable
Non-comput able Computable
Real world Formal system

This set of properties that distinguish disjoint categories of objects is

the subject of Rosen’s life work. Any brief summary will be lacking the
content developed to establish them in detail, but can give a sense of what
the categorization entails.
The categories are disjoint because of a central characteristic of the way
we go about practicing science. We create a surrogate world, the world of
simple systems, to enable our minds to try to grasp the complexity of the
world we wish to comprehend. Hence the real world is complex and must
be reduced to something less than its many faceted reality in order to be
studied. The surrogate simple world is a world of formal systems or models,
that replace the many faceted complex reality with something our minds
can work with. Our understanding depends on this process. The process
has consequences that can not be escaped. The table of characteristics is a
summary of those consequences, and they will be discussed briefly.

1.1. The largest Model

The formal system science has traditionally used to model the complex
world has a largest model. This model is the crux of scientific description
and can be called the “Newtonian Paradigm” using terminology adopted
by Rosen. The Newtonian Paradigm is a world constructed in terms of
non-linear dynamics in its many forms. The most modern forms all re-
solve to the largest model formulated as non-linear dynamics. Included are
chaos theory, cellular automata, and non-linear systems of partial differen-
tial equations or state space descriptions, among others. These formalisms
lend themselves well to the use of modern computing methods and therefore
computability becomes a hallmark of the largest model used to represent
32

simple systems. Real systems go beyond computable properties and can

only be approximated by computer simulation. This has been said in many
ways, such as “The map is not the territory”, and will often be acknowl-
edged in specific instances, only to be denied vigorously when proposed as
a general truth. In the working definition of complexity, this is manifest as
the need for distinct formalisms in order to completely represent complex
reality. In fact, there is no finite number of such descriptions.

1.2. Why is the whole more than the s u m of its parts?

A mantra in complexity research is the notion that any complex whole is
more than the sum of its parts. The world of simple systems created by
science does not escape this idea. We will give some examples of models
that can demonstrate that the same material parts can be assembled in
different ways to make very different functioning wholes. The key word
being “functional” for it is in function that this property of complex real-
ity is rooted. More interesting is the recognition that function is as much
an ontological aspect of complex reality as are the atomic and molecular
material building blocks. This must be so if the notion that the whole is
more than the sum of its parts is to have any meaning. Reducing a com-
plex whole to smaller units and eventually to atoms and molecules destroys
function and destroys the very identity of the complex whole. This thing
called “function” that characterizes complex wholes is as real as the mate-
rial aspects of the system. Simple systems can be reduced to their parts
and are called machines or mechanisms for this reason. This leads to an
unfortunate ambiguity between different meanings of the word “machine”.
An automobile is a man made machine and can seem to be identical with
the formal description used to create it. However, the real world automo-
bile is a complex entity only approximated by its formal description even
though it can be built from that formal description. It will be important
to keep the vernacular meaning of machine, the complex reality, separated
from its formal description as a machine or simple mechanism. In the rela-
tional systems theory being developed as an approach to complex system
description using the ideas introduced by Rosen, the material description
is sacrificed in favor of a functional description. The complex whole is re-
duced to a combination of functional components that can not be mapped
to material parts on a one to one basis. These functional components only
have definition in the context of the complex whole and are destroyed when
it is reduced to material parts. Hence the functional components are the
 33

way that the things that are “more than the sum of the material parts”
become entities having an ontology of their own. This idea is central to an
understanding of complexity.

1.3. Causality and information: Science of method and

science of content
Science has often been characterized by its methodology. This has been
the key feature that has kept science from being “contaminated“ by the
logical problems that are inherent in things like circularity and self ref-
erence. The limits imposed by this methodology also keep science from
dealing adequately with complex reality for this very reason. Within sci-
ence a very limited definition of 1Linformation77 was developed by Shannon
and Weaver to provide a tool in communications theory. The shortcomings
of this very sterile definition show up again and again because it does not
satisfy the intuitive notion if what the term “information” connotes. This
fact is closely linked to the relationship between syntax and semantics. The
concept of information they formulated deals only with syntax while what
is really needed is a way of also dealing with semantic content. Rosen intro-
duced another kind of information. The difference between the two versions
of information is related to the complexity/simple mechanism/machine di-
chotomy. Science, by dealing with mechanisms, led to questions about
iihow” things work. The other aspect to information deals with “why”
things are as they are. The question “how does it work?” is a reduction-
ist/mechanist question. The strict methodology of hard reductionist science
does not allow “Why?,’ questions because the method is impotent to deal
with them. Yet there is much information content in the answer to the
question “Why?” It was Aristotle that taught us that. Aristotle gave four
answers to the question “Why?” These causes or LLbecauses’’ are the ma-
terial, efficient, formal, and final causes. They are often illustrated as the
answers to the question “Why a house?”
Material cause: Bricks, wood, metal, glass, and other materials
that make up the house.
Efficient cause: The builders that put those materials into the
form of a house giving it a function beyond that
of the unassembled materials.
Formal cause: The plan or blueprint used to give the house its
form as a house.
Final cause: The purpose or function of the house, to be a
dwelling place.
34

In machines these causes can always be separated. In real complex sys-

tems they are entwined and not capable of being separated as in a machine.
The example to follow, the distinction between an organism and a machine,
will establish this in a clear example.

1.4. W h i c h i s generic, physics or biology?

The widely accepted myth that biology is special and that physics is generic
is totally wrong. In fact, far more can be learned about the material world
by a careful study of biology than can ever be learned from physics. Physics
is the model of hard reductionist science. Physics envy has done real dam-
age to fields that want to be considered “scientific” such as psychology and
sociology among others. Biological and social systems are clearly complex
only because they have resisted the attempt to find an acceptable largest
model as has been done in physics.

1.5. Analytic us. synthetic models

The distinction between analytic and synthetic models is a mathematical
one. It is based on the use of direct products and direct sums as the way the
models are constructed. The definition of these structures comes from cat-
egory theory. Rosen developed his own form of category theory but it does
not differ that much from other versions. Category theory is that branch
of mathematics created to overcome some problems in set theory. It is ac-
tually a way of modeling mathematics itself. To summarize the essential
difference between analytic and synthetic models is a matter of looking at
the way set like entities can be combined. Product sets are based on a set
theoretical operation called a Cartesian product. The Cartesian product
of a pair of sets A = (al, a2, “3) and B = (bl, b2) is symbolized as A x
B and consists of the collection of pairs (al, b l ) , (all b2), (a2, bl), (a2,
b2), (a3, b l ) , (a3, b2). The direct sum of these entities is merely the com-
bination of the members of A and B. In simple systems the direct product
can always be reduced to the direct sum. They are equivalent. This is sim-
ply a mathematical way of describing the reducibility or fragmentability
of reductionist models. In the case of complex systems, this is never true.
There is more involved because complex systems are modeled in terms of
their functional components and these functional components consist of di-
rect products that are not capable of being reduced to parts without the
destruction of the functional component.
 35

1.6. h g m e n t a b i l i t y
It follows directly from what has been developed so far that the complex
system with its context dependent functional components can not be frag-
mented into material parts. Simple mechanisms or machines can always be
fragmented into material parts.

1.7. Computability
This subject can fill a number of books. It is the subject of heated debates
and for those who have placed their faith in computers and computable
models, the stakes are very high. The sides of this debate can be exemplified
on one hand by the hard core proponents of “artificial life” and on the
other hand by Robert Rosen and many others who have come to see his
understanding of the complex real world as a fundamental breakthrough in
world models. The proponents of the reductionist/mechanistic view believe
that the Church-Turing thesis is correct.

2. SCIENCE AS A COMPLEX SYSTEM

Science is a complex system. It has many facets and many different ways of
being defined, described and analyzed. The most obvious use of the word
is as a noun. What does the noun stand for? It may stand for an ideal that
we sometimes call ”pure” science. Such ideals are approximated in reality
and never really realized completely. Science as a noun also stands for a
body of knowledge. Science is also a set of human institutions. Science can
also mean those things that the people we recognize as scientists do. When
we speak of science in that way we can also speak of an activity and we
sometimes refer to doing science so there is a verb form involved. We often
describe things as scientific as in a scientific study or a scientific expedition.
Then we see that there are adverbs and adjectives as well. There is also the
scientific method. There is a philosophy of science. This philosophy is very
important because it is the only way we can complete the system. That
is, if we were forced to study science using science we would be caught in
an impredicative self referential loop. This may be a partial explanation
of why scientists seem to avoid philosophy, hoping to escape the need for
another approach..
Science as a human institution also has a very important set of political
facets. The products of scientific activity are often a source of wealth and
power. The doing of science is an expensive enterprise and it must have
36

its patrons. Both complexity theory and the philosophy of science run into
this political aspect of science in a number of ways. By describing the
limits of the reductionist/mechanist paradigm there is a danger of casting
doubt on the ability of science to produce what its patrons expect from it.
Teaching such things to science students can place them in situations that
are difficult. Students must choose projects to satisfy their mentors who in
turn must satisfy their patrons. Reflecting these strong shaping forces back
on the definition we so roughly outlined, it should be clear that differences
between science as an ideal and science as what is practiced by scientists
can be very great if the ideal is involved with showing limits rather than
convincing patrons that their investment will pay off.
As the research tools and equipment grow in expense and sophistication,
the tendency to occupy one’s time using the equipment also grows. Thus the
activity we call science becomes driven by questions that can be answered
by the equipment rather than the desire to know in its “purest” form.
The methodology of science is not codified in any clear way. Books have
been written about the best way to accept candidates for scientific law and
scientific theory. Books have also been written about the failure of other
disciplines to satisfy the criteria of the scientific method.
Periodically there are so called paradigm shifts claimed. Often these
are not universally recognized as such, but what ever name one wants to
assign, change does occur in the method and this can be significant in its
magnitude. Proponents of complexity theory in its many forms often ask
for it to be a candidate for such status.
Is there a model for this kind of complexity in systems of human thought
and the activities that feed in to such systems? Clearly there is. Robert
Rosen chose Number Theory as his example. He reviewed the efforts of
the formalists to purge the field of all circularities and self referential loops.
To make a very long story very short, the attempt failed miserably. Each
attempt to present a finished product was met by demonstrations that im-
portant things about numbers were left out. This was because they were
insisting that the theory be self consistent and rightly so. It took Kurt
Godel to prove that such systems can not be both self consistent and com-
plete. By requiring self consistency they doomed their efforts to produce
an incomplete system. The only way to deal with complex systems was to
rely on something external to attempt the completion. But then the larger
system is subject to the same problem and an infinite regression results.
This issue has come up often in discussions among complexity theory pro-
ponents. One way some seem to satisfy themselves that they need not heed
 37

the problem is by claiming that Number Theory is mathematics and not

science and that Godel’s proof has not been shown to apply to physical
systems. This seems to be a false solution to the dilemma, especially since
all the models science uses are mathematical

3. COMPLEXITY AS AN ATTRIBUTE OF NATURE

The use of the term “complexity” has come to be commonplace both in
the scientific literature as well as in the lay press. A complex idea in its
own right, complexity presents a challenge to those who would define it or
codify its contents. Why is ”What is complexity” a question not so easily
answered? Universities and other research institutions have programs in
”complexity research” and journals carry this word in their title. What is
complexity? What does it mean to be ”complex”? The dictionary does not
help us here. There is more to the idea than what a dictionary definition
suggests.
We need to look within science, where words are carefully defined and
have more precise meanings, since it is here where the answer does lie. The
concept of complexity that is of so widespread interest was born within
the confines of the scientific community, had to struggle for existence, and
finally grew into a substantial force in scientific thinking. The answer lies
in the nature of science. Science helps us to understand the world around
us by creating a surrogate world that is understandable. The real world
is where complexity has its home. The world science gives us has become
simple and understandable.

3.1. Hard Science i s built o n Cartesian Reductionism

Hard science can be best-defined in terms of Cartesian Reductionism and
the machine metaphor. Descartes gave us the machine metaphor and he did
it in a very interesting way. In the case of living systems he saw the body
as a biological machine and the mind as something apart from the body.
This is called Cartesian Dualism and survives to this day as one approach
to the so-called mindlbody problem. What the machine metaphor did was
to set the tone for modern science. It is still with us. Descartes really did
not know what a machine is, or if he did, he never told anyone. Ironically,
not only do we not have a good definition of complexity, but we also lack
one for a machine. The duality between these concepts is the key to the
resolution of the problem. The importance of the machine metaphor is in
the intuitive concept of machine that almost everyone shares. A machine is
38

built up from distinct parts and can be reduced to those parts without losing
its machine-like character. We call this idea ”Cartesian Reductionism”. We
have seen that complex (real) systems can not be successfully reduced to
material parts without the loss of some significant attributes in the process.
This led to the axiom that the whole is more than the mere sum of its parts.
Adopting this axiom as a truth leads to a conclusion that is inescapable. If
the whole is more than the sum of its parts there must be more to things
than atoms and molecules. Reducing a complex whole in the real world
always destroys something. What is lost is an elusive entity, but is also the
central idea in the concept of complexity as applied to the real world. If
there is more to things in the real world than atoms and molecules then
this something that is there has an existence, an ontology. This is the
reason why as a concept complexity is a difficult one. It can not be both
ways. Either a real world whole is more than the sum of its parts or
it can be reduced to atoms and molecules with nothing being lost and
the world is made of machine-like things. Cartesian reductionism does not
work for making models of complex systems; it only reduces them to simple
mechanisms that may reflect some aspect, but merely as a shadow of the
complex whole

3.2. The Newtonian paradigm is the modern manifestation

of hard science
Newton gave us, among other things, three laws of motion, which were in-
tended to describe the motion of the planets. It turned out that methodol-
ogy spawned by the application of these laws could be applied in a seemingly
perfectly general way. This broader application has been the foundation of
the modern scientific method and will be referred to here as the Newtonian
Paradigm.
In the center of this paradigm is dynamics. Dynamics is the way the laws
of motion get applied. The local description of the motion is formulated as a
differential equation called an equation of motion. The equation of motion is
manipulated by using the calculus (integrated) and results in a trajectory,
which is an algebraic equation for calculating a particle’s position as a
function of time. Later, this was made somewhat more complicated by
quantum mechanics, but the central philosophy has never changed.
The paradigm has been generalized from particle motion to all systems
if we recognize that quantum mechanics is part of that generalization. The
world-view called reductionism has as a central theme that anything in
 39

the material world can be reduced to particle motion, that is, the motions
and interactions of atoms and molecules. When we look carefully at the
subject matter of physics, we see that it is the application of the Newtonian
Paradigm to the universe. This application then makes the world into
simple mechanisms. That is to say that the subject matter of physics is the
study of simple mechanisms. Note that in this context, "simple" means the
opposite of complex, not the opposite of complicated.

3.3. Complexity is the result of the failure of the

Newtonian Paradigm to be generic
A careful look at the real world tells us that it is complex. Yet, the success
of the Newtonian Paradigm cannot be ignored. Most of modern science
and technology is the result of it. For that reason alone it is difficult to
suggest that it has limits and to then make that suggestion stick. Not only
does the paradigm have limits, but also those limits are what gave rise to
a concept like complexity.

3.4. The way science i s done: The modeling relation

How is science done? It is a combination of using our senses to observe the
world around us and then to use some mental activity to make sense out
of that sensory information. The process is what we will call the modeling
relation. If we call the world we are observing and/or trying to understand
the Natural System and the events that make it change as we observe
causality, then that represents our object of study. What we do in our
minds is to encode the natural system into another system that is of our
making or choosing which we can call a formal system. Once we have chosen
a formal system, we can manipulate it in various ways with the objective
of mimicking the causal change in the natural system. These manipulative
changes in the formal system we will call implication. Finally, once we think
we have an appropriate formal system and have found an implication that
corresponds to the causal event in nature, we must decode from the formal
system in order to check its success or failure in representing the causal
event. Figure 1 represents the modeling relation we have just described.
If all the parts of the diagram are in harmony, in other words if 1 = 2+
+
3 4, we say that the diagram commutes and we have a model. A model of
the world is the outcome of a successful application of the scientific method,
but it can also arise in other, less formal ways. Whenever someone tries
to make sense out of the world, they are trying to construct a successful
40

DECODING
CAUSAL
EVENT IMPLICATION

SYSTEM

ENCODING

Figure 1.

modeling relation, or a model.

3.5. Complex systems and simple systems are disjoint

categories that are related by the modeling relation
The world therefore divides naturally into those things that are simple and
those things that are complex. The real world is made up of complex things.
It is something we encounter constantly and try to make sense of. The way
we do that is to encode it into formal systems. Therefore the world of simple
mechanisms is a fictitious world created by science or, more specifically, by
physics as the hard version of science as a formal system which hopes to
model the real world. This is the world of the reductionist. It is modeled
by the Newtonian Paradigm and simply needs sufficient experimentation
to make it known to us. Those experiments involve reducing the system to
its parts and then studying those parts in a context formulated according
to dynamics.
Therein lays the reason for the restrictions on the scientific method.
The method involves a rigid empiricism that feeds data into the formal
system. Here is the philosophical rub. The epistemology spills over into an
ontology. The modeling relation is forgotten (but it is always there) and
the formal system has become reality.
Now the definition of complexity is complete. The world, from which
we single out some smaller part, the natural system, is converted into a
formal system that our mind can manipulate and we have a model. The
world is complex. The formal system we chose to try to capture it can
only be partially successful. For years we were satisfied with the Newtonian
Paradigm as the formal system, forgot about there even being and encoding
and decoding, and gradually began to change the ontology so that the
Newtonian Paradigm actually replaced or became the real world (at least
 41

as seen through the eyes of science). As we began to look more deeply into
the world we came up with aspects that the Newtonian Paradigm failed to
capture. Then we needed an explanation. Complexity was born! This
easily can be formalized. It has very profound meaning.

4. THERMODYNAMIC REASONING AS A
TRANSITION TO COMPLEXITY SCIENCE
Rosen had little to say about thermodynamics in his critique of reduction-
ism in the form of the Newtonian paradigm. There is probably a good
reason for this. Clifford Truesdell [5] once made a very good case for ther-
modynamics having a certain “strangeness” as a part of physics, or we could
say as part of the reductionist Newtonian paradigm. This strangeness needs
to be considered in more detail for it reveals the seeds of the ideas Rosen
found to be true about the mechanistic approach to reality. Thermody-
namics is probably more poorly understood by mechanistic scientists than
any other branch of physics. The reasons for this are deep and revealing.
One of the most serious consequences of the inability of those doing
either mechanistic science or thermodynamics to see the problem clearly
is the resultant gap in the Newtonian largest model. It is possible to see
mechanistic reasoning and thermodynamic reason as different models of
complex reality in the spirit of the ideas discussed here. Instead, from its
beginnings thermodynamics was put under severe pressure to conform to
the mechanist’s largest model and thereby suffered a lack of development
as its own alternative to mechanistic physics.
Thermodynamics came into being for very practical reasons. The boring
of cannons, the brewing of beer and the steam engine are but a few of
the reasons that the frictionless world of Newton’s paradigm needed to be
patched up. Heat as a form of energy also had to be dealt with. Perpetual
motion machines had to be dealt with in a rational manner to curtail the
squandering of time and energy, sometimes by very bright scientists. As
progress was made, there were also immediate problems presented to the
those who had been content with the frictionless world where heat was
merely another form of energy and nothing more troublesome than that.
Since the material world was to be understood in mechanistic terms, a new
kind of mechanics called “statistical” mechanics had to be developed to
try to make a bridge between thermodynamic reasoning and mechanistic
reasoning.
Why is this so? The answer is one of the best demonstrations of the
42
42

power of Robert Rosen’s analysis even though he may not have ever seen it
himself. There is a fundamental difference between mechanistic reasoning
and thermodynamic reasoning and it can not be erased by the limited
successes of the use of statistical mechanics to bring them together.
Thermodynamics is about those properties of systems that are true
independent of their mechanism. This is why there is a fundamental asym-
metry in the relationship between mechanistic descriptions of systems and
thermodynamic descriptions of systems. From the mechanistic information
we can deduce all the thermodynamic properties of that system. However,
given only thermodynamic information we can deduce nothing about mech-
anism. This is in spite of the fact that thermodynamics makes it possible
for us to reject classes of models such as perpetual motion machines. (This
does not stop such models from appearing in subtle forms in the modern
literature.) This asymmetry is poorly understood because thermodynamics
is not a tool mechanists see as valuable. Some of this attitude is justified
by the structure of thermodynamics as it exists in texts and courses. In
some ways the field is a prisoner of its own history. It is useful to examine
that history with this problem in mind.

4.1. Classical or LLequilibrium”t h e r m o d y n a m i c s a n d i t s

limits
Thermodynamics recognizes distinct systems categories based on the con-
straints imposed on the system in order to study it systematically. This
thought process is very much like the defining of a “universeof discourse” in
other fields. A system is defined in terms of the way we allow it to interact
with the outside world. Often it is useful to see these interactions as having
to occur across some real or imaginary boundary or “skin” surrounding the
system. There are only three possible categories based on this mode of
reasoning:
(1) The isolated system. Isolated systems are isolated from the rest
of the world by a boundary that is impermeable to both energy
and matter flow. Such systems define the notion of “equilibrium”,
in the thermodynamic sense, as the state the system is in after
a sufficient amount of time is allowed to pass. The equilibrium
state is characterized by the disappearance of the “transient” flows
of matter and energy within the system that brought it to this
final state wherein nothing further can happen. That is true in the
large. Very locally, fluctuations around this equilibrium point are
 43

happening as a manifestation of the thermodynamic property called

“temperature”. The fluctuations are atomic and molecular motion
and increase in magnitude as heat energy is added to the system
and decrease in magnitude as it is removed. Isolating the system
will result in a system at constant temperature consistent with its
heat content.
(2) Closed systems are systems with boundaries that will allow the
transfer of heat energy through them but these boundaries are im-
permeable to matter flow. Thus closed systems contain a fixed
amount of material but can be warmed or cooled by manipulat-
ing their heat content. Such systems are capable of being put in
contact with reservoirs of heat energy at different temperature to
induce a flow of heat through them and are therefore capable of
being maintained in stationary states away from equilibrium.
(3) Open systems. These systems can exchange matter and heat energy
with their environment and need never reach an equilibrium. These
systems can exist in stationary states, periodic states, as well as
chaotic states.
The laws and ideas of thermodynamics were developed with great care
by doing thought experiments on equilibrium systems. Virtual “quasi-
stationary processes were conceived by imagining so called “infinitesimal”
changes to take the isolated system from one equilibrium state to another.
Sadi Carnot developed a clever method of bringing the system through a
cycle so that it always ended in its starting equilibrium state.
One central characteristic of equilibrium is the existence of a thermo-
dynamic variable called “entropy”, S, defined as the quotient of the heat
energy in the system, Q, and absolute temperature of the system, T:

S = Q/T
More useful was the notion of the entropy change associated with an
isothermal quasi-stationary process:

dS = dQ/T (2)
The entropy of a system, to oversimplify somewhat, is a measure of the
“quality” of the energy of that same system. This follows from reasoning
about heat engines being operated in cycles. In order for a heat engine (e.g.
a steam turbine) to produce work that can be used it must have a source of
44

hot steam and that steam must flow through it to a reservoir that is cooler.
The First law of thermodynamics is nothing new to physics since it is
simply the idea that energy can not be created or destroyed. Hence a stem
engine uses some of the heat energy, converts it t o mechanical work, another
form of energy, and allows the rest of the heat energy to pass through it with
the matter (in this case water and/or water vapor) that is conserved in the
process. The Second law of thermodynamics forbids the complete transfer
of heat energy to mechanical work as well as forbidding the operation of the
engine between a source and sink at the same temperatures. This is why the
entropy became a useful and necessary concept. The result of “cooling” the
matter in the process of extracting mechanical work, increased the heat/
absolute temperature quotient or the entropy of the system. One way of
stating the second law is that there is a requirement that any real process
must result in the increase of entropy overall. It is possible to make devices
that locally decrease entropy, but only if the global result is an increase.
At equilibrium, there is no change in the amount of entropy.

dS = 0 (3)
In any real process,

dS >0 (4)
The fact that so much of what we know about thermodynamics came
from reasoning involving equilibrium and isolated systems is ironic when we
recall that it was practical matters, the doing of things, the carrying out of
processes that motivated the entire development. Nevertheless, by resorting
to what now seems to be a rather clumsy thought process involving the
carrying out of processes by performing small incremental changes, much
was learned.
What is problematic is that the world of real processes is what is of
interest and equilibrium are the natural endpoints of some of those processes
in the situation where the system is totally isolated.

4.2. Dissipation, frzction, and irreversibility

Energy is conserved, but entropy must increase in the carrying out of any
process. The first two laws of thermodynamics are imbedded in that sen-
tence. The statement is true of all systems regardless of mechanism. The
result of this is that physics has two realizations. The original physics of
 45

Newton and his followers was the physics of systems that had dynamics
yet were without the creation of entropy. The “ideal” pendulum, for ex-
ample, can not exist and is the end of a limiting process where friction
is diminished to zero. Friction is another manifestation of the second law
of thermodynamics. It is the recognition that in any real process some of
the energy must be converted to heat energy. This heat energy can not be
converted back to the original form of energy by the same process. The
process is “irreversible”. One good way to see this is in a famous exper-
iment performed by Joule to measure the factor that must be used when
energy in the form of mechanical work is changed to heat energy by any
process. The number he obtained was called the mechanical equivalent of
heat. The device he used has value as a source of insight far beyond its
utility in obtaining that number.
The device was a jar of water surrounded by an insulating jacket having
the purpose of isolating the system thermally. Thus heat energy could not
enter or leave the system through its walls. At the top of the jar was
mounted a well insulated crank attached to a paddle wheel inside the jar.
Turning the crank allowed mechanical work to be performed on the water
in the jar. The system allows a simple mechanistic explanation for how the
mechanical work is converted to heat energy. The water is put into motion
in a directed manner by the paddle wheels, but looses that directed motion
eventually, the motion rather becoming “random” motion characteristic of
heat energy. Having a thermometer mounted so that the water temperature
can be monitored allows the direct relation between the work done turning
the crank and the appearance of heat in the jar, using the heat capacity of
water to calculate the amount of added heat energy from the increase in
measured temperature.
The process is irreversible. There is no way to reverse this randomiza-
tion of molecular motion to turn the paddle wheel and thereby gain back
the mechanical work. The energy has been converted once and for all as
long as it is confined to the thermally isolated jar.
This is a very revealing example of the nature of friction. Sometimes it
is difficult to see a mechanistic picture of how this frictional “dissipation”
of energy occurs. Chemical reactions have their own version of friction and
are as irreversible as any mechanical process. In electricity the resistance
to the flow of electrons results in electrical energy being converted to heat.
Knowing this, physics has created two categories for the systems it stud-
ies, conservative and non-conservative systems. Newtonian dynamics was
developed using conservative, frictionless systems (fictional) and then ex-
46

tended to the more difficult non-conservative systems that, as real world

processes, involved friction necessarily. These non-conservative systems
were irreversible and needed to be open if they were to do anything other
than wind down to an equilibrium state and stop.

4.3. Preserving the paradigm involved considering fraction,

irreversibility and dissipation
It wasn’t until the work of Onsager [6, 71 and others in the 1930’s that a
systematic physics of non-equilibrium, irreversible systems was developed.
This new physics was thermodynamic rather than mechanistic, but the field
of statistical mechanics was later extended to the non-equilibrium domain
making the problem of backing up the thermodynamic knowledge with me-
chanical interpretations in terms of atomic and molecular motion very much
more difficult. The increase in difficulty has severely limited the scope of
these endeavors. Onsager and the others developed a thermodynamic ap-
proach and escaped most of that difficulty. They did so in a way that had
its own inbuilt limitations, but that would not be seen for about forty more
years when Network Thermodynamics was developed by Oster, Perelson,
and Katchalsky and independently by Peusner [8]. The approach Onsager
and the others introduced is am “phenomenological” approach. This is
very different from the mechanistic physics that was used to describe non-
equilibrium dynamic systems up to that time. Onsager’s work was part
of what won him a Nobel Prize. Later, Prigogine won a Nobel prize for
building on the work of Onsager. One can speculate that Katchalsky may
have been there in Prigogine’s place had he not been gunned down by ter-
rorists in 1972. That event changed the history of science and the history of
the world. It was Kedem and Katchalsky who introduced non-equilibrium
thermodynamics to biology and changed the way we dealt with all the dy-
namic processes occurring in living systems. Then, along with Oster and
Perelson, he explored the new approach that married non-equilibrium ther-
modynamics to the huge body of theory generated for dynamic electric
networks.
The impact of all this and the work of others that built on it is largely not
understood by many even after complexity theory has grown in its impact
on science. The reasons are many and not easily identified without being
critical of the way science is practiced. The urge to preserve the mechanistic
paradigm and to see thermodynamics as only an appendage has been too
strong and in that way too destructive. When quantum theory was born it
 47

also was shaped by the same urges. The difference is that quantum theory
is so much more adaptable to the mechanistic reductionist approach. Its
mathematics and the interpretation of the mathematics could be given the
very same form as what it was clearly showing to be a failure of Newtonian
mechanics. Thermodynamics, on the other hand, uses a different form of
mathematics. To most scientists satisfaction quantum mechanics simply
helped further the knowledge generated by the Newtonian paradigm and
did it little harm. It allowed the reductionist philosophy to appear to have
established itself at all levels in the material description of reality. It became
possible to see a more or less universal bottoms up approach to science.
Yet there are findings from network thermodynamics that raise some very
interesting, and possibly troubling questions about this. The lack of interest
in these questions parallels a similar lack of interest in questions raised
about science by Robert Rosen as he explored the complexity of the world
science seemed to have mastered. It is very important that these questions
and the findings that have been generated by the use of thermodynamic
reasoning that is not constrained by the mechanistic mind set do not get
totally ignored and forgotten Yet that may be exactly what is happening
as reductionism forges ahead. Some insights into these events of scientific
history and the philosophical oversight or lack of same can be had thanks
to the contributions of Lakoff [9] who has applied cognitive linguistics to
similar situations in politics. His ideas can be married to a concept of
memes as “packages” of ideas that get passed along in a manner much
analogous to genes.

4.4. Framing the question in science:” Don’t think about

the whole system”
As we make models of the world we necessarily use what we have already
put together in our minds. The idea that complete objectivity is possible
arises when the nature of the modeling relation is misunderstood. A model
has many components. It consists of a number of subjective choices. The
choice of the formal system we use to encode sensory information in or-
der to think about how the real world is actually operating is a subjective
choice. The illusion of objectivity arises from the fact that the question has
been framed in a particular way. The idea that there is only one encoding
possible in order for the model to be “scientific” forces a choice. As soon
as we acknowledge that complex reality can and indeed must be encoded
into distinctly different formal systems, the illusion of objectivity is de-
48

stroyed. Lakoff uses the sentence “Don’t think of an elephant” to illustrate

the impossibility of escaping certain thoughts once such a framing event has
occurred. Modern science is framed by reductionist thought and this fram-
ing is so effective that the illusion of objectivity can do very well. A large
part of Robert Rosen’s work is devoted to helping us see the predicament
we are in. It is very interesting how different people react to these ideas.
If the concept of complexity being used here is not accepted by a person,
there is almost no chance that the person will understand why objectivity
is not ever possible. The irony is that that position is a subjective choice.
If this is understood, it becomes a simple matter to understand why
thermodynamic reasoning is only accepted when it has a mechanistic coun-
terpart. The largest model must be able to encompass everything scientific.
The idea that distinctly different formal systems must exist as valid models
is unacceptable. Thus the “useful” ideas from thermodynamics are made
to be derivable from mechanistic reasoning, namely statistical mechanics,
and any other conclusion drawn independent of that mechanical base are
suspect.

4.5. Reductionism needs a particular kind of mathematics

t o accomplish its goals
The formal systems reductionist science use are mathematical. Yet among
the rich formalisms in the total mathematical repertoire, only analytical
mathematics is able to fill the needs of the reductionist paradigm. Calculus
and differential equations are the essence of the Newtonian approach and
were indeed created for that use. Thermodynamics and Rosen’s approach to
complexity utilize another kind of mathematics. It is topology that enters
as the formalism that provides the needed structure for encoding relation-
ships in the real world. The analytical mathematicians also were forced to
utilize topology in the breakthroughs that occurred when the realization
that so much of interest had to be encoded using non-linear differential
equations as the equations of motion. Topology and differential geometry
then were found useful even to reductionists. Out of this the nature of
chaotic systems was born and chaos theory had a central role during a brief
period of the evolution of complexity theory. It is still not clear to everyone
that chaotic dynamics are part of the Newtonian paradigm and do not con-
stitute a distinct formalism different from its largest model. The topological
mathematics used to model complex systems and the topological mathe-
matics that must be used to arrive at certain unique new conclusions about
 49

the real world using thermodynamic reasoning are very closely related.
Rosen, as a student of Rashevsky recognized that topological mathemat-
ics was necessary to create the relational biology they both had envisioned
as the approach to living systems that was not locked to the severe limits
of the reductionist paradigm. The elusive qualities of living things that
distinguished them from non-living mechanistic things could only be dealt
with by encodings into topological mathematical formalisms. Hence Rosen
developed the metabolism-repair M, R system as the formalism he would
manipulate to come up with a clear distinction between the classes of things
we call “organisms” and the class of things we call “machines”. In doing
this he forced his audience to make a hard choice. Either they would accept
a formalism that “kept the organization” but left out the “physics” or they
were locked into the reductionist’s world of physics which necessarily lost
crucial aspects of the complex whole as it was reduced to mere material
parts.
Thermodynamics in its most powerful form mimics this situation. There
are many examples, but only a few need be mentioned to make this point.

4.6. Topological reasoning in thermodynamics leads to

powerful results
One of the most interesting examples of the comparison between reduc-
tionist methods and relational or topological methods is in the proof of the
second law of thermodynamics which was accomplished by Caratheodry af-
ter a discussion about the proofs existing up to then with Max Planck [8].
The existing proofs were done using the cyclic quasi-stationary processes
developed by Carnot. Plank found these a very unsatisfying way to prove
something having the global consequences of the second law. Caratheodry
was able to construct a topological proof that rested on one piece of empir-
ical input to confirm it. The second law had its global validity established
by Joule’s experiment for it was only necessary to give one instance of
irreversibility in the real world to make the model valid.
The discovery of a metric structure for thermodynamics had been for-
saken until Network Thermodynamics in the hands of Puesner revealed that
the network thermodynamic model of a system was canonical. This discov-
ery rests on the topological embedding of the system posited by Onsager,
Prigogine and the others in a higher dimensional space[8].
50

5. Will science extend to the modeling of complex reality

or will it be restricted to the limited domain of the
largest model formalism it clings to?
This is a question that should not have to be asked. It is the self-imposed
limits that scientists have used to frame the answer to the question “What
is science?” that make it necessary. The framing of the question in this
manner is strengthened by the enormous progress being made in technology.
The hunger for more and more practical results makes it very difficult to
persuade scientists to spend time and resources pursuing an approach that
is just beginning to be recognized by significant minority among them.
Should bright students be encouraged to work in this new area? Is it wise
to devote any significant portion of the scarce resources given to science to
the development of these radically new ideas? It seems to be easier for many
scientists to choose an alternative definition for the word “complexity” and
to make sure that this definition never allows those using it to step outside
the reductionist paradigm.
There are a few who will not be restricted in that way. Two examples
of recent progress in complexity science are very different in their area of
application yet they hold in common their dedication to the new challenge
to find distinct formal systems that can not be reduced to the Newtonian
paradigm.
In their book Into the cool [lo] using earlier work by Schneider and Kay
[ll]develop a thesis about the synergy between life on this planet and the
conditions that sustain that life based on a broader interpretation of the
second law of thermodynamics than used before. They see the second law
as demanding that systems respond to imposed gradients by some form of
self-organization in order to decrease the gradient. The alternative is to
have enormous driving forces stressing the system. Hence the biosphere
is believed to be the result of the tendency for the sun’s energy to create
stressing gradients. They present evidence for the role that life forms,
weather patterns, etc. play in keeping the system stable. Clearly, a concept
like this one is not going to be something we prove or disprove. At a time
when so much controversy centers around the impact of human activity on
planetary conditions, their endeavors do provide a model worth examining
further before dismissing it out of hand merely because it forces us to
abandon the restricted universe of discourse reductionist science has forced
upon us.
The second area where some bold new ideas are being offered ins in
 51

the domain of what we often call the “mind-body” problem. The work of
Bach-y-Rita as interpreted by Kercel [12, 131 and others goes tom the heart
of many modern controversies. Their interpretation of real experiments in-
volving the sensory behavior of humans is shaking quite a few foundations.
At the center of all their models is the central role played by closed loops
of causality. These loops are the self-referential heart of complex reality
and the human mind seems replete with them. This work and he work of
Louie [14], a student of Rosen, showing that Rosen’s {M,R} systems nec-
essarily have non-computable aspects suggests that “artificial intelligence”
and “artificial life” are indeed merely machine intelligence and simulations
of systems that have some limited life-like qualities. These technological
marvels are distinctly different from human intelligence and living systems
in many import ways.
These new investigations as well as the areas of thermodynamics that
have been largely disregarded because of their distinctly non-mechanistic
character strongly suggest that science will find a way to include other
formalisms and break free from the restrictions it has imposed in itself.

References
1. Rosen, R. Anticipatory Systems, New York, Pergamoii ,1985
2. Rosen, R. LifeIitself, New York, Columbia, 1991
3. Rosen, R. Essays on LifeIitself, New York, Columbia, 2000
4. Mikulecky, D. C. The circle that never ends: Can Complexity be made sim-
ple? In Complexity in Chemistry, Biology, and Ecology, D. Bonchev and D.
H. Rouvray, eds. New York, Springer, 2005.
5. Truesdell, C. Rational Thermodynamics, New York, McGraw-Hill, 1969
6. Onsager, L. Reciprocal Relations in Irreversible Processes I, Phys. Rev.,
1931a, 37, 405-426.
7. Onsager, L. Reciprocal Relations in Irreversible Processes 11, Phys. Rev.,
1931b, 38, 2265-2279.
8. Mikulecky, D. C. Applications of Network Thermodynamics to Problems in
Biomedical Engineering, New York, New York University Press, 1993.
9. Lakoff, G, Don’t Think of an Elephant! Know Your Values and Frame the
Debate White River Junction, Vermont, Chelsea Green Publishing, 2004
10. Schneider, E. D. and Sagan, D. Into the Cool, Chicago, University of Chicago,
2005
11. Schneider, E. D. and J. J. Kay, Life as a manifestation of the second law of
thermodynamics, in Modeling Complex Biological Systems, hl. Witten and
D. C. Mikulecky, eds, Special Issue of Mathematical and Computer modeling,
1994, 19, 25-48.
12. Kercel, S. W. Journal of Integrative Neuroscience, 2005, 4, 403-406.
13. Kercel, S. W., Reber, A. S. and Manges, W. W. Some Radical Implications
52

of Bach-y-Rita's Discoveries, Journal of Integrative Neuroscience, 2005, 4,

551-566.
14. Louie, A. H. Any Material Realization of the (M,R)-Systems Must Have
Noncomputable Models, Journal of Integrative Neuroscience, 2005, 4, 423-
436.
 ON THE IMPORTANCE OF A CERTAIN SLOWNESS
Stability, memory and hysteresis in complex systems*

PAUL CILLIERS
University of Stellenbosch
fpc@sun. ac.za

In philosophy the winner of the race is the one who can run most slowly.
Or: the one who gets there last.
Wittgenstein (Culture and Value)

1. Introduction
As a result of a whole range of what one could call “pathologies” in con-
temporary culture, the idea of “slowing down” has of late been mooted in
a number of contextsa. A few can be named briefly. The “Slow Food”
movement, which started in Italy but has a worldwide following, extols
the virtues of decent food made from decent ingredients without compro-
mise. The resistance shown to “junk food” is not only based on aesthetic
considerations, but also on ethical (and nutritional) ones. The movement
promoting “Slow Cities”, also of Italian origin, fosters an understanding of
cities which is more humane. Such cities should encourage walking rather
than driving, have small shops with local products rather than shopping
malls and, in general, provide opportunities for the community to interact,
not to live in isolation. “Slow schooling” is a movement which questions
educational processes in a world geared for instant results. It emphasises
the contextual nature of knowledge and reminds us that education is a
process not a function. On a more personal level, “slow sex” involves at-
titudes which try to prevent that the values of the marketplace also rule
in our intimate relationships. We need to recognize that the journey is

*This paper first appeared in Emergence: C o m p l e i t y and Organization. Permission to

use the material is gratefully acknowledged.
Honor6 (2004) for a discussion of the emergence of several movements which chal-
lenge the “cult of speed”.

53
54

more important than the destination, and that takes time. An immediate
or perpetual orgasm is really no orgasm at all.
There are a number of very important issues at stake in these examples.
In what follows, however, the focus will not be on these social movements
as such, but on some of the underlying principles which make the debate
on slowness an important one. Through an analysis of the temporal nature
of complex systems, it will be shown that the cult of speed, and especially
the understanding that speed is related to efficiency, is a destructive one.
A slower approach is necessary, not only for survival of certain important
values or because of romantic ideals, but also because it allows us to cope
with the demands of a complex world in a better way.
The argument will be made initially by briefly analysing current dis-
tortions in our understanding of time. These distortions result, on the one
hand, from the rational and instrumental theories we have about a modern
world, and, on the other, from the effects of certain technologies, especially
communication and computer technologies. In order to show why these are
“distortions”, or at least, to show why these distortions are problematic,
the temporal nature of complex systems will be discussed. The relation-
ship between memory and anticipation will be central to this discussion,
but attention will also be paid to the importance of delay and iteration.
These characteristics of complex systems have important implications for
our understanding of the formation of identity, both individual identity as
well as the identity of groups. In closing, a number of general cultural issues
involving the fast and the slow will be looked at.
It is important to realise that the argument for slowness is not a conser-
vative one; at least not in the political sense of the word. It is not merely
backwards-looking nor a glorification of what has been. Although it em-
phasises the historical nature of knowledge and memory, the argument for
slowness is forward-looking; it is about an engagement with the future as
much as with the past. Slowness is in itself a temporal notion, and in many
ways the opposite of the notion “static”. In point of fact, it is actually an
unreflective fastness which always returns you to the same place.
It should also be stated up front that there is no argument against an
appropriate fastness. A stew should simmer slowly, but a good steak should
be grilled intensely and briefly. The argument is against unreflective speed,
speed at all cost, or more precisely, against speed as a virtue in itself; against
the alignment of “speed” with notions like efficiency, success, quality and
importance. The point is that a system which has carefully accumulated
the relevant memories and experiences over time will be in a better position
 55

to react quickly than one which is perpetually jumping from one state to
the other.
Perhaps “slow” and “fast” are not exactly the correct terms to use.
Terms like “reflective” and “unreflective”, or “mediated” and “unmediated”
may be more accurate. Nevertheless, the debate taking place uses “slow”
and “fast”, and the terms do have a certain rhetorical significance. If we
stay with their use, it is done in a metonymical way. The whole point of
this paper is to give them a richer meaning.

2. Living in the Present

In Tame: The Modern and Postmodern Experience Helga Nowotny (1994)
argues for a certain shift in our experience of time. In short, in my para-
phrase, and incorporating insights from Bauman (e.g. Bauman 1992), the
argument is the following: One of the main aims of the instrumental ra-
tionality flowing from the Enlightenment was to create conditions in which
we are not controlled by contingency. To achieve these conditions, it is
necessary to understand, and preferably control, the future. This demands
co-ordinated and goal-oriented action in the present. Modernism becomes
a project which demands our total commitment against the forces of irra-
tionality and chaos.
The modernist project has two important effects on our understanding
of time. In the first place, our actions need to be ceordinated. This
can only happen if time is universalised in such a way that we all live in
the ‘same’ time. This was achieved mainly through technology, i.e. the
construction of accurate clocks, and by regulating time globally. Instead of
each person or local community living in their own time, it was necessary
to synchronise time in such a way that activities in say New York and Paris
could be correlated. The effects of this, however, go much further than
merely synchronising time in different parts of the globe. It also means
that private time and public time are synchronised. We have to live our
lives according to a generalised and controlled understanding of time. A
subjective, or should one say phenomenological, experience of time has to
be sacrificed in order to generate a universal temporal framework in which
we can operate efficiently.
The second effect of instrumental rationality on our understanding of
time is a result of the desire to control the future; for the future to be made
knowable. This would only be possible if the future, in some essential way,
resembles the present. We cannot anticipate what we do not know, and
56

therefore we should do everything in our power to create a future which

does not disrupt the steady progress we are making towards a better world.
This modernist strategy is perhaps exemplified best in Hegel’s dialectic of
history which is supposed to converge towards an ultimate solution. The
actual result of this ideology is to extend the present into the future, causing
us to live in a perpetual “present”. This collapse of the diachronic into the
synchronic allows instantaneous interaction between everybody; it creates
a world which is fast and efficient. The sacrifice made in order to achieve
this, however, is nothing short of sacrificing the very notion of temporality.
Nowotny (1994: 16) calls it “the illusion of simultaneity”.
The way in which contemporary society lives in an eternal present, or
what Eriksen (2001) calls the ‘‘tyranny of the moment”, is made possi-
ble, and augmented, by the surge in technology, especially computer and
telecommunication technology. We are instantaneously in contact with ev-
erybody everywhere. Not only has the distinction between home and the
workplace collapsed, but also the distinction between work time and private
or leisure time. It is expected that we are available, always and everywhere.
This state of affairs may have been less detrimental if it did not also demand
instant response. The very reason for mobile phones and e-mail lies in the
fact that immediate response is possible. It is in this “immediate” that
the main problem lies. There is less and less time for reflection. Reflection
involves delay, and in a cult of speed, delay is unacceptable. This move
away from reflection to immediate response has profound implications for
our understanding of what it is to be human (see Parkins 2004: 376-379),
t o which we shall return.
The “illusion of simultaneity”, the idea that if we live quickly and ef-
ficiently in the present we are somehow closer to reality, is nevertheless
exactly that: an illusion. We cannot escape our temporal nature, and our
persistence in trying t o do so can only lead t o pathology. The necessity
of delay and reflection needs to be re-evaluated. This can be done from a
number of perspectives. A Freudian analysis would show that instant grat-
ification is actually a destruction of pleasure. More sublime pleasure can be
found only if desire is delayed, anticipated as a memory of something still
t o come, yet something which should also in principle be able to surprise
us. Derrida calls the illusion of living in the present, of thinking that we
have access to an objective understanding of reality if we live “in” it, the
“metaphysics of presence” (Derrida 1976: 49). He introduces the notion
of dzffkrance specifically to undermine the metaphysics of presence (62).
DzffLrance is a notion which intertwines difference (as a spatial notion, one
 57

could say) and delay (to defer, a temporal notion) as the engines of meaning
(Derrida 1982). The present consists only as a combination of memory (of
what has been) and anticipation (of what is to come).
In his novel Slowness Milan Kundera (1996) uses the metaphor of some-
body riding on a motorcycle as being constantly in the present. Speed and
the demands of the machine reduces his horizon to something immediate.
Someone walking, however, is moving at a pace which allows for a much
wider horizon. The stroll unfolds in time in a way which opens up reflection
about where we are coming from and where we are going to, as we walk.
This theme of both the past and the future being present in a meaningful
experience of the present could be pursued in much more detail from both
a Freudian and Derridean perspective - and several others too - but the
argument for a meaningful temporality, i.e. something slower, will be made
here from the perspective of the dynamics of complex systems.

3. Complex Systems, Temporality and Memory

An important aspect of complex systems, one which certainly complicates
our understanding and modelling of such systems, is their temporal nature.
Complex systems unfold in time, they have a history which co-determines
present behaviour and they anticipate the future. Moreover, as we know
at least since the work of Prigogine, the behaviour of complex systems are
not symmetrical in time. They have a past and a future which are not
interchangeable. This being “situated in time” does not always receive
adequate attention in our analysis of complexity.
The central notion at stake when we talk of time and complexity is that
of “memory”. Memory is the persistence of certain states of the system,
of carrying something from the past over into the future. It is not merely
the remembering of something in the past as if belonging to that past, it
is the past being active in the present. We should therefore not think of
memory in abstract terms, but of memory as something embodied in the
system. In many respects the system is its memory. If one accepts a n
understanding of complexity which emphasises the relational nature of the
system, it is useful to think of systems as networks where the connections
between the nodes are more important than the nodes themselves. The
nature of these connections is a result of which states of the network are
“retained”, thus the structure of the system is a result of the sedimented
history of the system.b

bThis argument can also be made using the example of the brain, and links with many
58

It is important to remember that memory is not merely a cumulative

process. The structure in the network of relationships can only develop
if certain states of the network are not maintained. Memory is a result
of a process of selection. The states which are significant are repeated
more often and therefore form more permanent links in the network. Less
significant states will fade away over time. Memory is only possible if the
system can also forget‘. What is important to note at this stage is that
memory is not an instantaneous thing, it takes time to develop, it is slow.
If one characterises memory as the past being carried over into the
future, it follows that the future can only be anticipated in terms of the
memory of the system. Anticipation is not, or at least, should not be, simply
an extrapolation of the present. It is a complex, non-linear process which
tries to find some trajectory, some way of “vaulting” from that which has
already been experienced to that which has to be coped with. The quality
of the anticipation is a function of the quality of the memory. A more
varied, richer, deeper and better integrated memory will open up more
sophisticated anticipatory capabilities.
The obvious question now would be to ask how such a rich memory
is formed. This is a complex issue, but for the sake of the argument at
stake here, one can say the following: memory is information from the
environment which has been filtered, it is that which has been interpreted
- by the memory already sedimented in the system - as significant. The
identity of the system is, in some sense, its collection of dynamic memories.
The implication is that the system cannot reflect, or act upon, everything
which is going on in the environment at a given moment. If that were the
case, the system would always be merely a reflection of its environment and
would have no identity of its own. In order for it to be a system at all, a
system which has its own identity, which can react to the environment and
not merely mirror it, a certain hysteresis is required. The system must be
slower than its environment.
The notion of hysteresis is an important oned. It links to the notions of
delay and dzffirance discussed above. An event in the environment of the
system does not have inherent and immediate significance for the system.

Freudian arguments in an interesting way. See Cilliers (1998:4547,92,108)for further

discussion.

CThisprocess is known as the “use principle” or Hebb’s rule. For more detail see Cilliers
(1998: 17-18,93-94)
dHysteresis is the “lagging of effect when cause varies” (Oxford Concise)
 59

Its significance is established in time as it is reenacted in the system and

carried over into the future. In a way, the significance of an event has
always already been established (in terms of the memory of the system),
but never completely or finally since the significance is always also to be
determined by what is still to come. The system has to hang on to some
aspects with a certain tenacity; not let go of them too quickly. There is
risk involved in this, of course. The system has to invest resources in this
process. It cannot maintain everything; it has to select. If too many of
the wrong things are carried over it will impair the system’s performance.
However, if not enough is carried over, it will also fail.
To put it in slightly different terms: the system has to find a way to
discriminate between information and noise. If it follows every trend in its
environment, it will also be following noise. If it reacts too slowly it will only
follow the low-frequency trends, which may also be just noise. The system
must be stable enough not to be buffeted around by every fluctuation, and
it must be flexible enough to be able to adapt when necessary. Where this
optimal point lies is not a question which can be answered from an objective
viewpoint. The balance between stability and change is a contingent thing
which plays itself out in time. What one can say, though, is that merely to
be fast will destroy the system.
The argument for a certain slowness should start to take shape now. A
viable system has to be able to resist some of the dynamics in its environ-
ment. There should be a temporal space in which the past is allowed to
play itself out in interaction with present. There must be time for reflec-
tion and interpretation. The faster the system becomes, the shallower its
resources will be. Ultimately quick behaviour will be no more interesting
than Brownian motion.
It must be stressed again that the argument for a certain slowness is not
a conservative argument. A certain amount of conservation is a prerequisite
for a system to maintain itself, of course. The important point, to which
we shall return, is that a “slow” strategy is not a backwards-looking one.
If a somewhat slower tempo allows a system to develop a richer and more
reflective memory, it will allow the system to deal with surprises in its
environment in a better way. The argument of slowness is actually an
argument for appropriate speed. There is no objective or immediate rule for
what that speed is. If anything, it is a matter of experience, and experience
(as Aristotle urged) has to be gained, it cannot be “given” in an immediate
way. It is experience which determines which piece of meat should be
fried quickly and which should simmer slowly in the stew. She who fries
60

everything quickly will only have something nice to eat now and then, and
then purely by chance.

4. Integrity, Identity and Reflection

In his novel The Discovery of Slowness, Sten Nadolny (2003) g'ives us a
fictionalised account of the life of John Franklin. Franklin, a l S t h century
explorer primarily obsessed with finding the Northwest passage, was slow.
His advance (in the Royal Navy) is also slow, mainly because being slow is
confused with being stupid. Since he is not stupid he is gradually awarded
command, and those working with and under him discover the advantages
of being slow. Franklin is persistent, dependable and trustworthy. Even in
war, thorough reflection pays dividends not always immediately apparent.
His political career, as governor of Van Diemen's Land (now Tasmania),
ends badly only because he is disgraced by those out for quick and selfish
results. His principles are not negotiable.
Franklin is a worthwhile human being because he has integrity. There
is a substance to his personality which may seem opaque at first, but even-
tually shows itself as solidity. The nature of his integrity is directly coupled
to his slowness. He assimilates, integrates and reflects before he acts. This
is sometimes a ponderous process, and he pays a price for it. Under normal
circumstances it is easy not to notice someone like this, or to pass him by,
but when there is a crisis, it is him they turn to. He can be trusted, he
will come up with something. This is most significant. It is exactly when
one would think that being fast is what is required that slowness proves its
worth.
The link between slowness and integrity is also an issue in J.M. Coetzee's
(2005) novel Slow Man. Here we have a character who resists change,
despite the cruel demands being made on him. He clings to a set of values
important to him, and this gives his personality substance. However, he is
too stubborn, and eventually he cannot adapt to new circumstances. One
has tremendous sympathy for him, but he turns out to be too slow, and
pays the price for it. Even so, it is clear that when there is a choice between
the loneliness of the slow and the superficial companionship of the quick,
the author sides with the slow. Integrity is more important than a certain
kind of success.
Despite Coetzee's darker view, there is no reason why slowness should
be solitary and sad. Quite the contrary is true. In his novel Slowness,
Milan Kundera (1996) shows with great conviction how a certain slowness
 61

is a pre-requisite for being fully human. What is at stake in this novel is

not moral integrity, or a kind of Calvinist dependability, but the sensuality
of human interaction, the beauty of a relationship which unfolds in time,
the ecstasy of a love which has a history and a future. Being human implies
having a body, something with its own rhythms and demands. If we reduce
all of this to something merely instrumental, to transactions written in legal
terms (not in lyrical prose), if we demand results now, then we will stop
being human. Language cannot be reduced to a code; it plays itself out in a
certain context. What is more, even if we immerse ourselves in the context
we have to wait beyond the last sounds. When all is said, the meaning
has not finally arrived yet. It is the anticipation of what it could yet mean
which draws us forward. Einmal ist Leinmal.
Many may feel that the novel is an outdated art form, something to be
replaced with the fast and immediate communication of the digital code. In
his book The Moment of Complexity Mark Taylor (2003) seems to lean in
this direction. For him, something of a paradigm shift has occurred in the
last few decades. We live in a new world with new forms of communication
and new forms of learning and human interaction - something he seems
quite willing to sell. He resonates with a fast world, something new and
exciting.
His emphasis on the new goes hand in hand with a nearly interchange-
able use of the notions “noise” and “complex”. This problematic conflation
is the result of an understanding of complexity primarily informed by chaos
theory and of information as entropy. Such an understanding, inspired by
the theories of Shannon and Chaitin, will attribute the highest information
content to a purely random sequencee. Although these notions are impor-
tant in the context of computation, they are less useful when talking about
complex systems in general. Living systems, including the social systems
Taylor explores, are neither random nor chaotic. Despite the fact that they
are constituted through non-linear interaction and that they are capable of
novel and surprising behaviour, they are well-structured and robust. They
persist through time and maintain themselves. When we encounter be-
haviour that we do not understand, or cannot decode, it often looks like
noise, but once it is understood we can see the patterns. These patterns
are not merely or only an order imposed by the observer, but also charac-
teristics of the system itself. Complexity may look like noise, but all noise

“See Hayles (1999) for a discussion of these issues. Primary sources are Shannon (1949)
and Chaitin (1987).
62

is not something complex waiting to be decoded. Sometimes noise is just

noisef,
Taylor’s argument is seductive, but, to my mind, wrong if not harm-
ful. In his fervour to embrace the posthuman he looks at the history of
being human with a certain disdain. It seems as if he thinks that com-
plexity is a recent discovery and forgets that being human has always been
complex. He embraces the present and wants to deal with it quickly and
efficiently. We can be educated instantaneously by electronic means and
thus we should make a radical break with old methods. In his excitement
he forgets that complex systems, even those in our postmodern world, are
constituted historically, that they develop and change and that one of their
primary functions is to distinguish between information and noise. This
cannot be done at the press of a button.
The ideas of the posthuman and the cyborg are of undeniable impor-
tanceg, but in our enthusiasm to embrace new modes of being we should be
careful not to effect a transformation into something inhumanh. Machines
are fast, but they are machines. The present argument is not for an a prior2
rejection of the possibility of machines with human capabilities, or one that
denies the intimate relationship humans have always had with technology.
Our cultural existence presupposes the use of tools. The difference between
using a quill and word-processor may have huge implications on a practical
level, but they also share some essential features’.
The notion “posthuman” is thus an ambiguous one. If it signifies a tight
coupling between the body and technology, we have always been posthu-
man. If it signifies the obsolescence of the body, perhaps in the sense that
a “person” could be downloaded instantaneously as software and run on a
machine, it becomes a problematic notion at odds with the idea that human
identity is also the result of a certain temporal embodiment. The general
argument presented here maintains that any complex system, organic or
not, would have to incorporate a certain slowness.
The need for slowness, and a warning against the embracing of the

‘A similar, and more detailed argument is made in Cilliers (2005).

Wee Badmington 2000 for a collection of philosophical essays on the posthuman.
hSee Hayles (1999) for a detailed discussion of cybernetics, the development of the
posthuman and the importance of embodiment. See Braidotti (2005) for an affirma-
tive discussion of the posthuman which is neither a euphoric, uncritical acceptance of
advanced technology, nor a nostalgic lament for the decline of classical humanism.
‘There would definitely be a lot less drivel to wade through if it was not possible to write
so quickly.
 63

fast, can perhaps be motivated best from the perspective of philosophy.

Philosophy, in its most general form, is essentially the art of reflection.
Wendy Parkins (2004) analyses contemporary culture as one moving away
from reflection, and argues that what we need is an “ethics of time”. She
does not elaborate much on what such a n ethics should look like, but it is
something which needs careful attention, not only from a moral perspective,
but also from the purely pragmatic perspective of how to live and survive
in a fast world. Such an “ethics” will be complex in itself. It will have t o
unfold in time and be conscious of its own temporal nature.
For now, instead of rushing around like the red queen in a world where
change is virtuous merely because it is change, we can start by taking some
time out to reflect. At this point the argument for slowness becomes a
political one: we should put up some resistance to a culture in which being
fast is a virtue in itself. With a little more regularity we should say “no”.

References
1. Badmington, Neil (ed.). 2000. Posthumanism. London: Palgrave.
2. Bauman, Zygmunt. 1992. Intimations of Postmodernity. London: Routledge.
3. Braidotti, Rosi. “Cyberfeminsim
with a difference.” http://www.let.uu.nl/womensstudies/rosi/cyberfem.htm
(Accessed 8 August 2005)
4. Chaitin, G.L.J. 1987. Algorithmic Information Theory. Cambridge: Cam-
bridge University press.
5. Cilliers, Paul. 1998. Complexity and Postmodernism. Understanding Complex
Systems. London: Routledge.
6. Cilliers, Paul. 2005. Complexity, Deconstruction and Relativism. Theory Cul-
ture €4 Society, Vol. 22 (5), pp 255-267.
7. Coetzee, JM. 2005. Slow Man. London: Seeker and Warburg.
8. Derrida, Jacques. 1976. Of Grammatology. Baltimore: John Hopkins Univer-
sity Press.
9. Derrida, Jacques. 1982. ”Diffkrance” in Derrida, Jacques. Margins of Philos-
ophy. Chicago: The Harvester Press, pp. 1-27.
10. Eriksen, Thomas Hylland. 2001. Tyranny of the Moment: Fast and Slow
Tame in the Information Age. London: Pluto Press.
11. Hayles, N Katherine. 1999. How W e Became Posthuman: Virtual Bodies in
Cybernetics, Literature, and Informatics. Chicago: The University of Chicago
Press.
12. Honork, Carl. 2004. I n Praise of Slowness: How a Worldwide Movement is
Challenging the Cult of Speed. London: Orion.
13. Kundera, Milan. 1996. Slowness. London: Faber and Faber.
14. Nadolny, Sten. 2003. The Discovery of Slowness. Edinburgh: Canongate.
15. Nowotny, Helga. 1994. Time: The Modern and the Postmodern Experience.
Oxford: Polity Press.
64

16. Parkins, Wendy. 2004. “Out of Time: Fast subjects and slow living”. In Time
and Society, Vol. 13 No 2, pp. 363-382.
17. Shannon, C.E. 1949. Communication in the presence of noise. Proc. IRE.,
Vol. 37, pp. 1cb21.
18. Taylor, Mark C. 2003. The Moment of Complexity: Emerging Network Cul-
ture. Chicago: The University of Chicago Press.
 SIMPLICITY IS N O T TRUTH-INDICATIVE

BRUCE EDMONDS
Centre for Policy Modelling
Manchester Metropolitan University
http://bruce. edmonds.name

1. Introduction
The notion of simplicity as an important property of theories is traditionally
ascribed to William of Occam (1180), who extensively used the principle in
argument to rebut over-elaborate metaphysical constructions. It has been
invoked as part of the explanation why the Copernican account of plane-
tary motion succeeded over the Ptolemaic one whilst the evidence was still
equivocal. Newton made it one of his rules of reasoning: “we are t o admit
n o more causes of natural things than such as are both true and suficient to
explain their appearances f o r Nature is pleased with simplicity and aflects
not the pomp of superfluous causes.” [Ref. 1, page 31. Einstein chose the
simplest possible system of tensor equations to formalise his theory of Gen-
eral Relativity.2 Phrases like “for the sake of simplicity” are used to justify
many modelling decisions (e.g. Ref. 3, page 1). More fundamentally, some
go further, claiming (or more often, just assuming) that a simpler theory is
somehow more likely to be true (or closer to the truth) than a more com-
plex theory. For some surveys on the philosophy of simplicity see Refs. 4
and 5.
In this paper I will argue (along side others, including those in Refs. 6
and 7) that, in general, there is no reason to suppose that the simpler theory
is more likely to be true. In other words simplicity does not tell us anything
about underlying ‘model bias’.a In this I am saying more than just that
simplicity is not necessarily truth-indicative. For although I admit there
may be special circumstances where simplicity is truth-indicative, I do not
~~~~ ~ ~

aModel bias is that effect that the form of a model has upon its efficacy, for example
using a series of sinusoid functions to model some data rather than a polynomial of some
order.

65
66

see that there is any evidence that simplicity and truth are fundamentally
related. By analogy, there may be special circumstances where the colour
of an object is an indication of its mass (e.g. when someone has gone round
calibrating weights and painting them accordingly) but, in general, colour is
not at all an indication of mass. To say only that “colour is not necessarily
an indication of mass” would be highly misleading, for unless somehow
contrived, colour and mass are completely unrelated. I claim that there is
no more connection between simplicity and truth than between colour and
mass - they are unrelated properties that are only correlated where this
is contrived to be so (i.e. by prior assumption or arrangement).
Thus, in particular, when t w o theories have equal explanatory power
there is no particular reason to prefer the simpler other than convenienceb
- it may be that the more complex turns out to more useful in the future,

for example by being more amenable to generalisation or explaining new

data.
Our preference for simpler theories (apart from their obvious practical
advantages such as ease of manipulation) can be explained by the facts
that, firstly, humans are known to elaborate unsuccessful theories rather
than attempt a thorough revision and, secondly, that a fixed set of data
can only justify adjusting a certain number of parameters to a given degree
of precision. No extra tendency towards simplicity in the natural world
is necessary to explain our preference for simpler theories. Thus Occam’s
razor eliminates itself (when interpreted in this form).
I will start by arguing that a tendency towards elaboration along with
the pragmatic advantages are sufficient to explain our preference for simper
theories. Then I will briefly look at a couple of a priori arguments that
supposedly justify a bias towards simplicity. I follow this by reviewing some
evidence as to whether simpler theories are likely to be true taken from the
field of Machine Learning, since this is a case that helps distinguish between
induction in general and induction by humans. This is followed by a section
discussing some special cases where we do have some reason to expect there
to be a bias towards simplicity. I briefly consider some of the concepts that
have been called “simplicity” in the literature. I conclude with a plea for
the abandonment of the use of simplicity as a justification for modelling
decisions.

than the peculiar case where one has a limited set of data to fit as discussed in
the section on special cases below.
 67

2. Elaboration
If one has a theory whose predictions are insufficiently accurate to be ac-
ceptable, then it is necessary to change the theory. For human beings it
is much easier to elaborate the theory, or otherwise tinker with it, than to
undertake a more radical shift (for example, by scrapping the theory and
starting again). This elaboration may take many forms, including: adding
extra variables or parameters; adding special cases; putting in terms to
represent random noise; complicating the model with extra equations or
rules; adding meta-rules or models; or using more complicated functions.
In Machine Learning terms this might be characterised as a preference for
depth-first search over breadth-first search.
Classic examples of the elaboration of unsatisfactory theories include
increasing the layers of epi-cycles to explain the observations of the orbits
of planets in terms of an increasing number of circles and increasing the
number of variables and equations in the national economic models in the
UK. In the former case the elaboration did increase the accuracy on the
existing data because the system of epi-cycles can fitted arbitrarily well
to this data, but this is better done with ellipses. Given enough data
the system of epi-cycles can be used to successfully predict the orbits to
any desired degree of accuracyc, but ellipses will do it with less data and
considerably less calculation because the form of ellipses is more suited
to describing the true orbits.d Once the a priori bias towards circles is
abandoned the system of epi-cycles becomes pointless. In the later case of
the macro-economic models the elaboration did not result in the improved
prediction of future trends', and in particular these models have failed to
predict all the turning points in the UK economy.e
Why humans prefer elaboration to more radical theory change is not
entirely clear. It may be that it is easier to understand and predict the effect
of minor changes to the formulation of a theory in terms its content, so that,
if one wants to make a change where one is more certain of improvement,

'One needs t o be a little careful here, unlike ellipses, the system of epi-cycles does not
provide of itself any information about the future course of orbits, but given a postulated
orbit shape (including ellipses) epi-cycles can be used t o express such orbits and hence
used in accurate prediction of the future courses of the planets as well as ellipses.
dBeyond a certain point they will do better than an ellipse because they will be able t o
include the Eisteinian corrections, however this can still be done easier with corrections
t o an ellipse.
=Although one was apparently predicted, but this was due t o intervention by the modeller
on the basis of his expert knowledge (that destocking that can occur after an oil shock)
for details about this see Ref. 8.
68

minor changes are a more reliable way of obtaining this. In this case a
more certain but marginal improvement may well be preferred to highly
uncertain significant improvement. Alternatively, it may be that using a
certain model structure biases our view because we get used t o framing our
descriptions and observations in this way, using variations of the model as
our ‘language’ of representation - akin to Kuhns ‘theoretical spectacle^'.^
In other words, once we have started to think about some phenomena in
terms of a particular model, it becomes difficult to think about it in other
ways. Finally, it may be due to simple laziness - a wish to ‘fit’ the current
data quickly rather than going for longer-term fundamental success ( e g
prediction on unseen data).
Regardless of the reasons for this tendency towards elaboration, we are
well aware of this tendency in our fellows and make use of this knowledge.
In particular, we know to distrust a theory (or a story) that shows signs
of elaboration - for such elaboration is evidence that the theory might
have needed such elaboration in the past - for example because it had
a poor record with respect to the evidence. Of course, elaboration is not
proof of such a poor record. It may be that the theory was originally
formulated in an elaborate form before being tested, but this would be an
unusual way for a human to proceed. Thus when presented with alternative
theories developed by our fellows, one simpler than the other, we may well
guess that the complex one has been elaborated and this would be some
(albeit fallible) evidence that it has needed such elaboration. Here it is
knowledge of the (human) process that produced the theories that informs
us implicitly about their past record against the evidence and past record
against evidence is the only guide we have to the future performance of
theories.
This knowledge, along with an understandable preference for theories
that are easily constructible, comprehensible, testable, and communicable
provide strong reasons for choosing the simplest adequate theory presented
to us. An extra tendency for simplicity to be, of its own, truth-indicative
is not needed to explain this preference.
In addition to this preference for choosing simpler theories, we also have
a bias towards simpler theories in their construction, in that we tend to start
our search with something fairly simple and work ‘outwards’ from this point.
This process stops when we ‘reach’ an acceptable theory (for our purposes)
- in the language of economics we are ‘satisficers’ rather than ‘optimisers’f.

fAn ‘optimiser’is someone who searches for the best solution, whilst a ‘satisficer’accepts
 69

This means that it is almost certain that we will be satisfied with a theory
that is simpler than the best theory (if one such exists, alternatively a
better theory). This tendency to, on average and in the long term, work
from the simpler to the less simple is partly a consequence of the fact that
there is a lower bound on the simplicity of our constructions. This lower
bound might be represented by single constants in algebra; the empty set
in set theory; or a basic non-compound proposition.
An alternative approach might be to start from a reasonably complex
theory an look at modifications to this (in both simpler and more complex
directions). For example in ethology one might start from a description of
how an animal appeared to behave and then eliminated irrelevant aspects
progressively. In this case, unless one had definite extra evidence to the
contrary the sensible bias is towards the original, more complex, account.
It is important to constrain our theorising in order that it be effective, but
one does not have to use simplicity as this constraint. In" we argue for
the use of evidence to determine the starting point for model adaptation
rather than simplicity.

3. A Priori Arguments
There have been a number of a priori arguments aimed at justifying a
bias towards simplicity - I discuss two of these below. It is impossible
to disprove all such arguments in a single chapter so I will confine myself
to these two and then make some more general arguments why any such
attempt is likely to be mistaken.
Kemeny" makes an argument for preferring simpler theories on the
presumption that there is a sequence of hypotheses sets of increasing com-
plexity and that a completely correct hypotheses exists - so that once one
has reached the set of hypotheses that contains the correct one it is not
necessary to search for more complex hypotheses. Thus once the evidence
shows one has an acceptable theory one should not look for more complex
theories. However this does not show that this is likely to be a better or
more efficient search method than starting with complex hypotheses and
working from there. The conclusion is merely a reflection of the assump-
tions he has made. It also does not deal with the case where there is not
a single completely 'Lcorrect7'theory but a series of theories of increasing
complexity that more precisely fit the data as they get more complex.

a solution that meets a given standard.

70

Li and Vitbnyi12 show that it is possible to code hypotheses so that

the shorter codes correspond to lower measures of complexity so that this
coding can be interpreted as prior probabilities (in the Bayesian sense).
Thus they suggest that this is a sensible a priori basis for Bayesian prior
estimates of hypothesis probability. However, in this case, there is no nec-
essary relation between the complexity of the hypotheses and the length of
the codes that is evident before this relation to probabilities is established.
Just because such numbers can be interpreted as prior probabilities is not
sufficient reason for supposing that it is sensible to do so.
As an indication that all arguments that simplicity is a priori likely to
be truth-indicative are wrong, consider the following thought experiment.
In this experiment there are two lkg masses, A and B, of the same weakly
radioactive material, in which atoms currently decay at an average rate of
1 atom per minute. By each mass there is a Geiger counter which detects
when an atom in the mass decays and sends a particle towards the counter.
The task is to predict which counter will register a particle first after each
hour on the clock begins. Now any model which predicts A and B half the
time will, in the long run, do equally well. In this case it is abundantly
clear that simpler theories are not more likely to be correct - correctness
is determined by the proportion of A and B that the theory predicts and
nothing else.
Now, quite reasonably, one might object that a sensible model concern-
ing radioactive decay is not a directly predictive one but one which specifies
the unpredictability of the phenomena and concentrates on ‘second-order’
properties such as the probability distribution. However, this is beside the
point - it is a truism to say that those phenomena where our simple theo-
rising succeeds have some simple behaviour and those where such theories
do not hold require more complex ones. If the thesis that simplicity is truth-
indicative is restricted to only those aspects of the natural world where it
works, it has some legitimacy but this is merely begging the question - it
can not then be invoked to justify the selection of theory about phenomena
where it is not known before hand that the right theory is simple. For ex-
ample, we rightly do not attempt to predict the exact position of each grain
of sand with our mathematical models of sand piles but instead concentrate
on those aspects of that are amenable to our modelling techniques, such as
relation between the frequency and size of a~a1anches.l~ In general we are
highly selective about what we attempt to model -we usually concentrate
upon that tip of the natural world iceberg which is not too complex for our
abilities. In fact there is evidence that we find more complex patterns more
 71

difficult to per~eive’~, so that we tend to notice the simpler patterns and

attribute the more complex ones to meaningless background or randomness.
Theoretical results in Machine Learningl5>l6show that, in general, no
learning or search algorithm (i.e. method of induction) is better than an-
other. In particular that if a bias towards simplicity is sometimes effective,
there must be other domains in which it is counter-productive. To gain any
improvement in inductive ability one must apply knowledge about the par-
ticular domain one is concerned with. However, these results are extremely
abstract and dominated by search spaces that are seemingly random and
discontinuous almost everywhere. These theorems do not rule out the pos-
sibility that Nature is biased towards producing data that is more amenable
simple than these extreme cases. Thus we look to some evidence as to this.

4. Some Evidence from Machine Learning

We have two explanations for our preference for simpler theories once the
pragmatic advantages are factored out (all evidence being equal): firstly,
our knowledge that theories tend to be elaborated when unsuccessful and,
secondly, that there is an inherent bias towards simplicity in the natural
world. If we were to hold to Occam’s razor (in the form that simplicity is
truth-indicative) then we would choose the first because this is sufficient to
explain the phenomena - the postulated bias in the natural world is an
‘unnecessary entity’.
Since I don’t hold with this form of Occam’s razor I need to look for some
evidence to distinguish between the two explanations. Since the tendency
towards elaboration is a characteristic of a lot of human theory construction,
the critical evidence needs to come from non-human theory construction
- situations where theory construction is not based upon a process of

elaboration to see if simplicity is truth-indicative there. If a bias towards

simplicity was useful apart from our human prelediction for it, this would
support its general use. Recently there have been such studies in the field
of Machine Learning - where a computer program (rather than a human)
attempts the induction. This gives one a test bed, for one can design the
induction algorithm to use a simplicity bias or otherwise and compare the
results. In one of these studied7 a comprehensive evaluation of all possible
theories in a given formal language (to a given depth) were analysed against
some real-world data series as follows: firstly as to their effectiveness at
fitting some initial portion of the data (the in-sample part of the series),
secondly as to their success predicting the continuation of this data (the
72

out-of-sample part), and finally, as to the theory’s complexity (measured

in this case by the size or depth of the formal expression representing the
theory). The theories that were best at fitting the in-sample data were
selected. Within this set of ‘best’ theories it was examined whether the
simpler theories predicted the out-of-sample data better than the more
complex theories. In some cases the simpler hypotheses were not the best
predictors of the out-of-sample data. This is evidence that on real world
data series and formal models simplicity is not necessarily truth-indicative.
In a following study on artificial data generated by an ideal fixed
‘answer’18, it was found that a simplicity bias was useful, but only when
the ‘answer’ was also simple. If the answer was complex a bias towards
complexity aided the search. This result can be interpreted as exploiting
available information about the solution in the search for it - the same
lesson that one can draw from the “NO Free Lunch” theorem16.
Webblg exhibited an algorithm which systematically extended decision
trees (making them more complex) so that they gave the same error rate on
the in-sample data, and, on average, gave smaller error rates on the out-of-
sample data for several real-life time series. This method was based upon
a principle of similarity, which was used to restrict the set of considered
hypotheses. A useful survey of results in Machine Learning is Ref. 20
which could be considered a parallel paper to this one, concentrating on
the usefulness of a simplicity bias in Machine Learning and only touching
upon the wider, philosophical, issue.
Thus there is some evidence that, when considering non-human induc-
tion, a simplicity bias is not generally helpful or truth-indicative. Rather it
seems that it is often used as an ill-defined stand-in for domain knowledge
- perhaps this is a reflection of our own ~ 0 g n i t i o n . lThe
~ evidence does
not say that a biased towards simplicity is never useful but is sufficient to
exclude it being routinely assumed.

5. Special Cases
Although, there is no reason to suppose simplicity is in general truth-
indicative, there are special circumstances where it might be. These are
circumstances where we already have some knowledge that would lead us
to expect that the solution might be simple. That is the evidence points to
a particular theory (or class of theories) and those happen to be simple. I
briefly consider these below.
The first is when the phenomena are the result of deliberate human
 73

construction. Deliberate human constructions are typically amenable to an

almost complete analysis assuming a 'design stance', that is considering the
construction from the point of view of it being designed.21 That is, that
they are frequently modular, and the result of relatively simple principles
iterated many times. If someone asks you to guess the next number is the
sequence: 2 ,4,8, 16 you will correctly guess 32, because the nth power of
two is the most familiar pattern that describes these five numbers, and you
an rely on the fact that the human will have chosen such a pattern and
that those patterns tend to be simple (if possibly obscure). It would not be
sensible to guess the number 31, despite the fact that there is a rule that
would make this the correct answer.g However here we already have some
knowledge about what kinds of rule/pattern people are likely to choose and
that these are more likely to be simple.
The simplicity of these kinds of phenomena is only a hallmark of de-
liberate, conscious human construction. Products of our unconscious brain
or social constructs such as language may be extremely complex for these
were not the product of an intentional design process. Thus artists may
construct extremely complex artefacts because they do not consciously de-
sign every detail of their work but work intuitively a lot of the time with
parts and media that are already rich in complexity and meaning.
Apart from human construction there are other circumstances where one
has good reason to expect simplicity, namely the initial stages of processes
that start with the simplest building blocks and work from there. That
is the underlying process is known to be one of elaboration. Examples of
these might include the construction of progressively higher elements in the
early universe, the reactions of bacteria to external stimuli, or, possibly, the
first stages in the evolution of life. Evolution started from relatively simple
organisms and evolved from there. Obviously when life started the intro-
duction of variety by mutation would be unlikely to result in simplification,
since the organisms were about as simple as they could get while still being
able to reproduce in their environment. Thus the effective lower bound on
complexity of viable organisms means that there is a passive drift towards
greater complexity (as opposed to an active drive towards complexity22).
However this bias is only significant at the start of the process because the
space of possible organisms is so great that once an organism has evolved
to be somewhat complex it is almost as likely to evolve to be simpler as

~~~

gFor example, the number of areas that n straight lines, each crossing the perimeter of
a circle twice and such that no three lines intersect in a single point, cut that circle into.
74

more complex - the lower bound and the ‘inhabited’ part of the possibility
space do not impinge upon the possibilities that much so as to significantly
bias its evolution towards complexity.
Another situation is where one already knows that there is some correct
model of some minimum complexity. In this case one heuristic for finding
a correct model is to work outwards, searching for increasingly complex
models until one comes upon it. There are, of course, other heuristics - the
primary reason for starting small are practical; it is far easier and quicker
to search through simpler models. In more common situations it might
be the case that increasingly complex models may approximate the correct
model increasingly, but never completely, well or that no model (however
complex) does better than a certain extent. In the first case one is forced
into some trade-off between accuracy and convenience. In the second case
maybe no model is acceptable, and it is the whole family of models that
needs to be changed.
Clearly if one has some information about the complexity of the sought
after model before the search starts, using that information can make search
more efficient, but then this is merely a case of exploiting some knowledge
about the solution - it is not an reason for a general bias towards simplicity
in other cases. In such circumstances as those above there is some reason
to err towards simplicity. However in these circumstance the principle is
reducible to a straight forward application of our knowledge about the
phenomena that leads us in that direction - principles of simplicity do not
give us any ‘extra’ guidance. In these circumstances instead of invoking
simplicity as a justification the reason for the expectation can be made
explicit. Simplicity as a justification is redundant here. Likewise if one had
some evidence that the desired model or theory was complex before the
search starts, a bias away from simplicity would be helpful.

6. Versions of “Simplicity”
In order to justify the selection of theories on the basis of simplicity,
philosophers have produced many accounts of what simplicity as. These
have included almost every possible non-evidential advantage a theory
might have, including: number of parameter^^^; extensional p l ~ r a l i t ;y ~ ~ ~ ~ ~
fal~ifiability~~;
likelihood26,6; stability27 logical expressive power28; and
content2g. These approaches give an uncomfortable feeling of putting the
cart before the horse - for instead of deciding what simplicity could mean-
ingfully be and then seeing if (and when) it is useful to be biased towards
 75

simpler models when trying to explain or predict some observations, they

seem to have started with the presumption that a bias towards simplicity is
right and then try to find an account of simplicity that is compatible with
this idea. In some cases this has almost come full circle. Sober3’ charac-
terises simplicity as informativeness - so that instead of asking whether
simplicity is informative he seeks to show that simplicity (as informative-
ness w.r.t. a specified question) is, in fact, simple!
As a result of the prior philosophical need for a definition of “simplicity”
to represent a useful bias in model/theory selection, these definitions do not
tend to be properties that are the opposite of accounts of complexity. This,
in itself, is evidence for the failure of the project to justify a bias away from
complexity - it is not the case that a reasonable definition of complexityh
is the starting point and then this shown to be counter truth-indicative,
but that properties that have nothing to do with a lack of complexity are
labelled as “simplicity” in an apparently desperate attempt to justify this
assumed bias.
If, as I have argued, simplicity is not truth-indicative, this whole en-
terprise can be abandoned and the misleading label of ‘simplicity’removed
from these other properties. This epithet transferral, far from producing
insight has produced a fog of differing ‘simplicities’and ‘complexities’which
do much to hinder our understanding of the modelling process. Theories
can posses a lot of different advantages that are not directly linked to its
success at explaining or predicting the evidence, restoring the correct labels
for these advantages will help (rather than hinder) the elucidation of their
different and interacting roles.

7. An Example - Curve fitting by parameterisation

A particular case of hypothesis selection that has been discussed in the
literature is curve fitting. This is the case of deciding which of a variety
of hypotheses (in different functional curve ‘forms’) one will select given a
set of data (typically given as data points). Typically these ‘forms’ include
parameters that are adjusted to fit the data, so that each form corresponds
to a family of curves, and the member of that family determined by the
choice of values for the parameters.
Curve fitting can be a misleading example as it can be difficult to rid
oneself of ones intuitions about what sort of curves are useful to posit. One

hFor this reason a definition of complexity is not relevant here. For those who are
interested in what might be a reasonable definition of complexity see Ref. 31.
76

can have strong visual intuitions about the suitability of certain choices
which strongly relate to a set of heuristics that are effective in the domains
one happens to have experienced. These intuitions might not be helpful in
general.
In particular, one might happen to know that there is likely to be some
random noise in the data, so that choosing a curve that goes through every
data point is not likely to result in a line that reflects the case when more
data is added. In this case one might choose a smoother curve that might
approximate to a local average of the data. A traditional method of smooth-
ing is choosing a polynomial of a lower order or with fewer parameters. This
is not, of course, the only choice for smoothing; one might instead use, for
example, local regression32 where the fitted curve is a smoothed combina-
tion of lines to fit segments of the data. Thus an assumption that a curve
with a simpler functional form might be more appropriate for a certain set
of data can depend on: firstly, that one has knowledge about the nature of
the noise in the data and, secondly, that one chooses the simplicity of the
functional form as ones method of smoothing. If, on the other hand, one
knew that there was likely to be a sinusoid addition to the underlying data
one might seek for such regularities and separate this out. Here a preference
for simplicity is merely an expression of a search bias which encodes one’s
prior knowledge of the domain.
A recent series of argues that simplicity is justified on the
grounds that its use can result in greater predictive accuracy on unseen
data. This is based on results obtained in Ref. 35. Simplicity in this case is
defined as (effectively) the Vapnik-Chervonenkis (VC) dimension36 of the
set of curves which in very particular circumstances is equivalent to the
number of adjustable parameters in the equation form. The advantages of
‘simplicity’ in this account amount to the prescription not to try and fit
more parameters that you have data for, since the larger the set of hypothe-
ses one is selecting from the more likely one is to select a bad hypothesis
that ‘fits’ the data purely by chance. The extent of this overfitting can be
estimated in particular cases. The argument is that in these cases one can
know to choose a form with fewer parameters if one does not have enough
data to justify estimating any more, even if such more complex forms may
appear to fit the data better. However, this does not effect the general ar-
gument - if you have two models whose predictive accuracy, once adjusted
f o r its expected overfitting o n limited data, is equal then there would be no
reason to choose the family which might be considered simpler to have a
simpler form. In circumstances with a fixed amount of data the estimation
 77

of the extent of overfitting might or might not tip the scales to lead one to
select the simpler model.
This account gives no support for the thesis that the simplicity of a
model gives any indication as to its underlying model bias. Thus, in cir-
cumstances where one can always collect more data (so that effectively
there is an indefinite amount of data), these arguments provide no reason
to select a simpler model, but rather suggest one should collect more data
to distinguish which model is better in general. In this case, the decision of
when to stop seeking for a model which gives increased predictive accuracy
is a practical one: one has to balance the cost of collecting the additional
data and using it to search for the most appropriate model against the
goodness of the parameterised model.
Also the connection between the VC dimension and any recognisable
characteristic of simplicity in the family of curves is contingent and tenuous.
In the special case where the only way of restricting the VC dimension (or in
finite cases, the number of hypotheses) is through the number of adjustable
parameters, then it is the case that an equational form with more adjustable
parameters will require more data for accurate parameterisation. However
there are other ways of restricting the set of hypotheses to reduce the VC
dimension; as discussed abovelg successfully used a similarity criterion to
do this. Thus one can avoid overfitting by restricting the VC dimension of
the set of hypotheses without using any criteria of simplicity or parsimony
of adjustable parameters. In a related study3’ examined the connection
between the complexity of expressions and (indirectly) their ambiguity and
concluded that any measure that restricted the space of models would be
equally effective. Of course, one can decide to define simplicity as the VC
dimension, but this is begging the question again and one would need to
justify this transferred epithet.
To summarise this section, there is a limit to the accuracy with which
one can adjust a given number of unknown parameters given a certain, jixed
amount data - one is only justified in specifying a curve to the extent
that one has information on which to base this. Information in terms of
a tightly parameterised curve has to come from somewhere. However, in
the broader picture where different families of curves are being investigated
(for example, by competing teams of scientists continually searching out
more data) these considerations give no support to the contention that
the simpler family has an advantage in terms of predicting unknown data
better.
78

8. Concluding Plea
It should be clear from the above that, if I am right, model selection ‘for
the sake of simplicity’ is either: simply laziness; is really due to practical
reasons such as cost or the limitations of the modeller; or is really a rela-
belling of more sound reasons due to special circumstances or limited data.
Thus appeals to it should be recognised as either spurious, dissembling or
confusing and hence be abandoned.
Indeed rather than assuming that there will always be a simple adequate
theory (if only we were clever enough and worked hard enough to find it)
we should keep an open mind. On the face of it, it would seem reasonable
to expect that complex phenomena will require complex theories for many
purposes, and that simple theories will only be adequate for a restricted
range of phenomena. However we should always allow for the possibility
that occasionally some apparently simple phenomena will require complex
theories and apparently complex phenomena allow for simple theories. The
point is one can not tell in advance and so it is unwise to make assumptions
about this.
However, there is a form of Occam’s Razor which represents sound ad-
vice as well as perhaps being closer to its Occam’s original formulation
(usually rendered as “entities should not be multiplied beyond necessity”),
namely: that the elaboration of theory in order to fit a known set of data
should be resisted, i.e. that the lack of success of a theory should lead to a
more thorough and deeper analysis than we are usually inclined to perform.
It is notable that this is a hallmark of genius and perhaps the reason for the
success of genius - be strict about theory selection and don’t stop looking
until it really works.

Acknowledgements
I would like to thank all those with whom I have discussed or commented
upon these ideas (most of whom were not inclined to change their opin-
ions on this matter), including: the participants of the 20(th)Wittgenstein
Symposium in 1999; and those at the workshop on the Evolution of Com-
plexity at the VUB in 1997. Also to several anonymous referees who have
motivated me to make my position clearer (if more stark).

References
1. Newton, I. Philosophia naturalas Princapia Mathematica, (1686).
 79

2. Einstein, A., Relativity: the special and the general theory - a popular ex-
position, London: Methuen, (1920).
3. Gertsbakh, E. and Gertsbakh, I., Measurement Theory for Engineers,
Springer, (2003).
4. Hesse, M., Simplicity, in P. Edwards (ed.), The Encyclopaedia of Philosophy,
vol. 7., New York: Macmillan, 445-448, (1967).
5. Zellner, A., Keuzenkamp, H. and McAleer, M. (eds.), Simplicity, Inference
and Modelling. Cambridge: Cambridge University Press, (2001).
6. Quine, W. V. 0. Simple Theories of a Complex World. In The Ways of
Paradox. New York: Random House, 242-246, (1960).
7. Bunge, M., The Myth of Simplicity. Englewood Cliffs, Prentice-Hall, (1963).
8. Moss, S., Artis, M. and Ormerod, P., A Smart Macroeconomic Forecasting
System, The Journal of Forecasting 13,299-312, (1994).
9. Kuhn, T. S. The Structure of Scientific Revolutions. Chicago, University of
Chicago Press, (1962).
10. Edmonds, B. and Moss, S., From KISS t o KIDS - an anti-simplistic mod-
elling approach. In P. Davidsson et al. (eds.): Multi Agent Based Simula-
tion 2004. Springer, Lecture Notes in Artificial Intelligence, 3415, 130-144,
(2005).
11. Kemeny, J. G. Two Measures of Complexity. The Journal of Philosophy, 52,
722-733, (1953).
12. Li, M. and VitBnyi, P. M. B. Philosophical Issues in Kolmogorov Complexity,
in Automata, Languages and Programming, 19th International Colloquium,
Lecture Notes in Computer Science, 623,1-15, Springer, (1992).
13. Bak, P. How Nature Works: The Science of Self Organized Criticality. Ox-
ford, Oxford University Press, (1997).
14. Charter, N. The Search for Simplicity: A Fundamental Cognitive Principle?
The Quarterly Journal of Ezperimental Psychology, 52A,273-302, (1999).
15. Schaffer, C., conservation law for generalization performance. In Proceedings
of the llth International conference on Machine Learning, 259-265. New
Brunswick, NJ: Morgan Kaufmann, (1994).
16. Wolpert, D. The lack of a priori distinctions between learning algorithms.
Neural Computation, 8, 1341-1390, (1996).
17. Murphy, P.M.; Pazzani, M.J. Exploring the Decision Forest: an empirical
investigation of Occam’s razor in decision tree induction, Journal of Artificial
Intelligence Research, 1, 257-275, (1994).
18. Murphy, P. M. An empirical analysis of the benefit of decision tree size biases
as a function of concept distribution. Technical report 95-29, Department of
Information and Computer Science, Irvine, (1995).
19. Webb, G . I. Further Evidence against the Utility of Occam’s Razor. Journal
of Artaficial Intelligence Research, 4, 397-417, (1996).
20. Domingos, P. Beyond Occam’s Razor: Process-Oriented Evaluation. Machine
Learning: ECML 2000, llth European Conference on Machine Learning,
Barcelona, Catalonia, Spain, May 31 - June 2, 2000, Proceedings, Lecture
Notes in Artificial Intelligence, 1810, (2000).
21. Dennett, D. C., The Intentional Stance. Cambridge, MA: A Bradford Book,
80

MIT Press, (1987).

22. McShea, D. Meatzoan Complexity and Evolution: is there a trend? Ewolu-
tion, 50, 477-492, (1996)
23. Draper, N. R.; Smith, H. Applied Regression Analysis. New York: John Wi-
ley, (1981).
24. Goodman, N. The Structure of Appearance. Indiapolis: Bobbs-Merrill,
(1966).
25. Popper, K. R. Logic of Scientific Discovery. London: Hutchinson, (1968).
26. Rosenkrantz, R. D. Inference, Method and Decision. Boston: Reidel, (1976).
27. Turney, P. The Curve Fitting Problem: A Solution. British Journal for the
Philosophy of Science, 41, 509-530, (1990).
28. Osherson, D.N. and Weinstein, S. On Advancing Simple Hypothesis. Philos-
ophy of Science, 57, 266-277, (1990).
29. Good, I. J. Corroboration, Explanation, Evolving Probability and a Sharp-
ened Razor. British Journal f o r the Philosophy of Science, 19, 123-43, (1969).
30. Sober, E. Simplicity. Oxford: Clarendon Press, (1975).
31. Edmonds, B., Syntactic Measures of
Complexity. Doctoral Thesis, University of Manchester, Manchester, UK,
(1999). (http://bruce.edmonds.name/thesis/)
32. Cleveland W. S., Devlin S. J., Grosse E. Regression By Local Fitting - Meth-
ods, Properties, And Computational Algorithms. Journal Of Econometrics,
37, 87-114, (1988).
33. Forster, M. and Sober, E. How to Tell when Simpler, More Unified, or Less
Ad Hoc Theories will Provide More Accurate Predictions. British Journal
fo. the Philosophy of Science, 45, 1-35, (1994).
34. Forster, M. Model Selection in Science: The Problem of Language Invariance.
British Journal for the Philosophy of Science, 50, 83-102, (1999).
35. Akaike, H. Information Theory and an Extension of the Maximum Likelihood
Principle, in Petrov, B. N. and Csaki, F. (eds.) 2nd International Symposium
on Information Theory, 267-281. Budapest: Akademai Kiado, (1973).
36. Vapnik V. N. and Chervonenkis A. Y. , Necessary And Sufficient Conditions
For The Uniform-Convergence Of Means To Their Expectations, Theory Of
Probability Applications, 26, 532-553, (1981).
37. Pearl, J. On the Connection Between the Complexity and Credibility of In-
ferred Models, International Journal of General Systems, 4, 255-264, (1978).
CAN THE WHOLE BE MORE THAN THE COMPUTATION
OF THE PARTS? A REFLECTION ON EMERGENCE

CAMILO OLAYA
Institute of Management
University of St. Gallen (HSG)
Dufourstrasse 4Oa, CH-9000,St. Gallen, Switzerland
E-mail: camilo.olaya@unisg.ch

The term ‘emergence’ is frequently found in Complexity Science; however, it gives

the impression that earlier debates on this topic are not taken fully into account. By
considering this situation, the article underlines the need of a consistent discussion
when addressing emergence if this term is going t o have an added and meaningful
value for learning about complex systems. The promising suitability of process
thought and the significance of detailed rulebased modeling are derived from this
reflection as further paths t o explore.

1. Introduction
Lately everything seems to emerge. Qualities, properties, matter, life,
mind, ideas, inventions, computed macro-states, and many other things,
now ‘emerge’. In particular, in what is loosely known as Complexity Sci-
ence (CS),” the term ‘emergence’ is widely invoked as a pivotal notion,
although it is used in diverse senses. And it is not the purpose here to offer
a new definition; the article presents important distinctions that usually
are absent in CS, regardless of been heavily discussed in other areas, in
particular in philosophy. The central point is that emergence, taken as a
metaphysical inquiry, seems to be overlooked when developing explanations
by means of computer simulation. However, as John Holland reminds us -
in his book ‘Emergence’,’ CS is committed to make models of the world.

“In order to frame the subject, somehow, CS refers here t o approaches related t o non-
linear dynamics, agent-based modeling, chaos theory, artificial life, social simulation,
complex adaptive systems, system dynamics, and alike, that claim t o address ‘complex-
ity’ using typically computer simulation. Yet this does not exclude indirect connections
to related areas that also address emergence like molecular and evolutionary biology,
quantum theory, matter theory, and the philosophy of mind.

81
a2

W h a t type of explanation regarding emergence can CS really deliver? The

paper introduces a reflection on this question. As a result, an ontologi-
cal shift to process philosophy arises as a natural option so as to address
novelty and change, the motivating concerns behind the term ‘emergence’.
The suitability of bottom-up computer modeling is particularly underlined,
arising from this shift.

2. The Emergent Question

Emergence still remains as one of the hardest notions to demarcate beyond
the sentence ‘the whole is more than the sum of the parts’. Usually the word
‘more’ refers to unpredictable novelty. It has produced an intense debate
since we have to face the appealing paradox: How can something that arises
from a set of parts be more than the sum of those parts? Fundamentally this
question is a metaphysical inquiry, or at least it used to be so; the original
thesis proposed the generation of new units, new kinds, new entities, new
qualities. This task has been historically cumbersome, to say the least.
Jones2 recently remarks:

Emergence is largely used as an explanation: such and such an ob-

ject - ranging from atoms through multicellular organisms to con-
sciousness - is an emergent property of some ensemble of parts. But
this leaves an inadequate level of understanding The term emer-
gence has a somewhat equivocal press. For some it is an explana-
tion for the appearance of wholes or coherent objects in any system
of things or understandings; while for others it is either meaningless
or, conversely, an invocation of something mystical (p. 418).

The heated dispute may have been started in 1920 with Samuel Alexan-
der for whom emergence necessarily admits no explanation; he claimed that
it has to be accepted as a ”brute empirical fact” (as cited by Pap).3 The
reply of Stephen Pepper in 1926 posts the warning sharply: either it is the
sum of the parts or it is not4 (no paradox here). Looking again at the quote
above of Jones, it seems there has not been too much advance since then.
What follows is a brief account of some important distinctions through this
philosophical debate;b these distinctions (highlighted in italics) are rele-
vant today, on a first basis because many of them seem to be trivialized or
overlooked.

bFor comprehensive overviews see e.g. [5], [6].

 a3

Going back to 1926, Pepper, ignoring the advice of Alexander, addresses

the subject. First, he identifies emergence as a kind of change; second, this
change, he summarizes, is derived from three propositions: (i) there are
different levels of existence in terms of degrees of integration; (ii) there are
marks that make these levels distinguishable from one another over and
above the degrees of integration; (iii) these marks can not be deduced from
marks at other higher or lower levels. He focuses on laws as the means to
describe what emerges, though he remarks that "accurately speaking, we
must first observe, laws can not emerge. Emergence is supposed to be a
cosmic affair and laws are descriptions" (p. 242).4 Pepper challenges the
implicit puzzle of the last two premises: On the one hand, (ii) affirms that
there is cumulative change; however, he argues that if we work with laws
then what take place are shifts at levels of description, where variables can
always be distinguished and placed at proper levels. On the other hand,
(iii) affirms that change is not predictable; Pepper asserts that either change
is predictable (as in "part of the system") or it is epiphenomena1 since "if
they represent irreconcilable inconsistencies in the physical system, they are
not laws at all, but statements of chance occurrences" ( ~ . 2 4 4 )The
. ~ alter-
native to laws is to a r m that what emerge are qualities (sensory qualities
- metaphysical simples); Pepper dismisses this, rejecting such approach as a
theory of epiphenomenalism (no added value) and therefore unsatisfactory
for metaphysics. Therefore, for him emergence is meaninglessc
Soon after in 1929, Watts Cunningham' underlines that novelty should
not imply unintelligibility, i.e. if we have dynamic systems whose very na-
ture change with time, there are emergents which, though non-predictable,
can be intelligible; he affirms this assertion for both substancial emergence
(a constituent or entity appears de novo) and structural emergence (a novel
arrangement of old material). In 1939 Ablowitz conceives the emergence
of unpredictable and intelligible qualities as due to a new togetherness of
the partsg (hence structural emergence). However Malisoff, in the same
year, emphasizes the level of description defined by the observer (thus, we
can have alternative descriptions depending on the degree of organization
that the observer attributes to the system); this eliminates the mystery
on the question, he argues, since it is subject to the mode of observation,
i.e .relative to particular specifications or postulates." Stace replies that
emergence produces novelty, but what is interesting, he underlines, is abso-

C A full discussion and criticism of the argument of Pepper is developed by Meehl and
~el~ars.'
84

lute novelty (anything which had never appeared before) and not relative
novelty (relative to a particular situation).l' Gotshalk, in 1942, emphasizes
that novel emergents are irreducible and that they bring into prominence
individuality and quality; for him emergents are underivative ingredients of
the real and not merely causal reductions;12 it is implicitly a criticism to
Pepper (emergence now is meaningful) and Malisoff (for whom the question
is only epistemological). Also in 1942 Campbell Garnett emphasizes that
the failure of the theory of emergence to establish a vera causa (as it is was
suggested by Alexander) implies the incompatibility with scientific meth-
ods; as a result, he adheres to the organismic interpretation: a single nature
within which two irreducibly distinct types of process are functionally in-
tegrated in one organic whole;13 hence, he rejects ontological emergence
since complete independence is proclaimed between layers. In the same
journal, Henle criticizes the prominent role of the observer for addressing
predictability (as in "quality A is predictable by person B on evidence C
with degree of assurance D"); for him novelty exists in the universe though
this does not entail emergence since two conditions have to be met: it must
be logically novel relative to all previously exemplified characteristics, and
it must not be merely a spatio-temporal characteristic.'* Bahm in 1947
supports organicism: things are organic unities, i.e. they are both one
and many, identity and difference, permanence and change, stability and
novelty, substance and function; for him, emergence is just a natural con-
sequence of recognizing those levels or dimensions15 (and therefore there is
no paradox). Pap in 1952 attacks the notion of absolute unpredictability
("pure metaphysical obscurantism"); he proposes the possibility of having
laws that can correlate an a priori unpredictable quality Q with certain
causal condition^.^ Berenda, in similar lines, supports the ability for un-
derstanding (structured in reason) as different from the ability for learning
(developed in time), for him prediction is just a matter of time, therefore
he rejects logical unpredictability.16
Somehow summarizing these earlier debates, Kekes in 1966 characterizes
emergence with novelty (when a token of a type appears which never existed
before) and unpredictability (this occurrence of novelty is unpredictable as
based on the theories of lower levels of existence); his type of novelty is in
the lines of Berenda, a priori unpredictability, prior to the occurrence of the
event but explicable afterwards. Kekes postulates both novelty and unpre-
dictability as necessary and sufficient conditions of emergence. Moreover,
for him qualities and laws as well can emerge, and the emergence of any of
 85

them would imply the emergence of an entity in general;d finally, he rejects

the triviality of having a type of emergence where, in spite of an adequate
lower level theory, logical (metaphysical) unpredictability persists.”
More recently, Klee proposes an unified treatment of emergence favor-
ing properties over laws as the emergent units within a micro-deterministic
view of nature, i.e. parts determine wholes; however, the preference for
properties is for practical reasons given the more complicated path, he ar-
gues, implied by a discussion around a theory of lawlikeness.18 In 1997
Bedau proposes weak emergence which is based on the necessary condition
of deriving dynamically macrostates via computer simulation; here what
emerges is a macrostate of a systemlg (though he seems to equate proper-
ties with behaviour). Recent significant questions on ontological emergence
(and non-reductive physicalism) can be found in the work of Kim; e.g. he
criticizes the notion of downward causation and he suggests the conditions
of novel causal powers and non-functionalizability for having ontological
emergen~e.~
In a needed call for order, Silberstein and McGeever2’ urge to distin-
guish between ontological and epistemological emergence; they underline
that it is common to draw fallacious conclusions on the existence of ontolog-
ical emergence (a feature of the world) based on epistemological emergence
(fundamentally a description, an explanation, an artefact).e Cunningham
in 2001 makes a similar clarification suggesting three types of emergence:
ontological emergence (based on a duality, close to the proposals of Camp-
bell Garnett of 1942 and Bahm of 1947) and two types of epistemological
emergence: multiple realizable properties (determined fully by the organi-
zation of the parts) and highly configurational and holistic properties which
is roughly similar to the notion of weak emergence of macrostates of Be-
dau since it is the power of computation that determines the possibility

dHe assumes order in nature which can be described by laws (determinism), then he
argues: ”At any rate, emergent laws describe regularities holding in nature which must
have been novel at some time. These regularities describe the behaviour of entities which
may or may not have emergent qualities; yet the regularities described by emergent laws
could not be described by laws formulated to describe the behaviour of entities on a
lower level of integration” (p. 366).
‘Van Gulick’l makes a classification using this distinction, finding four types of episte-
mological emergence and six types of ontological emergence; e.g. Georgiou” favors the
epistemological approach because for him emergent properties provide the identity of a
system since they allow an observer to say ‘this is a system of such-and-such and not
otherwise’ for they act as unifying reference points for the group of interrelated parts
which constitute the system” (p. 242).
86

of addressing system-level marks (these are properties for Cunningham);

then, he places complex dynamic systems studies in general as an example
of epistemological emergence.23
And the story goes on.
After examining the discussion, a first impression is that, in CS, either
emergence is implicitly taken as a basic principle not to be discussed (as
the proposal of Alexander and thus rather unintelligible) or it is understood
only as an epistemological issue (sometimes with hasty ontological claims).
Or even worse, sometimes it reminds a basic premise treated also as a result
or conclusion - Petitio P r i n ~ i p i i There
.~ are various complications. First
of all, regardless of the development of the previous discussions and the
diverse proposals available, ambiguity seems more the rule than the excep-
tion; e.g. sometimes emergence seems to denote metaphysical assumptions,
and sometimes it seems to point at a level of description or explanation;
and many times what we find is a sophisticated (and sophistical) muddle;
beyond a discussion around words like ‘ontology’ or ‘epistemology’, it is
manifest the confusing inconsistent use of the term, e.g. in the lines of the
warning made by Silberstein and McGeever, presented ab0ve.g This should
not be a problem as long as assumptions are held consistently taking into
account the relevance of that distinction (not very frequent in any case).
A second complication is that the emergents take a wide range of forms;
in the literature we witness the emergence of properties, behaviours, en-

Roughly speaking: ” Why is it emergent? Because it is complex”. ”Why is it complex?

Well, because it is emergent.” Apart from the risk of explaining ‘emergence’ in terms of
the perhaps more obscure term ‘complexity’, a short example may picture the panorama:
On the one hand, ”Complexity is a marvelous and essential property of our universe. The
existence of complexity provokes the emergence of behaviors full of richness but hard t o
investigate”(p. 33).24 On the other hand, ”If ‘complexity’ is currently the buzzword of
choice for our newly minted millennium, as many theorists proclaim, ‘emergence’ seems
t o be the explanation of the hour for how complexity has evolved. Complexity, it is
said, is an emergent phenomenon” (p. 18).25The problem for a research program like
CS is not only the seemingly unavoidable ambiguity of the terms as such (though a
modern Tower of Babel is not a desirable situation), the question aims at reflecting on
assumptions, meaningfulness, consistency, and relevance.
gThis illustrative example, cited in [20] may suffice: ”Chaos brings a new challenge t o
the reductionist view that a system can be understood by breaking it down and studying
each piece. This view has been prevalent in science in part because there are so many
systems for which the behaviour of the whole is indeed the sum of its parts. Chaos
demonstrates, however, that a system can have complicated behaviour that emerges as a
consequence of simple, non-linear interaction of only a few components. The interaction
of components on one scale can lead to complex global behaviour on a larger scale that
in general cannot be deduced from knowledge of the individual components” (p. 186).
 87

tities, laws, structures, events, qualities, patterns, among others, treated

in a seemingly careless way; even sometimes these terms are used on an
exchange basis. As a further consideration, in order to hold a consistent
position (specially if we are going to assume, accept, or address metaphys-
ical emergence), we can consider to challenge also our ontological ground
and not only our conceptualization of ‘emergence’, a term that as time goes
by becomes too sophisticated, more disperse, and less useful. The presented
historical account emphasizes some difficulties behind the real motivation
of the endeavour: to address novelty and change. Could an ontological shift
be a more suitable way for tackling the emergent question?

3. Explanations of Uniformity
Emergence used to be a cosmic affair (as in ”emergence is a feature of the
world”); in fact, a number of discussions within CS seem to point at absolute
emergence, but what seems to prevail - at the end - is a particular level
of description that habitually takes the form of causal accounts and/or
universal laws; such view involves a sort of uniformity that makes harder
to address novelty and change. This is presented next.
Explanations based on causal accounts hold that there is something spe-
cial about causal interactions that increases our understanding of the world;
causality is the crucial relationship which makes information explanatorily
relevant.26 An account of ontological emergence is naturally a formidable
challenge for such approach, e.g. it has been widely addressed in terms
of multiple complex causal networks or sophisticated law-like expressions.h
Silberstein and McGeever summarize the popular position:20
Complexity studies is inclusive of chaos theory, connectionism and
artificial life, in that it seeks a grand unified theory of the funda-
mental and universal laws that govern behavior of all complex and
non-linear systems. It would be fair to say that many complexity
theorists are after a unified theory of ‘emergence’ (see Kauffman
and Wolfram for examples). Complexity theorists (often) reject
part-whole reductionism, but they hold that there are fundamental
laws of the universe which pertain not only to all of the physical,
chemical and biological systems, in the universe, but also to all
systems (such as the economy and other social systems) that reach
certain degree of ‘complexity’. These ‘meta-laws’ or ‘power-laws’

hRather recent examples are [27], [28], [29].

88

govern the emergence or evolution of a system’s behavior in the

world (p. 190).

Such quests seem inappropriate for coping with emergence; for instance,
universal laws are opposed to individuality, ignoring the remark of Gotshalk
of 1942: ”The most obvious philosophical importance of the principle of
emergence, as I see it, is that it brings into intellectual prominence factors
in the universe that actually have been present and in their way important
all the time, namely, individuality and quality. In modern scientific causal
analysis, both of these factors have tended to be lost from view”12 (p. 397,
emphases added). Seemingly they are still lost from view since it can be
said that many discussions on emergence have taken an uniform nature for
granted (or explicitly as e.g. Kekes17, presented above), as it is introduced
next.
We try to make sense of a higher level ‘property’ (let’s say) called ‘emer-
gence’ that is not only in function of the parts although it is derived from
them. This should challenge some basic frameworks.’ Why? Within a
causal view, we seem to follow Aristotle where there is a necessity in causal
relations coming from the fact that they derive from the essential charac-
teristics of material things.31 A very brief note on Aristotelian properties:
”essential properties of a thing are those on which its identity depends.
They are the properties of a thing which it could never have lacked, and
which it could not lose without ceasing to exist or to be what it is. Ac-
cidental properties of a thing are properties which that thing could either
have had or lacked”(p. 374).32 Such a frame imposes strong constraints
for having a notion of emergence within a world of properties ascribed to
static substances which in turn seem to be assumed as self-identically en-
during stuff - see e.g. [33],[34]; hence substancial emergence (as in Watts
Cunningham,’ presented earlier) based on essential qualities, becomes dif-
ficult to address.
This world depicted above is usually expressed with laws: ”The state-
ment that an object follows a law involves the presumption that there is
some informant agency that makes an object to behave in a certain way

‘For instance, Humphreys suggests the examination of the common ontological minimal-
ism, a doctrine based on a small set of premises: (i) to hold a small set of fundamental
constituents of the world; (ii) the primacy of intrinsic (non-relational) properties; (iii)
all non-fundamental individuals and properties are composed from those fundamental
entities. These premises are rooted in two doctrines: a pre-eminence of the logical re-
construction of concepts, and a Humean account of causality.30
 89

and not to behave in any other way...If there is only one informant agency,
for all objects of a kind in all time, we call it a law”(p. 9).35 We can ap-
preciate why individuality and novelty are hard to meet, nature is assumed
uniform;j more specifically this use of laws neglects endogenous change,
e.g.: ”the events the law describes also do not change endogenously unless
an exogenous force is introduced into the system. The model is universally
deterministic. Given complete information about the initial and subsidiary
conditions, the law allows us to retrodict events precisely on to the past
and to predict them to precisely on to the future”(p.
Claims of emergence, framed in such assumptions and focused on univer-
sal and abstract modeling, are - seemingly unnoticed - reductionist accounts
(regardless of regularly affirming the opposite) whenever order is taken as
reducible to a single mode of relatedness, but apparently believing, at the
same time, in something ‘more than the sum of the parts’; this is indeed a
mystic path. Already for Pepper was clear that we can not have both. If
there is a function (non-linear usually; it comes to mind the condition of
non-functionalizability of Kim), if there is a law (restricted in any form), or
if there are causal statements, then to deal with absolute novelty beyond
‘the parts’ is not uncomplicated - see e.g. [37], [38]. The old warning of
Gotshalk made in 1932 helps to rephrase the point and the relevance:3g

Uniformity, if anything significant, is a relation between a genuine

diversity. But the causal theory, as set forth, actually applies to
no genuine diversity. We raise no question of the theory’s abstract
truth; the theory, as stated, is abstractly true enough; the same
cause, under similar conditions, no doubt should produce the same
effect, again and again It merely states (in sum) that A is always A,
and fails, so far as experience goes, to state a relationship between
real diversity. In short, it amounts to a true but trifling proposition
(pp. 142-143).

In such scenario, prediction is very narrow: on the basis of empirical

data”we may have a well-confirmed ‘emergence law’ to the effect that when-
ever a system instantiates basal condition M it instantiates an emergent,
E” (p. 8);5 to label it ‘prediction’ is almost a compliment; and this type

JThis is perfectly illustrated in the quote of the physicist Wigner: ”It is ...essential that,
given the essential initial conditions, the result will be the same no matter where and
when we realize these ...If it were not so ...it might have been impossible for us to discover
laws of nature” - as cited by Schlesinger (p. 529).36
90

of inductive predictions is not the interesting matter around the emergent

question and unpredictability. What we would like to understand is why A
is not always A , and not why A is always A.
In short, immutable substances related to a sort of causal forces or law-
like statements for complex systems is a contradiction in terms; it is the
search for predictive theories of ‘novelty’.

4. Addressing Novelty
Complexity Science is supposed to address change, at least if the habitual
allusion to ‘emergence’ is going to be conceived as a kind of change. There
are already accounts of emergence understood as processes of interacting,
immutable (unchanging) entities (or as transitions between different enti-
ties) where results can be repeated as long as the circumstances are similar.
But a static concept of entities is at the center of the puzzle, that is, the ne-
glect of transition within entities. Time is left out. Why? Leclerc expresses
it ~ n m i s t a k a b l y : ~ ~

A stuff endures by virtue of its inertness; its endurance is conceived

to be the retention of its self-identity... [which]is irrelevant to any
temporal passage; the stuff is what it is quite irrespective of any
transition ...it means that stuff cannot endure... it is eternal ... For
if all substance be eternal, what then is duration? What is time?
An ontology of eternity can in the end only explain duration as an
appearance and thus as illusory (pp. 227-228, 229).

How to include time? One natural option is to include duration in

the very character of substances, i.e.”it must be the nature of substance
to be in passage, in transition”(p. 230).33 A natural option, then, is to
conceive them as processes of activities, e.g. a complex system might be
conceived as a process of evolving agents; these agents are thus activities;
moreover, they are agents because they act, which in turn can be conceived
as the use of rules in interaction processes. The significant term is ‘pro-
cess’ which takes the center of the stage. The basic shift is from static to
changing entities. Our current foundations do not ease such a view given
that we are embedded in Western metaphysics which since Aristotle favors
things and static s ~ b s t a n c e sYet,
. ~ ~ process can become a fundamental no-
tion with which events and experience might be (re)-interpreted;*l Leclerc
remarks the shift: ”For Aristotle substance is ‘informed’matter; for White-
head substance or actual entity is ‘informed’activity” (p. 242).33 Therefore
 91

emergents would not be merely things/properties/etc. that surface at some

point of time within a particular ‘process’. Emergence would be fundamen-
tally a process in the lines of thought outlined by thinkers like Bergson or
Whitehead (and surely back to Heraclitus of Ephesus). An agent i s a pro-
cess of activities as well. And the process of multiple inter-acting agents,
i.e. rule users, that unfolds through time is what we distinguish as emer-
gence. No interactions, no emergence. No process, no emergence. After all
emergence is a kind of change.
For instance, the earlier dilemma seems less complicated now. Emer-
gence arises from unique interactions unfolding through time, in this sense
is constituted by and generated from interacting dynamic agents. And this
process only arises as long as we have time and interactions, in this sense is
autonomous at the system level (by principle a process is not separable as
such, it is a whole). hrthermore, regarding hierarchies, the spatial multi-
layered scale is confronted with a more important temporal scale where
”higher levels reproduce themselves over longer durations than lower lev-
e l s ” ( ~9);42
. this statement suggests a different way to explore the concept
of ‘different layers’ of existence. In short, the system-level process of in-
teractions unfolding through time as a unit cell-complex is what we would
call emergence; the system i s that process whose behaviour we assess as
emergent from local interactions. Emergence then is just the process at the
system level, nothing more. To name the system is to name ‘what emerges’
so to speak. No need of something ‘ e ~ t r a ’ . ~
The main goal at this point is to consider this exploration for CS. To
develop the connection between emergence and process thought is not the
purpose of this article.’ But for now suggestive ideas would be: To act i s
to exist. Being is becoming. To understand the becoming of a dynamic

k T ~ proposals
o of dynamic interactions that would constitute non-magical explanans
for ‘emergent’ phenomena are presented in [43]and [44].
‘A broad view on process thought is introduced in [40]. The philosophy of organ-
ism, as proposed by Whitehead, particularly in his major work of 1929 ‘Process and
can be a starting point for approaching process thought more formally. Yet,
beyond this point, a process philosophy may have diverse interpretations and this ar-
ticle is not necessarily committed with particular assumptions; we can expand it in
different ways, e.g. presentationalist/idealist paths closer t o the ideas of Whitehead or
re-presentationalist/realist paths closer to the Evolutionary Epistemology of Campbell
and Popper or the Evolutionary Realism of Dopfer. Also limitations and specific criti-
cisms can be found, e.g. shortcomings related to the nature of composite entities in this
framework have been debated, e.g. [33],[45],[46],[47]; in particular Douglas examines
and rejects the possibility of characterizing emergence within process thought.48
92

process is to explain it as such. Substances are processes of activities.

Agency is constant transition and it is by nature social. Actual occasions
are unique entities. There are no essential causes, entities exist in change
and are social by nature. Durations, activities, and phases take place iden-
tifying an operating unit. Novel togetherness is a principle, not a derivative.
But there is no morphological analysis by components of static ‘stuff’ al-
though processes indeed can be characterized. The shift would underline
that emergence, altogether with contingency, novelty, and creativity, be-
come fundamental categories of metaphysical ~nderstanding.~’ This does
not necessarily resolve the issue; but the debate would be seemingly more
intelligibly. Intuitively a kind of change could be better addressed within
an ontology of change; after all process philosophy presents an ontology
committed to novelty.

5. And Individuality
Another central point from the discussion on emergence is individuality.
How to approach it? This question is related to the methods and types of
explanation that CS aims to provide. The reflection above brings elements
to bear in mind. The general concern is still model building which in fact
is a major feature of CS: the use of computer simulation models. This
particular characteristic is also a current challenge in the realm of explana-
tion, as Berger notices: ”It is reasonable to say that modelling explanations
dramatically increase our understanding of the world. But the modelling
explanations found in contemporary scientific research show that the in-
teresting claims of causal accounts are untenable...An adequate account of
scientific explanation must accommodate modelling explanations, because
they are simply too central to ignore”(pp. 329, 330).26
Keeping in mind the previous discussion, simulation models can be char-
acterized as systems of interacting agents using evolving rules. Addressing
the evolution of dynamic sets of rules (structure, origination, adoption fre-
quencies, selection criteria, individual and collective retention, etc.) we
can explain the becoming of emergents which are constituted by interacting
evolving agents. This is particularly relevant for modeling social systems;
for instance, the ‘parts’ can be related to agents using rules, e.g. cogni-
tive, behavioural, and blueprint rules, see [49]. Each agent is unique and
is changing constantly; they may belong to a generic class but each agent
does not necessarily apply in the same way his particular set of rules, and
a given agent does not act necessarily in the same way today than yester-
 93

day, even in similar circumstances; moreover, the set of rules of each agent
evolves as well. And in a further level, the agents also evolve as such, that
is, the generic class of agents evolves too; perhaps over a longer time scope.
In one word, there is variety: ”many individuals of a kind, where each
constitutes an actualization of a distinct idea. There are, on the one hand,
many individuals that are many matter-energy actualizations, and, on the
other hand, many ideas that can be actualized”(p. 14).35Furthermore, this
variety changes through time - variation (e.g. mutations in biology).
Explicitly, the implication is to question abstract modeling (e.g. ”Kauff-
man’s style”). A foundation on processes of activities underlines the pri-
macy of acting individuals (agents), variety and variation. The inclina-
tion here is for a detailed, qualitative, rule-based, ’expert system’ approach
where details do matter.m

6. Looking Ahead
Bearing in mind the emergent question, the last point to comment is that
an approach based exclusively on computation has to specify clearly how it
is going to meet ontological emergence when such claims are asserted. It is
not straightforward to declare non-reductionism using computer simulation.
Being closed software systems, it is not evident how something arising from
a computation can be (or even just represent) more than essentially that
computation (which is a formal deduction, or at least a positive calculation
anyway) beyond suggestive metaphors or the association of reductionism
with linearity. If we are going to address emergence then the type of episte-
mological bridge between computer simulation and metaphysical emergent
entities has to be built - and carefully - if we are really going to meet
it. Can it be done? After all ontological emergence used to be the original
question. Already W i m ~ a t noticed:
t~~ ”Most scientists regard a system
property as emergent relative to properties of its parts if it depends upon
their mode of organization - a view consistent with reduction”(p. 269).”
He refers to areas such as non-linear dynamics, connectionist modelling,

“‘Indeed, Chris Langton, leading researcher on artificial life, has already stated that ”the
quest for global rather than local control mechanisms (rules) is the source of the failure
of the entire program of modeling complex behaviors up to the present, especially, much
of the work on artificial intelligence”(p. 24).50 - see also [51].
“It is also a matter of what is understood by ‘reduction’; Wimsatt clarifies: ”Most
scientists see reductionism as a species of explanatory mechanistic materialism which
may not even involve laws or theories, as those terms are traditionally understood by
philosophers”(p. 292).
94

chaos, and artificial life. In principle, epistemological - or ‘weak’ - emer-

gence (close to equating ‘emergent’with ‘computed’) does not exclude with
what might be labelled ontological reduction i.e. all higher level entities
and processes are nothing more than certain combinations of lower base
level entities and processes,52which habitually is what seems to be rejected
in the discourse of CS. But there is nothing wrong with recognizing reduc-
ti0nism.O What is wrong is whenever scientists claim to have non-reducible
accounts of (ontological) emergence on grounds that do not necessarily al-
low such attempts. Reduction concerns explanation; and it seems that
when the term ’emergence’ is used in CS, many times can be successfully
understood only as a type of reductive explanation. However, maybe there
are real emergent (novel and unpredictable) entities, and maybe they can
be suitable of being consciously targeted (or rejected) by CS. An ontolog-
ical shift to process could be fully explored; even more, perhaps then, the
question on emergence might be dissolved.
A further step takes place in the roles and functions ascribed to simula-
tion models; detailed modeling does not mean to dismiss the system-level
focus of explanations. Quite the contrary. For example, we can follow a
path closer to the semantic view where a theory is the whole class of its
models (all possible interpretations), such a family of models would be a
highly abstract conception, ”carrying no information other than the struc-
ture of its parent formal system”(p. 301).58The ‘synoptic vision’, closer to
the metaphorical view of explanation, has been also suggested5’. Or we can
develop for instance the proposal of Hayek“ with his ‘explanation of the
principle’ where we focus on types of patterns to expect; instead of ‘predic-
tion’ we can refer to ‘orientation’ and we can prefer the term ‘cultivation’
over ‘control’.
To sum up, process thought arises as a natural option for addressing
unpredictable novelty, the traditional concern behind the term ‘emergence’.
It does not mean the usual notion of process in uniform nature where is a
channel to represent the interaction of unchanging parts in order to label

OIn his stimulating paper, Bedaulg exposes the promising possibilities of working with
weak emergence, though he also rejects the meaningfulness of addressing absolute emer-
gence - ”logically possible but uncomfortable like magic; and irrelevant”. In related
lines, B e ~ h t e lhas
~ ~to ‘defend’ and to make a case for the compatibility of complex
systems and reduction; he is right and a defence should not be necessary; unfortunately
it seems it is needed; moreover, the role of mechanism, as a kind of explanation, has
been recently developed in this context - see [54], [55], [56];in particular for complex
systems see [57].
 95

the change of macro-states at the system level as ‘dynamic’; that is, it

is not merely the variation between two or more consecutive states; this
is misleading since processes of activities are left out, that is, agency (as
acting) is left out. W h a t is underlined here is a foundational shift, with
a first indication t h a t favors detailed modeling. This may shed a light
on understanding novelty; and it may provide an intelligible discourse on
emergence. After all, since 1926, the challenge of Pepper still deserves
a satisfactory answer within CS. But all of these concerns are related to
the questions of change and novelty coherent with a creative and diverse
(complex) world; right there is where the discussion should take place.

Acknowledgments
T h e author thanks three anonymous reviewers for their helpful criticism
and suggestions. Further thanks go t o Peter Douglas, Kristjan Ambroz
and Markus Schwaninger, for their valuable comments on earlier drafts of
this paper. T h e usual disclaimer applies.

References
1. J. Holland, Emergence. Cambridge MA., Perseus Books (1998).
2. S. Jones, Organizing Relations and Emergence. In: R. Standish, M. Bedau
and H. Abbass (Eds.), Artificial Life VIII: The 8th International Conference
on Artificial Life, Cambridge, MA., MIT Press, 418-422 (2002).
3. A. Pap, The Concept of Absolut Emergence. The British Journal for the
Philosophy of Science, 2 (8), 302-311 (1952).
4. S. C. Pepper, Emergence. The Journal of Philosophy, 23 (9), 241-245 (1926).
5. J. Kim, Making Sense of Emergence. Philosophical Studies, 95, 3-36 (1999).
6. A. Stephan, Emergence - A Systematic View on its Historical Facets. In: A.
Beckermann, H. Flohr and J. Kim (Eds.), Emergence or Reduction? Essays
on the Prospects of Nonreductiue Physicalism, Berlin: Walter de Gruyter,
25-48 (1992).
7. P. E. Meehl and W. Sellars, The Concept of Emergence. In: H. Feigl and
M. Scriven (Eds.), Minnesota Studies i n the Philosophy of Science, Volume
I: The Foundations of Science and the Concepts of Psychology and Psycho-
analysis, University of Minnesota Press, 239-252 (1956).
8. G. Watts Cunningham. Emergence and Intelligibility. International Journal
of Ethics, 39 (2), 148-166 (1929).
9. R. Ablowitz, The Theory of Emergence. Philosophy of Science, 6 ( l ) , 1-16
(1939).
10. W. M. Malisoff, Emergence without Mistery. Philosophy of Science, 6 ( l ) ,
17-24 (1939).
11. W. T. Stace, Novelty, Indeterminism, and Emergence. The Philosophical Re-
view, 48 (3), 296-310 (1939).
96

12. D. W. Gotshalk, Causality and Emergence. The Philosophical Review, 51

(4), 397-405 (1942).
13. A. Campbell Garnett, Scientific Method and the Concept of Emergence. The
Journal of Philosophy, 39 (18), 477-486 (1942).
14. P. Henle, The Status of Emergence. The Journal of Philosophy, 39 (18),
486-493 (1942).
15. A. J. Bahm, Organic Unity and Emergence. The Journal of Philosophy, 44
(9), 241-244 (1947).
16. C. W. Berenda, On Emergence and Prediction. The Journal of Philosophy,
50 (9), 269-274 (1953).
17. J. Kekes, Physicalism, the Identity Theory, and the Doctrine of Emergence.
Philosophy of Science, 33 (4), 360-375 (1966).
18. R. L. Klee, Micro-Determinism and Concepts of Emergence. Philosophy of
Science, 51 ( l ) ,44-63 (1984).
19. M. A. Bedau, Weak Emergence. N o i s , Supplement: Philosophical Perspec-
tives, 11, Mind, Causation, and World. 31,375-399 (1997).
20. M. Silberstein and J. McGeever, The Search for Ontological Emergence. The
Philosophical Quarterly, 49 (195), 182-200 (1999).
21. R. Van Gulick, Reduction, Emergence and Other Recent Options on the
Mind/Body Problem: A Philosophic Overview. Journal of Consciousness
Studies, 8 (9-lo), 1-34 (2001).
22. I. Georgiou, The Idea of Emergent Property. Journal of the Operational Re-
search Society, 54, 239-247 (2003).
23. B. Cunningham, The Reemergence of ‘Emergence’. Philosophy of Science.
Supplement. Proceedings of the 2000 Biennial Meetings of the Philosophy of
Science Association. Part I Contributed Papers, 68 (3), 62-75 (2001).
24. E. Mizraji, The Emergence of Dynamical Complexity: An Exploration Using
Elementary Cellular Automata. Complexity, 9 (6), 33-42 (2004).
25. P. Corning, The Re-emergence of ‘Emergence’: A Venerable Concept in
Search of a Theory. Complexity, 7 (6), 18-30 (2002).
26. R. Berger, Understanding Science: Why Causes are not Enough. Philosophy
of Science, 65 (2), 306-332 (1998).
27. D. V. Newman, Emergence and Strange Attractors. Philosophy of Science,
63 (2), 245-261 (1996).
28. A. Clark, Twisted Tales: Causal Complexity and Cognitive Scientific Expla-
nation. Minds and Machines, 8 ( l ) ,79-99 (1998).
29. P. Thagard, Explaining Disease: Correlations, Causes, and Mechanisms.
Minds and Machines, 8 ( l ) ,61-78 (1998).
30. P. Humphreys, Emergence, Not Supervenience. Philosophy of Science. Suple-
ment. Proceedings of the 1996 Biennial Meetings of the Philosophy of Science
Association. Part 11 Symposia Papers, 64,337-345 (1997).
31. R. Harre, Laws of Nature. In: W. Newton-Smith (Ed.), A Companion to
the Philosophy of Science, Malden/Oxford, Blackwell Publishers, 213-223
(2000).
32. J. Bigelow, B. Ellis and C. Lierse, The World as One of a Kind: Natural
Necessity and Laws of Nature. The British Journal for the Philosophy of
 97

Science, 43 (3), 371-388 (1992).

33. I. Leclerc, Whitehead’s Tkansformation of the Concept of Substance. The
Philosophical Quarterly, 3 (12), 225-243 (1953).
34. J. Barnes, The Cambridge Companion to Aristotle. Cambridge, U.K., Cam-
bridge University Press (1995).
35. K. Dopfer, Evolutionary Economics: A Theoretical Framework. In: K.
Dopfer (Ed.), The Evolutionary Foundations of Economics, Cambridge, U.K.,
Cambridge University Press, 3-55 (2005).
36. G. N. Schlesinger , Qualitative Identity and Uniformity. NoZis, 24 (4), 529-
541 (2000).
37. M. Code, Order and Organism. Albany, State University of New York Press
(1985).
38. W. C. Wimsatt, Emergence as Non-Aggregativity and the Biases of Reduc-
tionisms. Foundations of Science, 5,269-297 (2000).
39. D. W. Gotshalk, Uniformity and Induction. The Journal of Philosophy, 29
(6), 141-152 (1932).
40. N. Rescher, Process Philosophy. In: E. N. Zalta (Ed.), The Stanford Encyclo-
pedia of Philosophy, Summer 2002 Edition, The Metaphysics Research Lab,
Stanford University, (2002).
41. A.N. Whitehead, Process and Reality. Corrected edition by D. R. Griffin and
D. W. Sherburne.Cambridge MA., New York, The Free Press (1978).
42. A. Gare, Process Philosophy and the Emergent Theory of Mind: Whitehead,
Lloyd Morgan and Schelling. Concrescence, 3, 1-12 (2002).
43. J. Bickle, Understanding Neural Complexity: A Role for Reduction. Minds
and Machines, 11, 467-481 (2001).
44. F. M. Kronz and J. T. Tiehen, Emergence and Quantum Mechanics. Philos-
ophy of Science, 69, 324-347 (2002).
45. G. Gentry, Prehension as Explanatory Principle. The Journal of Philosophy,
35 (19), 517-522 (1938).
46. W. H. Leue, Process and Essence. Philosophy and Phenomenological Re-
search, 21 ( l ) ,62-79 (1960).
47. M. Code, On Whitehead’s Almost Comprehensive Naturalism. Process Stud-
ies, 31 ( l ) , 3-31 (2002).
48. P. Douglas, Whitehead and the Problem of Compound Individuals. Process
Studies, 33 ( l ) , 8&109 (2004).
49. K. Dopfer, The Economic Agent as Rule Maker and Rule User: Homo Sapiens
Oeconomicus. Journal of Evolutionary Economics, 14, 177-195 (2004).
50. P. Thompson, Biology. In: W. Newton-Smith (Ed.), A Companion to the
Philosophy of Science, Malden/Oxford, Blackwell Publishers, 16-25 (2000).
51. R. M. Burian, Comments on Complexity and Experimentation in Biology.
Philosophy of Science. Supplement. Proceedings of the 1996 Biennial Meet-
ings of the Philosophy of Science Association. Part II. Symposia Papers, 64,
279-291 (1997).
52. R. N. Brandon, Grene on Mechanism and Reductionism: More Than Just
a Side Issue. PSA: Proceedings of the Biennial Meeting of the Philosophy of
Science Association, Vol. 1984, Vol. 11: Symposia and Invited Papers, 345-
98

353 (1984).
53. W. Bechtel, The Compatibility of Complex Systems and Reduction: A Case
Analysis of Memory Research. Minds and Machines, 11, 483-502 (2001).
54. P. Machamer, L. Darden and C. F. Craver, Thinking About Mechanisms.
Philosophy of Science, 67 (l),1-25 (2000).
55. S. S. Glennan, Rethinking Mechanistic Explanation. Philosophy of Science,
69, 342-353 (2002).
56. J. G . Tabery, Synthesizing Activities and Interactions in the Concept of a
Mechanism. Philosophy of Science, 71, 1-15 (2004).
57. W. Bechtel and R. C. Richardson, Discovering Complexity: Decomposition
and Localization as Strategies in Scientific Research. Princeton, N. J., Prince-
ton University Press (1993).
58. M. Hesse, Models and Analogies. In: W. Newton-Smith (Ed.), A Companion
to the Philosophy of Science, Malden/Oxford, Blackwell Publishers, 299-307
(2000).
59. A. L. Plamondon, Whitehead’s Organic Philosophy of Science. Albany, State
University of New York Press (1979).
60. F. A. Hayek, Degrees of Explanation. The British Journal of the Philosophy
of Science, 6 (23), 20S225 (1955).
 WHY DIACHRONICALLY EMERGENT PROPERTIES
MUST ALSO BE SALIENT

CYFULLE IMBERT
IHPST / Universit Paris 1 Panthon - Sorbonne

In this paper, I criticize Bedau’s definition of ‘diachronically emergent properties’

(DEPs), which says that a property is a DEP if it can only be predicted by a
simulation (simulation requirement) and is nominally emergent. I argue at length
that this definition is not complete because it fails t o eliminate trivial cases. I
discuss the features that an additional criterion should meet in order t o complete
the definition and I develop a notion, salience, which together with the simulation
requirement can be used t o characterize DEPs. In the second part of the paper, I
sketch this notion. Basically, a property is salient when one can find an indicator,
namely a descriptive function (DF), that is such that its fitting description shifts
from one elementary mathematical object (EMO) to another when the property
appears. Finally, I discuss restrictions that must be brought t o what can count as
DFs and EMOs if the definition of salience is t o work and be non trivial. I conclude
that salience (or a refined version of it) can complete the definition of DEPs.

In the literature about emergent properties, attention has recently been

paid to properties that emerge diachronically in complex systems, for ex-
ample when a tornado ends up appearing in a usually quiet area.
The central idea behind diachronic emergence is that these emergent
properties can be explained or derived, at least in principle, from the system
microdynamics (so they have nothing mysterious), but only in a complex
way, e.g. for the apparition of the tornado by numerically solving Navier-
Stokes equations.
In the past decade, different definitions have been proposed to charac-
terize more precisely diachronically emergent properties (DEPs). Stephan’
emphasizes that DEPs are highly unpredictable, as demonstrated in the
case of deterministic chaos. Bedau2 claims that a property is diachronically
emergent (he uses the term ‘weakly emergent’) if it can be derived from the
system’s microdynamics and external conditions but only by simulation. I
label this latter condition the ‘simulation requirement’ (SR).
In this paper, I focus on Bedau’s definition, which I am definitely sup-

99
100

portive of, and I try to refine it further by making it more stringent. I

agree that the definition is sufficiently broad to be fruitfully applied to a
large variety of systems that exhibit properties that seem to be DEPs. Yet,
I argue that the SR is not enough to define DEPs and that the definition,
as stated by Bedau, encompasses too many properties. I try in a second
step to build a notion, namely salience, that is appropriate to complete the
definition and capture only the right set of properties.

1. Preliminaries
In order to present the problem, I shall restrict for simplicity to an example
belonging to discrete mathematics, namely the case of a one dimensional CA
system composed of cells, each colored black or white.a A rule determines
the color of each cell at each step. Below is rule 110. It says for example
that when a CA and its neighbours are black, the CA turns white at the
next step.

Figure 1. Rule 110

The evolution of such a CA system, composed of automata following

rule 110, depends on its initial condition. Figures 2 and 3 represent this
evolution when the initial condition is a single black cell. Each line repre-
sents the system state at a given step and the evolution is downwards.
One can see on these figures that in this case, the evolution is neither
highly regular nor completely random. We can witness the emergence of
patterns such as occasional big triangles or as the vertical column of trian-
gles in the right.
For the discussion that follows, it is appropriate that I give pet names to
the properties that the discussion is going to revolve upon. ‘Blackcell’refers
to the property ‘having CA 10 in a black state’ and it is true at time 20.
‘Bigtriangle’ refers to the property ‘having a big triangle appearing around
CA number 20 on the lattice’ and it is true around time 80. ‘Microstate50’

aFor a clear presentation of cellular automata, see for example [3].

 101

Figure 2. Rule 110, starting from a single black cell, 250 steps. Figure taken from A
New Kind of Science, chap. 2, reproduced with permission of Stephen Wolfram, LLC.

refers to the property ‘being in the configuration that the system is in at

step 50’ and it is true at time 50. ‘Verticalcolumn’ refers to the property
‘having a vertical column of triangles appearing around CA 60’ and it is true
around time 350. These are property names but for linguistic commodity,
I shall also use these names to refer to the instantiation of these properties
at the mentioned times and also as if they designated emerging things.
I also need to introduce scraps of terminology. I shall use ‘microprop-
erty’ to refer to a property of a basic constituent. In our examples, the two
basic microproperties of a CA are ‘being black’ or ‘being white’ and you
can specify them by including time, e.g. ‘being black at time 3’. Following
the uses in statistical physics, I shall call ‘microstates’ conjunctive prop-
erties that consist of the ordered enumeration of all the microproperties
of the parts of the system S. I shall also call ‘macroproperties’ proper-
102

Figure 3. Rule 110,500 steps. Figure from A New Kind of Science, chap. 2, reproduced
with permission of Stephen Wolfram, LLC.

ties that can only be possessed by the whole system but not by its part.
Such properties are sometimes referred to in the literature as ‘nominally
emergent’.4 Structural properties, involving microproperties and a relation
between them are a subspecies of nominally emergent proper tie^.^ Exam-
ples of macroproperties are Blaekcell, Microstate50 or, for a cup of water,
fluidity or transparency, because molecules of water cannot be said to be
fluid or transparent. I will not venture here in giving a more detailed general
definition of macroproperties because I shall not need that much to define
DEPs. Bedau himself acknowledges that “full understanding of nominal
emergence would require a general theory of when macro entities have a
new kind of property that their constituents cannot have”.4 That should
not worry us too much because in our CA system example, we can very
easily, if we need to, depart on a purely logical ground what counts as a
macroproperty from what does not.
 103

2. Statement of the problem (Pb)

I can now phrase precisely why I am not satisfied with Bedau’s definition
of DEPs. The consequence of equating the fact of being a DEP with the
fact of satisfying the simulation requirement (SR) is that

(1) Blackcell, Microstate 50 and any coined conjunctive property that

you feel like building up by the conjunction of microproperties, e.g.
‘having CA 24 black at time 3 and CA 47 white at time 567’- I
call these “deceptive properties’;
(2) respectable properties such as Verticalcolumn, Bigtriangle, or any
emerging pattern - I call these “target properties”;

must be treated on a par and considered as DEPs because we need to run

a computer simulation to predict them all.
I think this is an unpalatable consequence of the definition, which shows
that the SR is not enough to define DEPs.

2.1. Two ways to face (Pb)

I shall now examine two ways to answer my claim that the SR is not suf-
ficient to define DEPs and I shall show why I think these answers are
unsatisfactory.

2.1.1. Is nominal emergence enough to solve (Pb)?

I have assumed in my phrasing of the problem that the SR is what matters
in the definition of DEPs. Bedau may perhaps argue that I have not faith-
fully reported his view, because he says: “Assume that P is a nominally
emergent property possessed by some locally reducible system S. Then P is
weakly emergent if and only if P is derivable from all of S’s micro facts but
only by simulation”. So according to Bedau the SR applies only to nomi-
nally emergent properties. Thus I can be reproached for having forgotten
the condition about P being nominally emergent or else for having misused
the notion: Blackcell (‘having CA 20 black’) seems to be a property of the
whole system and so to be nominally emergent ... but isn’t it because I used
a logical trick to forge Blackcell, instead of simply talking of a less intricate
microproperty which applies to CA 20 at time 10, namely ‘being black’?
Well, maybe. I leave to the reader’s metaphysical intuitions whether the
property ‘having a big wart on her nose’ (which is true of the witch that
gives the apple to Snow White) is not a very respectable property too and
104

is really the same as ‘being in a warty state’ (which, for a coarse-grained

description of the witch, is true of the descriptive cell corresponding to the
witch’s nose). In any case, the conclusion should at least be that if we want
to be able to determine which properties are DEPs we really need much
more theorizing about nominal emergence to select the right properties.
But I strongly doubt that a good theory about nominal emergence will
do. A reason is that, from a descriptive point of view, properties such as
Bigtriangle or the appearance of a tornado in a fluid, which are paradigms
for diachronical emergence, describe local features indeed,b and that these
features do not necessarily involve the whole system for being predicated
but can be predicated of just a few CAs, even if they are generated by the
whole system evolution (just like Blackcell). So from the nominal emergence
perspective, the difference between Blackcell and Bigtriangle is very slight,
whereas my guess is that Blackcell is not a DEP in our example, whereas
the appearance of a tornado is.
Another problem is that properties which are pure conjunction of micro-
properties and involve the whole system, such as Microstate50, will still be
nominally emergent, however unremarkable they may be. It is still possible
to hope that a strengthening of nominal emergence may make it possible
to get rid of them too. Yet it seems to me that with this strategy, we are
progressively lead towards a different kind of emergence leaning on a strong
distinction between whole and parts. The solution I propose in the second
part of this paper, with the use of the notion of salience, does not make use
at all of nominal emergence.

2.1.2. Is there more to diachronical emergence than mere

computational irreducibility?
A second way to face (Pb) is to hold tight like this: “Yes indeed, the
SR is what only matters to define DEPs and there is no reason to make
further distinctions between properties, because as far as weak emergence
is concerned, these properties are really on a par. If you want to predict the
apparition of Blackcell or Microstate50, you have to run a simulation and let
all these properties diachronically emerge. These properties are so to speak
“as diachronically emergent as” Bigtriangle or Verticalcolumn because the
same computational effort (modulo time linearity to account for different
appearance times) is required to predict them all. This computational effort

bSee also note g.

 105

corresponds precisely to the running of the dynamics of the whole system

- and this is where the notion of whole system has a part to play.”
I do not think that Bedau would hold tight like this. In [2], the examples
are about the emergence of periodic or stable configurations in the Game
of Life and about adaptation of mutation rates in a model of evolving life.
That is not exactly what one would call trivial or insignificant properties
of a system, as Blackcell or Microstate50 can be assessed to be. So if the
definition purports to include in its scope such insignificant properties, this
should be made clear and the examples are misleading because they point
at a specific class of very remarkable DEPs.
Besides, if there is nothing more to diachronic emergence than the SR,
then it is maybe better not to talk of emergence at all and to speak merely
of properties of computationally irreducible systems6 , or of properties that
are for a system computationally irreducible: if one views the evolution of
a system as a computation, then a system is computationally irreducible if
given its initial state, there exists no shorter computation that enables to
predict a later state of the system. In other words, there exists no shortcut
to predict the properties of the system.
Yet, I am quite reluctant to content with an equivalence between di-
achronic emergence and computational irreducibility. Giving a clear defini-
tion of computational irreducibility and finding that there are systems that
are computationally irreducible, as in the case of low-dimensional chaos7, is
one thing. Then, it could be a brute fact about nature that the properties
that cannot be predicted except by a simulation are not really worthy of
interest. Take for example a classical gas. Perhaps it is true that there is no
way to predict exactly its future microstates (that is to say, to predict the
position and velocity for each of the particles), except by a simulation.
But who really cares? We can still use statistical physics to (try to) cal-
culate more remarkable properties. So computational irreducibility would
not be such a big issue, if it was just met in the case of such anonymous
properties.
But this is not the case. And a good incentive to talk about diachron-
ically emergent properties is the two-piece fact (S) that, as abundantly
illustrated by [6] and as can be seen in figure 3 with Bigtriangle or Verti-
calcolumn:
(Sl) there seems to be computationally irreducible systems and
(S2) these systems generate among other things some properties (e.g.
periodicity when Verticalcolumn appears) that can be as respectable, re-
markable, new, worthy of interest or what you like as the properties we
106

ordinarily predict using analytical methods.

My conclusion is that there is really something wanting in a definition
of diachronic emergence that only relies on the SR.
Various requirements can be found in the literature’, for emergent prop-
erties, in particular novelty and irreducibility.c The SR requirement seems
to fulfill the irreducibility requirement. So my claim is that, in the case of
diachronic emergence, something like the novelty requirement must also be
honored one way or another.

2.2. How to complete the simulation requirement?

It did not go unnoticed that I left (S2) quite vague and used so far unsat-
isfactory adjectives (‘interesting’, ‘remarkable’, etc.). The purpose was to
keep my different claims distinct so that one can agree with my diagnosis
(the SR is not enough) and possibly disagree with the way I try to solve
the problem (what is missing).
What I want to do now is to give a criterion that enables to depart the
computationally irreducible properties that seem to have nothing special
(deceptive properties) from the ones that seem to have something more
(Verticalcolumn, Bigtriangle). But before presenting my solution, I think
it is appropriate to list the constraints that weigh on this criterion.

2.2.1. T h e criterion m u s t n o t rely o n intrinsic features of the

property but m u s t be contextual (Cl)
As I stated above, properties that are produced by a computationally irre-
ducible system can be the same as the ones we predict in other analytically
solvable cases, e.g. periodicity of a configuration. Therefore one should
not be after a criterion that would be based on a logical analysis of each
property, or on a complexity measure or any other feature characterizing
intrinsically the properties. To put it differently, what is striking is not to
witness a complex, or ordered or random behavior in a CA system but,
for example, to witness the apparition of a complex or random behavior
emerging out of an initial simple output or, like in the example of CA 110,
of structures emerging out of an almost uniform initial condition.d In other

‘This is in no way is in a definition of or a necessary condition for emergent properties.

In particularly, I do not assume that there is a single concept of emergence or that any
concept of emergence must meet these requirements.
dSee [6, pp.149-1521 for examples.
 107

words, what matters is which new features the system dynamics is able to
generate out of a context that did not possess these features, and that is
why I say the criterion we look for is contextual.

2.2.2. The criterion must be local, that is to say must rely on

preceding microstates only (C2)
Most of the listed properties (deceptive and target properties) are new in
a very weak sense: they are properties that were not possessed before (the
system was never in a state identical to Microstate50 before). Verticalcol-
umn and Bigtriangle seem to be new in a stronger sense because the system
seems to locally undergo a change. In the same time, nothing precludes that
similar instances of target properties may have appeared before or may ap-
pear again later: tornadoes do re-emerge in the same areas. I conclude that
the desired criterion, if it is to characterize novelty, must be local in the
sense that it is possible to determine if a property is novel only by looking
at the microstates of the system just before the property appears.
Local novelty is also an attractive feature because it is a weak require-
ment in contrast with other kinds of novelty such as absolute novelty (if
the supposedly new features were never possessed before). And I am after
a weak criterion because I believe that the SR is what matters first in di-
achronical emergence and I want only to refine it by disqualifying a very
specific set of deceptive properties.
The requirement of local novelty is also there to do credit to the idea
that the apparition of an emergent property corresponds to an observable
change in the system state. Emergent properties may involve no new causal
power, as this is precisely the case with diachronic emergence, where all
causal effects can be traced back to the system’s microdynamics. But one
can still require that the emergence of these properties correspond to the
apparition of novel features characterizing the system.

2.2.3. The criterion must be contextually absolute (C3)

Once a context - that is to say the previous microstates - is given, the
novelty of a property must be assessed on the basis of objective properties.
For example, there seems to be an objective sense in which one can say
that a tornado is a new structure that locally appears, because one can
find objective indicators to signal this appearance, such as a sudden local
increase of average vorticity. In particular, although I call ‘salience’ the
property that emergent properties must possess, I do not want salience to
108

be relative to an observer and her cognitive apparatus.

3. Salience
In the last section of this paper, I try to develop a concept, salience, that
fulfills the previously listed requirements and I claim that DEPs must also
be salient. Salience is an independent notion and I believe that it can prove
to have a wider application than the question of emergence.e The definition
of salience also raises its own set of problems, but since it is not my goal
to fully develop this notion here, I shall be quite sketchy, leave aside most
of these problems and be content to show how salience could be used to
complete the definition of DEPs.
The idea I am going to develop is that a property is salient if one can
find a descriptive indicator which can be calculated for any state of the
system and that undergoes a change when a salient property appears. To
make things clearer, I start with two clear-cut examples.
In the case of phase transitions, order parameters (when one manages
to find one) yield such indicators. For example, in the case of the transition
from ice to water, the quantity pice - psystem,where p indicates the density,
is zero till the transition and then grows.
Another more elaborate example from dynamical system field, which
I borrow from Rueger', is the case of the damped non linear oscillator,
+
which is described by van der Pol's equation: x - a(1- x2)x x = 0, where
x = d x / d t and a is a damping parameter. For a dynamical system, it is
appropriate to study properties of the phase space, which plays for us the
role of a descriptive indicator. With a = 0, the oscillator is undamped
and the phase space can be portrayed by concentric ellipses, each ellipse
representing a trajectory (see figure 4). If the damping is gradually turned
on, trajectories are no longer periodic and the phase space is made of spi-
rales converging to a limit circle (see figure 5). So the turning off of the
damping makes new salient properties appear (e.g. periodicity) and this
can be indicated by the topological change of the phase space, which is de-
scribed by topologically inequivalent objects (two objects are topologically
inequivalent if one cannot smoothly deform one object to transform it into
the other).
This specific and very well chosen example enables Rueger to provide a
very clear criterion of novelty, using the notion of topological inequivalence.

eFor example for the question of what it is to identify a non trivial property in a data
basis obtained by a number crunching simulation or by experiments.
 109

Figure 4. Undamped oscillator, taken from Synthese, 124, 2004

Figure 5. Damped oscillator, taken from Synthese, 124, 2004

Yet it seems to me that he actually shows how novelty can be defined in a

very particular case (dynamical system studies) and not in general. I am
not sure my criterion of salience will be as clear and elegant, but it aims at
being more general.

3.1. Preliminary definitions

I assume that at each time, the system can be suitably represented for the
properties that are studied by data originating in n measurements and I call
‘state of the system’ this n-tuple of measurements. Examples of states are
a triplet (P,V,T) for a gas at equilibrium, data describing a velocity field
110

for a fluid, what I called ‘microstates’ for the above CA system or subparts
of them (for example the subpart corresponding to CA number 15 to 25,
that is to say where Bigtriangle does appear).
I call ‘trajectory’ the sequence of states that one gets by varying a
parameter describing the system. This parameter can be time, like in the
case of the above CA system, but it need not always be. For example,
temperature can be chosen for an ice to water phase transition. What is
only required to get a well-defined trajectory along a parameter is that
one single state can be ascribed to each new value of the parameter along
the trajectory. Suppose for example you study properties of cooling of
glasses. Since, for a similar initial condition, the end state of the glass and
its properties depend on the cooling rate (see figure 6)’actual trajectories
depend on the cooling rate. Therefore temperature cannot be taken as the
varying parameter defining trajectories. Specifying exactly the trajectory
does matter because salience is a contextual notion (Cl) and because the
part of the trajectory before a property appear provides the context against
which the property stands out.
I call ‘descriptive function’ a single-valued mathematical function that

0.12 -
0.10 --

0.08 --
0.06 -.

0.04 --
0.4’Wsec

100 200 300 400 500 600

Figure 6. Glass cooling modelled by a two well system (the fraction of the systems in the
higher well is plotted). Figure taken from Dynamics of Complez systems,by Bar-Yam,
reproduced with permission of Addison-Wesley.
 111

takes as an argument the n-tuple describing the system.f

This is in a no way a definition of a new object. This definition only
provides a label that is stamped on already existing functions. Here are
examples of descriptive function:
for a spin lattice, the total magnetization or typical size of clusters
of similarly oriented spins;
0 an estimation, using an auto-correlation function, of the correlation
between measurements at spatial or temporal distance r or t ;
0 the probability value of an event, for example of having two similar
colors following each other on the above CA system;
0 a complexity measure such as Kolmogorov complexity of a state;
0 etc.
Descriptive functions are aimed at giving a common indicator all along
the trajectory, so that it be possible to make comparisons and thus deter-
mine if the system exhibits a new behavior. In the case of Verticalcolumn,
we can for example study the spatial correlations around CA number 60
in the 100 steps before Verticalcolumn is instantiated. On this interval,
which plays the role of a context, the system undergoes fluctuations and
correlations are low. When Verticalcolumn appears, fluctuations stop and
the correlations suddenly get high.
To be more precise descriptive functions should themselves be described
by other more elementary mathematical object (EMO) which enable to fit
them locally. In the case of the oscillator, the phase space was described
by different geometrical figures (circles and spirales). The novelty is then
signaled by the shift in the description of the descriptive function from one
E M 0 to another one. In the case of single-valued functions, EMOs are ele-
mentary functions such as z2 and power laws, l o g ( z ) ,e z p ( z ) ,trigonometric
functions, etc.

3.2. Towards a definition of salience

I can try now to propose a first definition of salience.

A property P of a system S, which appears at t or on interval I is

fThis is actually a restriction that I make for simplicity because one may need more com-
plicate objects than singlevalued function to describe suitably the systems. In Rueger’s
example of the oscillator, the system was described for each value of the damping pa-
rameter by trajectories within a phase space,
112

salient for trajectory T of S iff there exists different EMOs A and B and a
descriptive function DF of the states of S’s trajectory such that:

0 DF is described by A on an interval C of T (the context) before t

or I;
0 the description of DF shifts from A to B at t or on I with the
apparition of property P.

This definition, because of the existence condition, does not provide an

algorithm to determine if a property is salient. So the burden of the proof
is for those who want to claim that a given property is salient, because they
need to find an appropriate indicator, which can be far from obvious, even
for phase transitions. Further, this definition, stated like this, can easily
be found wanting. For example, it would be necessary to have a way to
determine what the minimal size of the context needs to be in each case.
In the following paragraphs, I shall study a few obvious corrections that
should be brought to the definition in order to make it work better and also
indicate in which directions some work needs to be done if a definition of
this kind is to be given a chance to be well-grounded in the end.

3.3. How to avoid ad hoc descriptive functions?

Remember salience aims at eliminating undesirable properties such as Mi-
crostate50 or Blackcell. An immediate objection is that it is very easy
to find descriptive functions that will exhibit a change in behavior when
Microstate50 or Blackcell appear. ‘Blackcell’ refers to the property of hav-
ing CA 10 in a black state at time 20. If I take the color of CA 10 as
a 1-tuple state, and if ‘black’ is encoded by 1 and ‘white’ by 0, and if
CA(t,x) denotes the state of CA number x at time t , then the function
t Hb(t - 20).6(1- C A ( t ,lo)), where 6 denotes Dirac function, does ex-
hibit a a sudden change in behavior when Blackcell appears. A similar trick
can give the same result for Microstate50. That is why it is necessary to
require that descriptive functions do not include functions of the parameter
defining the trajectory (such as t H6(t - 20)) that have an ad hoc salient
feature, which is made to appear at t or I (see definition above). Further,
similar variables of the n-tuple should be treated similarly or contribute
formally in a similar form to the descriptive function in order to avoid ad
hoc treatments. The consequence is that most descriptive functions, like
in the above examples, will be sums or integral functions.
A cautionary remark needs to be made here. Such restrictions are not
 113

aimed at getting rid of properties like Blackcell or Microstate50 in any

circumstances. Suppose that CA 10 -happens to be white from step 0 to 19
and then turns black at step 20. Then it is legitimate to say that Blackcell
is salient and for that we no longer need an ad hoc function as above.
Function CA (t,10), which is tolerated by the definition, will nicely do the
job because it will have a zero value till step 19 and then will abruptly take
value 1 at step 20.g Similarly, if it happens that Microstate50 is random
and appears after a succession of non-random states, then this provides a
good ground to call it ‘salient’.

3.4. Requirements about describing EMOs

Notice first that I do not define in the abstract what can count as EMOs.
What EMOs are will depend in each case of what kind of mathematical
object the descriptive function is. For example, EMOs will be different if
you want to study the evolution of a phase space or of an order parameter.
A fully assumed consequence of my definition is that a curve can have
different salient parts even if you can fit it with one single curve. Take for
example a curve that you can fit with function x Hx2 0 . 1 ~ With + ~ . my
definition, it can be said to have different salient parts. Around x = 0, the
curve is correctly described by x2. For large x, it is best described by 0.1x3.
A more worrisome problem is this. Take a curve F that always has 0
value. Then you can very easily find two polynomials A and B such that
1) A has value 0 till x = l and then has a more complicated behavior 2) B
has value 0 after x = 1 and a complicated behavior before.h A and B are
2 EMOs and the description of F shifts from one to the other. So it seems
that it is possible to find salience really everywhere. To guard against that,
it is necessary to require that there is no E M 0 that fits F on both intervals
where A and B are supposed to fit it and that is in the same time simpler
than A and B. Here, the function x H 0 is simpler than A and B.
The outcome of this is that, in order to fully develop the notion of
salience, I need to be able to estimate the simplicity of mathematical ob-
jects. In the previous example, one must be able to explain why x 2 0.1x3 +
gNote that in this case, nominal emergence would disqualify Blackcell.
hIn short, I need to set restrictions on which EMOs can be used to fit the curves. It
probably did not go unnoticed to the reader that the discussion of salience turns out to
be reminiscent of discussions about induction, curve-fitting and projectibility, and here
in particular of Goodman’s paradox about grue emeralds (see [9,chap. 3-41 and [lo,
chap. 51. A complete definition of salience should show how to avoid the goodmanian
pitfalls or how to get out of them.
114

is less simple than x2 or 0.1z3. Since the last two expressions are composed
of one element and the former of these two elements, the answer may seem
obvious. But this seems all too relative to the mathematical framework that
is taken as basic. Taking as basic the functions z H g(z) = z2 0.1z3 +
and z H h ( z )= z2- 0.1z3, +
one gets z2= 0.5(g(z) h ( z ) )and z2seems
now less simple than g(z). As a reviewer points out, this makes salience
relative to a privileged descriptive mathematical framework. A way out
of the deadlock would perhaps be to argue that there are good objective
grounds, in each case of data mining and salient properties detection, to
privilege a descriptive mathematical framework and basic EMOs on the ba-
sis of which projectable quantities characterizing the system the descriptive
function and more generally the mathematical description represent.
I do not go any further in the discussion of the concept of salience, since
as just shown, this would require a close scrutiny of much debated ques-
tions, like simplicity and curve fitting. One more thing still. In this paper,
the concept of salience was only aimed at giving a way to single out the
properties that are considered as remarkable in the study of a system, e.g.
in physics. But determining what patterns or properties are salient for a
subject is also a question that is crucial in cognitive sciences.' I make no
claim in this paper about the link between the two notions, which may be
somewhat related, even if not identical. This latter point can be illustrated
as follows. Indeed, a salient property (as defined above) in a data basis can
be imperceptible for a subject. That is why data mining requires resort to
statistical tests made by computers. In the same time, perceptual abilities
for pattern recognition also prove very useful to detect interesting p r o p
erties, for example in hydrodynamic simulations, and the existence of the
supposedly detected properties can be checked afterwards with statistical
indicators. Finally, a system (e.g. a Hopfield network3) could be trained to
detect in any circumstances a given pattern of CAs whereas this pattern
need not always be salient (in my sense), since the notion is contextual.
Besides, the pattern detection system may treat differently inputs of the
same kind ( i e corresponding to similar variables of the n-tuple in the
mathematical description) and thereby use for detection a function that
would not qualify as a descriptive function.

'See Ill] for example.

 115

4. Conclusion
We can now check that the concept of salience meets the requirements
listed above, or that a well-grounded and refined version of it could be
hoped to. It is local, because it depends only on the preceding part of the
trajectory. It is contextual because any property can happen t o be salient
if it is generated by a context against which some of its features stand out.
It is objective (at least provided a descriptive mathematical framework is
given) because it relies on the construction of mathematical indicators and
not on our epistemic interests, perceptual abilities or practical goals: for
example, the position at a given time of a planet, on which one wants t o
send a shuttle, is very unlikely t o be salient.
Salience, as I built it, is a very weak and purely descriptive notion,
which aims a t grasping the idea that the apparition of a non-trivial prop-
erty is simultaneous with a significant change in some of the values of the
observable quantities characterizing the system. Once again, I think that
using a concept as weak as possible t o complete the definition of DEPs is
appropriate because I believe that the SR is the crucial element in it. But
my conclusion is that DEPs satisfy the SR and are salient.

Acknowledgements
Special thanks to Anouk Barberousse, Jacques Dubucs and Paul
Humphreys for discussing different versions of this paper and also to an
anonymous reviewer for very helpful and very apposite critical comments,
which helped a lot t o improve the paper. All remaining shortcomings are
of course mine.

References
1. A. Stephan, “Phenomenal Emergence”, Networks, vol. 3-4, 91-102 (2004).
2. M. Bedau, “Weak emergence”, Philosophical Perspectives: Mind, Causation
and World, 11, Oxford Blackwell Publisher (1997).
3. G. Weisbuch, Dynamique des systbmes complexes, Une introduction a m
rkseauz d’automates, InterEditions/Editions du CNRS (1989).
4. M. Bedau, “Downward causation and the autonomy of weak emergence”,
Princzpia, vol. 6, 5-50 (2003).
5. T. O’Connor, Hong Yu Wong, “The Metaphysics of Emergence”, N o h , forth-
coming.
6. S. Wolfram, A New Kind of Science, Wolfram Media, Inc. (2002).
7. J.P Crutchfield, J.D Farmer, N.H Packard, R.S Shaw, “Chaos”, Scientific
American, vol. 255, 46-57 (1986).
116

8. A. Rueger, “Physical Emergence, Diachronic and Synchronic”, Synthese, vol.

124, 297-322 (2000).
9. N. Goodman, Fact, Fiction and Forecast, Harvard University Press (1983).
10. N. Goodman, Language of Arts, Hackett Publishing Company, Inc. (1976).
11. D. Marr, Vision, A computational investigation into the human representa-
tion and processing of visual information, New York: Freeman (1982).
 ON THE RELATIVITY OF RECOGNISING THE
PRODUCTS OF EMERGENCE AND THE NATURE OF THE
HIERARCHY OF PHYSICAL MATTER

KURT A. RICHARDSON

1. Introduction
On the matter of privileging a so-called ‘scientific method’ over all other
methods in the discovery of ‘truth ’ the philosopher Paul Feyerabend said:

It is thus possible to create a tradition that is held together by

strict rules, and that is also successful to some extent. But is it it
desirable to support such a tradition to the exclusion of everything
else? Should we transfer to it the sole rights for dealing in knowl-
edge, so that any result that has been obtained by other methods
is at once ruled out of court?’

Like Feyerabend my answer to the question is “ a firm and resounding

NO.” In this chapter I would like to explore my reasons for this answer by
exploring the ontological status of emergent products. I am particularly
interested in demonstrating the relativity of recognising the products of
complexity, i.e., that emergent-product recognition is problematic and not
‘solved’ by the application of one particular method over all others. I would
also like to explore the consequences of this in regard to the hierarchy of
physical matter and the hierarchy of human explanations.

2. A Theory of Everything
My starting point for such an exploration is the assumption that the Uni-
verse, at some level, can be well-described as a cellular automaton (CA).
Of course by selecting this starting point I am making the assertion that
not only does some form of cellular automata represent a plausible theory
of everything, but that a theory of everything does indeed exist. This would
seem to be a radically reductionist starting point, but I am claiming a lot
less than may at first appear. Rather than defend this starting point in

117
118

the main part of the chapter, I have added a short appendix that discusses
the reasonableness of the CA assumption. One thing is clear, a theory of
everything doesn’t even come close to offering explanations for everything so
we needn’t get too concerned about the ‘end of science’ or human thinking
just because such a theory might exist and might be discoverable.
Whether one accepts my starting point or not it should be recognised
that this is a philosophical exercise; to explore the ontological and episte-
mological implications of assuming that the Universe is well-described as a
CA at some level. It doesn’t necessarily follow that the Universe actually
is and only is a CA. The details of the CA, however, do not change the
philosophical conclusions. Richardson15 has explored these issues at length
and argued that, whether or not a CA model is a plausible theory of ev-
erything, a philosophy of complexity based on this assumption does seem
to have the capacity to contain all philosophies, arguing that each of them
is a limited case of a general, yet empty, philosophy (akin to the Buddhist
‘emptiness ’. This seems to be a fantastically audacious claim, particularly
given the heated debate nowadays over global (intellectual) imperialism.
However, one should realise that the resulting general philosophy is quite
empty - which I suppose is what we would expect if looking for a philosophy
that was valid in all contexts - and explicitly avoids pushing to the fore one
perspective over any other. Quoting Feyerabend’ again, “there is only one
principle that can be defended under all circumstances and in all stages
of human development. It is the principle: anything goes.” In a sense, all
philosophies are a special case of nothing!

3. S o , What is Emergence?
In the founding issue of the journal Emergence, Goldsteinlo offers the
following description of ‘emergence ’:

Emergence ... refers to the arising of novel and coherent structures,

patterns, and properties during the process of self-organisation in
complex systems. Emergent phenomena are conceptualised as oc-
curring on the macro level, in contrast to the micro-level compo-
nents and processes out of which they arise.

For the rest of this paper I would like to consider how patterns and
properties emerge in simple CA, and how the macro is distinguished from
the micro. I will argue that the recognition of emergent products, which is
often how the macro is distinguished from the micro, is method dependent
 119

(and therefore relativistic to a degree). The recognition of emergent prod-

ucts requires an abstraction away from what is, and as with all abstractions,
information is lost in the process. Of course, without the abstraction pro-
cess we would have all information available which is no better than having
no information at all. "Only a tiny fraction of [the richness of reality] affects
our minds. This is a blessing, not a drawback. A superconscious organism
would not be superwise, it would be p a r a l y ~ e d ". ~I do not intend to offer
a full review of the term 'emergence' here. The interested reader might
find Goldsteinlo or Corning3i4 useful as additional sources on the evolution
of the term itself.

3.1. Complex cellular automata and patterns

Figure l b shows part of the space-time diagram for the simple CA &36 ( n
= 20, k = 3). Even though the initial conditions selected for the particular
instance of CA were selected randomly, a pattern emerges that is neither
totally ordered (an example of which is given in Figure la) nor quasi-chaotic,
or quasi-random (Figure lc). CA &36 is an example of a (behaviourally)
complex CA as defined by Wuen~che'~.There isn't a lot else we can say
about the pattern (although there are a wealth of statistical analyses we
might perform to quantify certain aspects of the pattern) that emerges
except that maybe it is rather aesthetically pleasing and that its appearance
from such a simple set of rules, as well as a random initial configuration, is
quite a surprise. So even though it is often said that these patterns emerge
from the application of the CA rule, it is not particularly clear as to what
is emerging - what are the emergent products here? Is it appropriate to
suggest that the space-time pattern is emergent, given that it is a perfect
image of the ' micro '-level? What would the macro level be in this example?
It is not immediately clear at all.
There is no autonomy in this example, as all cells are connected to all
others - the pattern is simply the result of applying the CA rule time step
after time step. One thing we can say is that the pattern was not wholly
predictable from the CA rules, and so in that sense the pattern is novel
given our lack of knowledge of it before the system was run forward. But
did it really emerge? Or, are our analytical tools just too feeble to charac-
terise the patterns beforehand, in which case the fact that the patterns are
regarded as emergent simply reflects our lack of knowledge about this simple
system - our theories are simply not powerful enough. Following Bedeau',
Goldstein" refers to this aspect of emergence as ostensiveness, i.e., the
120

Figure 1. Examples of (a) ordered, (b) complex, and (c) chaotic automatons

products of emergence are only recognised by showing themselvesa. This is

also linked to the idea of radical novelty'' , meaning that the occurrence of
these patterns was not predictable beforehandb.
These patterns are of interest in that they illustrate some aspect of the
concept emeTgencec, but they do not tell us much at all about the dynamics
of the space-time plot. If we look very closely however we begin to notice
that there are patterns within the patterns. The triangles have different
sizes and their evolution (shape and movement) 'down' the space-time
plot seems to indicate some kind of interaction between local areas of the
overall pattern. We might also notice that certain areas of the space-time
plot are tiled with a common sub-pattern - a pattern basis. Nansen and
Crutchfieldll demonstrated how an emergent dynamics can be uncovered
by applying a filter to the space-time plot. Figure 2 shows the result of a
' pattern-basis filter' applied to a particular instance of the Space-time plot
of CA rule &36. Once the filter is applied an underlying dynamics is dis-
played for all to see. What can we say about this alternative representation
of the micro-detailed space-time plot?
Firstly, we can develop an alternative macro-physics for this apparent

'Even though for the products t o show themselves, an observer must recognise them as
products.
bThough having run this particular CA on many occasions, the qualitative nature of
the pattern was of no surprise to me at all - otherwise I would've found it very dificult
indeed to preselect an appropriate case. It would be a very straightforward exercise
indeed for me to produce a library of CA patterns from all possible initial conditions -
though I would be limited by such resources as time and, the related resource, processing
power. The point is that I would have to run all the cases fully, there is no shortcut -
no algorithm - I could implement to speed the task up, i.e., CAs are intrxtable.
CIt is still a little dissatisfying to regard these patterns as emergent as they are simply
an alternative expression of the CA rule itself, cut-off at an arbitrary time step.
 121

Figure 2. CA &36 after pattern basis filtering

dynamical behaviour that is entirely different (from a language perspective

at least) from the micro-physics of the CA. If we see these two ‘physics’
side by side (Figure 3) it is plainly clear how different they are. The micrc-
physics description uses a microlanguage16 (which is sometimes referred
to as the ‘language of design’) comprised only of 0s and Is, whereas the
macrolanguage is comprised of an entirely different alphabet: a , p, y+,
y-. There is no simple mapping between these two languages/physics, i.e.,
there is no way (currently) to move between the two languages without
some kind of information loss - the translation is lossy (as it is in the
translation of human languages). As S t a n d i ~ h ’remarks,
~ “An emergent
phenomenon is simply one that is described by atomic concepts available
in the macrolanguage, but cannot be so described in the microlanguage”(p.
4). This imperfect relationship comes about because, although the macro-
physics is a good theory, it is not a complete description of the micro-physics,
it captures some of the details but not all. A qualitative change in language
is equivalent to a change in ontology, but what is the ontological status of
the emergent products?
At the micro-level, only cells (that can only be in states ‘0’ or ‘ 1’)
absolutely exist. But can we say with equal confidence that a-particles,
for example, exist? Given the filtered version of the space-time diagram
showing the apparent evolution of a set of quasi-particles, as well as their
compelling macro-physics, it is very tempting indeed to answer in the af-
122

111 110 101 100 011 010 001 000

-
a+y.--r y-+a+2y+
y+ + a 2y + a + 'y
R + VI - b y +'

0 0 1 1 0 1 1 0
+
; fj-,v-
Y++Y"P
. .
7' + a + y- + y + a + y'
y' + p + y- + 0
(a) (b)

Figure 3. (a) The micro-physics of level 0 and (b) the macro-physics of level 1

firmative. However, we need to bear in mind how the original patterns

were created and how we uncovered the quasi-particle dynamics. The pat-
terns were generated by recursively applying a micro-rule to each cell in
the system. Each cell is treated exactly the same as the others - none re-
ceive any special or unique treatment. The application of the rule does
not change as the system configuration evolves. And, most importantly,
each cell in the configuration depends on the states of all the other cells - a
particular sequence of 0s and 1s representing the evolution of a particular
cell, only occurs that way because all the other cells evolve in their par-
ticular way. Although the rule is only applied locally, the dynamics that
emerge are a result of global interactions. So a-particles only emerge be-
cause P-particles emerge (as well as the other particles) - the emergence of
a-particles does not occur independently of the emergence of ,&particles.
As a result it would be a mistake to assume that some absolute level of
ontological autonomy should be associated with these particles, though of
course the macro-physical description provides evidence that making such
an assumption is for many intents and purposes useful. However, as with all
assumptions, just because making the assumption leads to useful explana-
tions, it does not follow that the assumption is correct. Even the motion of
these particles is only apparent, not real; it is simply the result of reapply-
ing the CA rule for each cell at each time step. Whilst discussing Conway's
Game-of-Life (a form of two-dimensional CA), philosopher Daniel Dennett6
asks:
[Slhould we really say that there is real motion in the Life world, or
only apparent motion? The flashing pixels on the computer screen
are a paradigm case, after all, of what a psychologist would call
apparent motion. Are there really gliders [or, quasi-particles] that
move, or are there just patterns of cell state that move? And if we
 123

opt for the latter, should be say at least that these moving patterns
are real?

Choosing ‘0’ and ‘1’ to represent the two cell states perhaps further
suggests the illusion of motion and ontological independence.
The choice of ‘0 ’ and ‘ 1’ is a little misleading as it can easily be inter-
preted as being ‘on’ and ‘off’ or ‘something’ and‘nothing’. Whether a
cell is in state ‘0 ’ or ‘ 1’, the cell itself still exists even if a‘black square’
is absent. We could easily use ‘ x ’ and ‘ N ’ to represent the two states
and the result would be the same. In uncovering the macro-dynamics we
had to, however reasonable it may have seemed, choose a pattern-basis to
be ‘removed’ or filtered-out. By clearing-out the space-time diagram we
allowed ‘defects ’ in the space-time diagram to become more clearly appar-
ent. We also chose to set to ‘0’ those areas of the space-time diagram that
were tiled evenly with the pattern-basis, and ‘ 1’ those regions which repre-
sented distortions in the pattern-basis; we assumed an even ‘background ’
and then removed it to make visible what was in the ‘foreground’d. Again,
it is tempting to interpret the ‘0’-covered (white areas) as ‘nothing’ and
the ‘1’-covered (black) areas as ‘something’. This would be incorrect
however, as all the cells that comprise the CA network are still there, we
have simply chosen a representation that helps us to ‘see’ the particles
and ignore the rest. However, it must be noted that though ‘0’ and ‘1’
do not map to ‘nothing’ and ‘something’ at the CA-substrate level, this
is exactly the mapping made at the abstracted macro-level.
So in any absolute sense, the emergent products do not exist as such.
They do however exist in the alternative (although deeply - complexly -
related) macro-level description of the CA. The change in perspective is
equivalent to an ontological shift, where the existence of the emergent prod-
ucts is a ‘feature’ of the new perspective though not of the absolute CA
perspective; what existed at the micro-level does not exist at the macro-
level and vice versa. This new perspective is an incomplete description,
yet the fact that the new description is substantially complete would sug-
gest the substantial realism of the particles on which the new description is
based. For this reason we say that though the particles do not absolutely
exist, they do indeed exhibit substantial realism and as such, for many in-
tents and purposes, can be treated as if they were real (for this reason they

~~~

dAt the CA-substrate level the background is not ontologically different from the fore-
ground - any difference is achieved through filtering.
124

are often referred to as ‘ quasi-particles’ rather than just ‘particles ’).

Quoting Feyerabend again,
Abstractions remove the particulars that distinguish an object from
another, together with some general properties ... Experiments fur-
ther remove or try to remove the links that tie every process to its
surroundings - they create an artificial and somewhat impoverished
environment and explore its pec~liarities.~
The process of abstraction is synonymous with the filtering process de-
scribed above. Feyerabend goes on to say (p. 5):
In both cases, things are being taken away or ‘blocked off’ from
the totality that surrounds us. Interestingly enough, the remains
are called ‘real‘, which means they are regarded as more important
than the totality itself. (italics added)
So, what is considered real is what still remains after the filtering process
is complete. What has been removed is often called background ‘noise’,
when really what we have accomplished is the removal of ‘details’ not
‘noise’, for the sake of obtaining a representation that we can do ‘science’
with.

4. The Hierarchy of Reality: Nested versus Convoluted

This view of existence changes the popular image regarding the hierarchy
of physical matter considerably. The popular image of the hierarchy of
matter (illustrated in Figure 4ae) is that starting with atoms (for argu-
ment sake - of course more fundamental (or, more real) entities are known
to the physical sciences) molecules are formed which are little more than
agglomerations of atoms. Here, atoms are regarded as the ‘micro’ and
molecules are regarded as the ‘macro’, and the filter applied is some mys-
terious process known as the ‘scientific method ’ (although any competent
and experienced scientist would rightly question the existence of such a
universal met hod). From molecules, molecular complexes are formed, from

=In considering Figure 4b it is very tempting indeed to assume that the extra dots (in
comparison to Figure 4a) are no more than noise. However, it is important to note (but
not necessarily easily grasped) that the existence of the seemingly more real pattern
(represented in Figure 4a) depends upon the existence of what is easily labelled ‘noise’
and vice verse. Each ‘noisy’ dot is potentially significant in the future evolution of
the system as a whole as it is a part of the system, not just the result of some impure
measurement.
 125

which cells are formed, from which multi-cellular organisms are formed, and
so on right up to galactic clusters and beyond. In such a neat nested hier-
archy every ‘entity’ has a place which distinguishes it from other entities
as well as determining which ‘science ’ should be employed to understand
it. Furthermore, every ‘thing’ at one particular ‘level’ in the hierarchy is
made-up off those ‘things’ said to exist at the ‘level’ below. This would
seem to suggest that molecular physics/chemistry, say, should in principle
be derivable from atomic physics. However, even though such a trick is
supposedly possible (assuming such a neat hierarchical organisation), in re-
ality scientists are very far indeed from pulling off such a stunt. Nowadays
the reason offered for this is simply the intractability of emergence which
is well known to nonlinear researchers. Rather than derive biology from
chemistry, scientists are forced to assume that the ‘entities’ of interest
(which are the result of a particular filtering, or coarse-graining) to their
particular domain do exist as such. Once such a step is taken, a science can
be developed independently of the science that deals with the supposedly
‘simpler ’ ‘ lower-level ’ ‘entities ’. Making such an assumption (which is
often made implicitly rather than explicitly) has worked out rather well,
and an enormous amount of practical understanding has been accumulated
that allows us to manipulate these ‘entities ’ in semi-predictable ways, even
if they don’t happen to truly exist! Technology is the obvious example of
such success.

4.1. Non-composite secondary structures

As well as the background ‘ pattern basis’ and the fundamental particles
listed in Figure 3, CA &36 “supports a large number of secondary struc-
tures, some having the domain and particles as building blocks, some not
describable at the domain/particle 1evel”ll . Though it has been acknowl-
edged above that a complete ‘physics‘ of one level cannot be derived from
the ‘physics’ of the layer below, it is often assumed that the lower-level
contains all that is necessary to allow the higher level to emerge. This
assumption needs to be explored further, but just because we don’t have
access to a perfect mapping from the lower level to the next level up, it
doesn’t mean that some day such a bootstrapping method will not be un-
covered.
In the CA model the fundamental ‘lowest’ level - level 0, say - does
indeed contain ‘everything’ there is to know about the CA universe; the
space-time diagram is indeed a complete description. However, the move to
126

Figure 4. (a) Abstracted ‘reality’ versus (b) reality?

the particle level description - level 1 - is an imperfect ontological shift. As

such, the set of quasi-particles describing the particle-level description is not
s a c i e n t to describe all the structures that are observed at the lower level
(as a result of ’background’ filtering), i.e., one cannot reverse the filtering
process and perfectly reconstruct the micro-level. Hansen and Crutchfield’s
discovery that secondary structures also exist, indicate that another level of
‘existence’ - level 2 - can be uncovered. if all of these secondary structures
were made up of the particles already described ( a ,p, etc.) then we could
argue that a ‘composite ’ particle level (2) exists ‘above‘ the quasi-particle
level (1). In the natural sciences this would suggest that chemistry is indeed
derivable from physics, or that all of chemistry emerges from all of phgszcs.
However, the existence of “secondary structures’‘ that are not “describable
at the domain/particle level” suggests that there are entities existing at
the higher composite level (2) that can’t possibly be explained in terms
of the particle level (1) description, even intractability was not an issue,
i.e., there exist higher-level entities that are not combinations of a , p, or y
quasi-particles. Another way of saying this is that matter is not conserved
between these ascending levels. Assuming these observations have merit
in the ‘real‘ world, this would suggest for example that chemistry cannot
 127

possibly be derived from physics in its entirety. This perhaps is no great

surprise; physics is an idealisation and so a chemistry derived directly from
such a physics could only represent an idealistic subset of all chemistry even
if we could somehow overcome the computational irreducibility of predicting
the macro from the micro.
This observation indicates a different meaning to the aphorism “the
whole is greater than the sum of the parts.“ Traditionally this has only really
meant that, because of networked causality, the properties of the whole can
not be directly derived from an analysis of the interacting parts; basically,
even though our ontology is (assumed to be) correct, our analytical tools
aren’t quite up to the challenge of reconstructing the whole from its parts.
Novelty in this particular interpretation simply comes from the fact that our
‘ science ’ is unable to predict the macroscopic behaviour from microscopic.
The existence of ‘ non-composite secondary structures ’ suggests that, even
if we could bootstrap from the micro to the macro, the ‘parts’ do not
offer a complete basis from which to derive the ‘whole’f; the ‘whole’ is
not greater than the sum of its ‘parts’ because the ‘whole’ is not even
constructed from those ‘ parts’ (except in idealised computer models). In
this context we would expect to observe chemical phenomena that could
never be explained in terms of physics (however complete physics might get
- in this view of the Universe physics can never be complete), and that
might even contradict predictions made at the level of physics. There are
entities that ‘exist’ within a chemical description of ‘reality’ that could
never be explained in terms of atoms, say.
The fact that each science has facilitated such enormous strides in our
ability to rationally affect our environment suggests that, though they may
be based on incomplete ontologies, they are in fact very good partial reflec-
tions of ‘reality’. The hierarchy of sciences is sustainable and justifiable,
but is not an accurate representation of what is. However, the assumption
that the physical matter the sciences attempt to explain organises into a
nested hierarchy would seem to be questionable. Regarding our hierarchy of
‘explanations ’ as convoluted rather than nested would be more appropriate
in light of the CA model.

‘Unless of course we were bootstrapping from level 0 to level 1, as entities existing at

level 0 are by definition deemed to absolutely exist.
128

4.2. Does level of abstraction correspond to degree of

realism ?
Given that as we move away from the absolute CA reality information
is lost, it would be tempting to suggest that the ‘entities’ that ‘exist’
at subsequently higher levels of description, though substantially real, are
increasingly less real. If our descriptions at one level were always tied
to descriptions at lower levels then this argument would be supportable.
But given that particular scientific communities will often operate with
little regard to the ontological and epistemological standards of another
scientific community, it is quite possible that a macro-explanation contains
entities that are more real than those comprising, what seems to be, a micro-
explanation. If this wasn’t the case and we did indeed tie ourselves to purely
quantum physical explanations, say, it is quite possible that, as a result of
the multiple abstractions that would be necessary to get from quantum to
social reality, we would learn next to nothing about that particular ‘reality‘
(as we would have a partial description of a partial description of a partial
description. . . ).
The fact that we are comfortable with ignoring quantum descriptions
and approaching ‘social reality’ in terms that seem more familiar to us (af-
ter all our human perceptions have been honed to give us rather good tools
for dealing with this particular level of reality) enables us to make rather
more progress in regard to the development of practical understanding for
manipulating social reality.
In short, a hierarchy of vertical ontologies can be inferred that, though
they may actually be inconsistent (or even, incommensurable) with each
other, they can still be incredibly useful in providing practical understand-
ing. The hierarchy of physical reality is not a nested hierarchy like the
hierarchy of the sciences has been traditionally regarded.

5. Multiple Filters and Horizontal Ontologies

Given the fact that the process of filtering-out pattern bases to reveal
‘macroscopic ’ dynamics works so well in the example above, it would be
tempting to assume that even though higher-level descriptions are incom-
plete, there is in fact a ‘ best ’ way to obtain these descriptions. So although
the discussion thus far supports the assertion that the recognition of emer-
gent products is tied to our choice of methodology, it also supports (or
at least it does not deny) the possibility that one particular methodology
is better than all others. I would like to briefly challenge this possibility
 129

next. This section of the chapter represents a move away from the more
scientifically-basedphilosophy presented thus far to a more speculative phi-
losophy.
As we move up our constructed hierarchy of explanations, the reality
that is ‘explained’ by our explanations seems to be more complex. By
complex here I mean that we seem to have increasing difficulty in locat-
ing the boundaries, or the ‘quasi-particles’ that describe our systems of
interest. This issue is perhaps exemplified at the ‘social level’, and more
clearly illustrated by the methodological diversity of the social sciences.
Whereas recognising the ‘atoms’ of the physical sciences is seemingly a
rather straightforward affair (though still only approximately representa-
tive of what is), the recognition of the ‘atoms’ of social science seems rather
more problematic. There are many more ways to ‘atomise‘ social reality in
an effort to explain a particular phenomenon (of course our choice of ‘atom-
isation scheme’ determines the phenomena that then takes our interest, and
to some extent, vice verse - although there are potentially many different
(possibly contradictory) atomisations ’ that may lead to seemingly good
explanations for the same phenomena). This is probably why social prob-
lems are often described as ‘messy’ problems” ; the articulation of the
system of interest itself is problematic, let alone the development of expla-
nations. The fact explanations of social phenomena often have low ‘fitting
coefficients‘ whereas explanations of physical phenomenon often correlate
very well indeed with perceived (measured) reality, may further suggest the
legitimacy of multi-methodological approaches in the social science+.
I don’t want to spend much time at all defending the view that multi-
methodological approaches for the social sciences are a must. For now I am
content to point the reader towards the book Multimeth~dology~~ which
argues rather well for a pluralist attitude in the social sciences. I think that
the failed attempts in the past to develop a ‘Theory of Society’ satisfies
me, for now at least, that pluralism is a feature of the social sciences rather
than a failure of social scientists to construct such an all-embracing theoryh.

gLow ‘ fitting coefficients ’ for explanations of social phenomena is clearly an indication

that the explanations proposed are quite incomplete, suggesting the necessary consider-
ation of alternative explanations. Whereas at the level of superstrings we might get very
close t o the ‘truth’ with only one model, we might find that ‘truth’ at the social level,
say, is at least the sum of all possible models, i.e., a monist notion of truth may work for
‘low-level’ explanations, where as a pluralist notion of truth is essential for explanations
of ‘ higher-levels ’.
hThe fact there is no complete theory of anything would suggest the importance of
pluralism in all realms, but it is clear that some realms of human understanding have
130

-mE
._
00 0 0
0 0 0
0
-

O O G
O&-
a Level 0 - CA ‘Reality’: no hierarchy

Figure 5. The convoluted hierarchy of existence. Each ellipse represents the set of
‘entities’ (a theory’s alphabet) that supposedly account for certain phenomena at a
.
certain level ’

The different explanations of the same phenomena offered in the social

sciences often rely on different ontological assumptions. Given our histori-
cal tendency to rank theories in terms of hierarchies of physical matter, we
might ask ourselves as to how we might go about ranking social theories.
In other words, are the entities used to construct explanation A more real
that those entities used to construct explanation B? I very much doubt
that a consistent answer exists for this particular question. The reason
this issue is raised here is simply to support the suggestion that as we ab-
stract away from the fundamental substrate of existence (level 0) towards
the more familiar social level we experience a kind of ontological explosion
(see Figure 5). This explosion occurs in explanation-space and represents
the theoretical diversity inherent in our attempts to explain phenomenon
in our terms rather than the Universe’s terms (whatever they might be).
As the explosion grows in intensity we find that there are many more legit-
imate ways to approach our explanation building. Each approach relies on
different ways to recognise the ‘atoms’ of what is and how they interact
with each other, i.e., each approach recognises the products of emergence
in a different way. Each approach is only capable of explaining certain
aspects of the phenomenon of interest, encouraging the increased need for
pluralist approaches. The notion of a ‘level’ becomes rather vague, as does
the notion of a hierarchy; even a hierarchy of explanations is challenged.
Taking this need for pluralism to its extreme leads us into the realms of
relativistic philosophies, as opposed to the realist philosophies that tend to

developed very well indeed without taking pluralism too seriously. For those fields to
continue to develop though I would expect pluralism to become more common. So all
my comments above regarding social sciences are true also for physics, just not equally
so.
 131

reign in the physical sciences. So how do we decide which are the ‘best‘
approaches that lead to the ‘best ’ theories? How do we avoid the choice
of ‘filter’ from being completely arbitrary? In the physical sciences consis-
tency with currently accepted theories is often used as an additional filter.
Equally important is partial validation through experimentation, and lim-
ited empirical observation. Physical scientists are rather fortunate in that
many of their systems of interest can be isolated to such a degree that the
scientists can actually get away with the assumption that the systems are in-
deed isolated. This helps a great deal. In the social sciences there is a more
‘try it and see ’ attitude as the application of reductionist methodologies is
highly problematic, because of the considerable difficulties in reproducing
the same experiment time and time again. Often new situations, or new
phenomena, force the social scientist back to square one and a new ‘ex-
planation ’ is constructed from scratch with minimal consideration of what
theories went before (although what went before may prove inspirational).
The fact that pattern-basis filtering of complex CA leads to alterna-
tive representations of the CA reality that account for a good percentage
of the systems behaviour would seem to suggest that an ‘anything goes’
relativism in particular contexts is not supportable. The success of physics
itself at least supports the notion that some theories are considerably bet-
ter (in a practical sense at least) than others. However, the fact that these
alternative representations are incomplete would suggest that an ‘anything
goes ‘ relativism might be valid for all contexts taken together at once. Once
a pluralist position is adopted, the key role of critical thinking in taking us
from a universal ‘anything ’ to a particular ‘something’ becomes apparent.
I do not want to explore any further the epistemological consequences
of CA research as this paper is primarily concerned with the ontological
status of emergence products (which, as has been indicated, is impossible to
explore without concerning oneself with epistemology). If our commitment
to a CA model of the Universe is to persist then we need to find CA systems,
and the tools for analysing them, that when abstracted multiple times (to
‘higher levels’ of existence) lead to representations that can be usefully
‘atomised’ and analysed in a variety of different ways (analogous to the
social sciences).

5.1. A note on intrinsic emergence

Before closing this chapter with some concluding statements I’d like to
mention the notion of intrinsic emergence introduced by Crutchfield as it
132

offers a computational-based explanation as to why some theories / perspec-

tives / abstractions / etc. are better than others and why we should resist
an “anything goes” relativism. According to Crutchfield’ , “The problem
is that the ‘newness ’ [which is referred to as ‘novelty ’ herein] in the emer-
gence of pattern is always referred outside the system to some observer
that anticipates the structure via a fixed palette of possible regularities.”
This leads to the relativistic situation mentioned above. Although there is
clearly a relativistic dimension to pattern recognition (and therefore theory
development) how might we resist a radical relativism (that would suggest
that anything goes in any context all the time)? Crutchfield’ goes on
to say, “What is distinctive about intrinsic emergence is that the patterns
formed confer additional functionality which supports global information
processing... [Tlhe hypothesis ... is that during intrinsic emergence there is
an increase in intrinsic computational capability, which can be capitalised
on and so can lend additional functionality.” Another way of saying this is
that, if we privilege a computational view of the Universe, the structures
that emerge support information processing in some way - they increase
the computing ability of the Universe. This would suggest that ‘real’ pat-
terns are those patterns which can be linked to some computational task.
For example, considering Figure 2 again, if a filter has been chosen that
was not based upon the pattern basis then a pattern would still have been
uncovered, but it is unlikely that, that pattern would have any relevance to
the computational capacity of the system - the odds are the pattern would
also appear quite disordered. This of course opens up the possibility that
the best ‘filters’ are those that lead to regular patterns, but given that
assessments of regularity (or disorder) lie on a spectrum how do we decide
which patterns are sufficiently regular to allow us to associate with them
some functional importance? Even though certain patterns play a func-
tional role within the Universe (which suggests in absolute terms that not
all patterns are ‘real’), it does not follow that we, as participants in the
Universe, can unambiguously determine which patterns are significant and
which are not. It is not even clear if an ‘outsider’ (God, perhaps) could
achieve absolute clarity either.
The notion intrinsic emergence shows that the patterns that we might
consider ‘real’ allow the system of interest to process information in some
way, suggesting that certain patterns are internally meaningful and not
arbitrary. However, it does not follow that just because some patterns a
more ‘real’ than others we can determine which are which. In a universal
and absolute sense, not all filters and patterns are made equal and certain
 133

filters (and their consequent patterns) are more meaningful than others. In
the absence of direct access to absolute and complete descriptions of reality,
however, an ‘anything goes ’ relativism (such as Feyerabend’s) cannot be
dismissed on rational grounds alone and offers the possibility of a more
genuine engagement with the (unseen) ‘ requirements ’ of certain contexts.

6. Summary and Conclusions

In summary I’d like to ‘atomise’ (and therefore abstract in an incomplete
way) the assertions made herein concerning the nature of emergents:
0 Emergent products appear as the result of a well-chosen filter -
rather than the products of emergence ‘showing’ themselves, we
learn to ‘see’ them. The distinction between macro and micro is
therefore linked to our choice of filter.
0 Emergent products are not real though ”their salience as real things
is considerable.. . “ 6 . Following Emmeche et aL7, we say that they
are substantially real. In a limited sense, degree of realism is a
measure of completeness relative to level 0.
0 Emergent products are novel in terms of the micro-description,
i.e., they are newly recognised. There is no such thing as absolute
novelty. Even an ‘idea‘ has first to be recognised as such. There
will be novel configurations of cell states, but no new real objects
will emerge, though new substantially real objects can emerge.
0 In absolute terms, what remains after filtering (the ‘foreground’)
is not ontologically different from what was filtered-out (the ‘back-
ground ’); what is labelled as ‘not-real ’ is ontologically no different
from what is labelled ‘real ’.
0 Product autonomy is an impression resulting from the filtering prc-
cess.
0 The products of emergence, and their intrinsic characteristics, oc-
curring at a particular level of abstraction do not occur indepen-
dently - rather than individual quasi-particles emerging indepen-
dently the whole set of quasi-particles emerges together. The hier-
archy of explanations correlates only approximately with degree of
realism.
0 The emergent entities at level n+l are not derived purely from the
entities comprising level n, even if a perfect bootstrapping method
that could overcome the problem of intractability was invented.
0 Emergent products are non-real yet substantially real, incomplete
134

yet representative;
0 Determining what ‘macro’ is depends upon choice of perspective
(which is often driven by purpose) and of course what one considers
as ‘micro’ (which is also chosen via the application of a filter).
0 All filters, at the universal level, are equally valid, although certain
filters may dominate in particular contexts.
One last remark. Much philosophical debate circles around what the
most appropriate filters are for understanding reality. For example, radi-
cal scientism argues that the scientific methods are the most appropriate,
whereas humanism argues that the filters resulting from culture, personal
experience, values, etc. are the most appropriate. An analysis of a CA
Universe suggests that it is impossible to justify the privileging of any filter
over all others on rational grounds (though there is a way to go before we
can say that this conclusion follows naturally from the CA view). Whereas
political anarchism is not a particularly effective organising principle for
society at large, a complexity informed theory of knowledge does tend to-
wards an epistemological and ontological anarchism, at least as a Universal
default position.

7. Appendix A: On the Reasonableness of Assuming that

the Universe is, at Some Level, Well-Described as a
Cellular Automaton
Given that the strength of one’s claims is related (but not necessarily soi)
to one’s starting point, I would like to discuss the reasonableness of making
the CA-assumption. I would also like to explore briefly what does not follow
from assuming the existence of a theory of everything.
Firstly, although some form of CA could possibly turn-out to be the
theory of everything, I do not believe that inhabitants of the Universe will
ever (knowingly) pin-down the details of such a model - the reason for this
being that, due to the nature of nonlinearity (assuming the Universe to
be inherently nonlinear), there simultaneously exist an infinity of nonlinear
(potentially non-overlapping) models that will provide (potentially contra-
dictory) ‘explanations ’ for all observed ‘ facts ’. Laplacian determinism
may well operate, but it certainly does not follow that we can observe this
exquisite determinism in any complete sense (though clearly we can to a

‘It is quite possible for the conclusions to be correct whilst the reasoning could be falla-
cious.
 135

very limited degree).

Secondly, even if we serendipitously stumbled upon THE theory of ev-
erything (though we would never know that we h a d ) , just because we have
a theory of everything, it by no means follows that we can derive ezplana-
tions for every thing. All we would have is a theory of how the fundamental
units of existence interact. Bootstrapping from that primary CA substrate
to ’higher levels’ of existence is considerably more challenging than dis-
covering such a fundamental theory in the first place. As Hofstadter12
remarks:

a bootstrap from simple molecules to entire cells is almost beyond

one6 power to imagine. (p. 548)

I argue herein that not only is information lost as we ‘abstract ’ to more

familiar ‘objects’ (which is the very nature of abstraction), but that there
is no single foolproof method to select the ‘best ’ abstraction (even if there
is such a ‘best’ abstraction).
So, claiming that the Universe might be well-described at some level as
a CA is not as big a claim as it might first seem. In the argument I develop
above, the details of such a CA are not really important for the validity of
the conclusions I draw. But why start with a CA, even if the details are
not particularly essential. My reason is simply parsimony; a CA model is
possibly the simplest construction currently known to us that comes close
to having the capacity to contain everything - this possibility arises because
of the model’s recursive nature. With his Life game Conway has shown
“that the Life universe . . . is not fundamentally less rich than our 0wn.”14.
There are a couple of things we can additionally suggest about such
a CA model of the Universe. All CA universes are finite universes and
as such certain configurations will eventually repeat. When this happens
the Universe would fall into a particular attractor in which a sequence of
configurations would be repeated for eternity. The idea that our Universe
is currently racing around some high-period attractor cycle seems unlikely
(given the novelty we observek),or at least at odds with our perspective on

j ” ...and
of course the clear and certain truth no man has seen nor will there be anyone
who knows about the gods and what I say about all things. For even if, in the best case,
one happened to speak just of what has been brought to pass, still he himself would not
know.” Xenophanes of Colophon (fragment B34).
kNot that the observation of novelty by a participant within the Universe is proof enough
that novelty ‘exists’. Plus, the idea of the big bang being following by a big crunch may
well suggest that there is indeed some colossally-high period attractor leading cosmic
136

things. The phase spaces of CA Universes contain many such attractors,

highlighting the potential import of initial conditions'.
If our CA Universe is not living some endlessly repetitive nightmare
(claimed by big bang / big crunch models), what might be going on instead?
Many of the attractors of CA systems exist at the end of fibrous tentacles of
configurations that do not occur on any attractor cycle. Figure 6 illustrates
this. These transients are the 'gaps' between the system's initial conditions
and its final attractor. These transients can have very complex structures
and can even support lifelike behaviour. So here is the CA Universe;
from a range of initial conditions (all of which could lead to a statistically
identical evolution) the Universe would pass through an inordinate number
of unique configurations which would never be repeated again, before it
finally settled into a fixed pattern of cycling configurations in which no
such novelty would ever exist again (unless of course the rules of the game
were changed or some external force pushed the system back onto one of
the transient branches - which would of course simply result in a partial
rerun). We are quasi-entities living on a transient branch of a CA attractor!
It would seem that in addition to long transients such a CA would
also have to display complex transients, as opposed to ordered or chaotic.
Wuenschelg defines complex CAs as those with medium input-entropy and
high variance. An example of a CA displaying complex transient dynamics
is rule &6cle53a8 (lc-5). The space-time diagram and the phase space for
this rule are shown in Figure 7.
Though it by no means demonstrates the validity of the CA-hypothesis
in itself, an interesting calculation might be to determine the size of CA
needed that would have transients at least as long as the time the universe
has existed thus far. We'll consider the rule above for this little exercise.
Let's assume that one time-step in our CA Universe is equivalent to the
small measure of time known as Planck Time"'. Figure 8 plots the maxi-
mum transient length for rule &6cle53a8 for increasing CA size (increasing
n). If we assume that the Universe is approximately 15 billion years old
(or 4.7 x 1017 s) then we require transients at least 8.77 x lo6' steps long.
If we approximate the relationship between maximum transient length and

evolution.
'Though there are CA models whose phase spaces are characterised by a single attractor
- in such circumstances initial conditions are quite independent of the system's end state
(cycle).
mPlanck time, which is 5.4 x s, is the age of the Universe at which energies were
available to enable particle interactions to occur across Planck distances.
 137

Figure 6. The largest attractor for CA rule &6cle53&, ~ 1 7k.5 , showing complex
transient trees of length 72 steps. The period of the attractor cycle is 221 steps

network size shown in Figure 8 by a power law and extrapolate to 8.77

x 1060 (which is a rather extreme extrapolation I know) then we obtain
a colossal value of for n, the network size, whereas we get a value
for n of 600 if we assume an exponential relationship". This number
is either outrageously large or small compared to what we might expect.
Chown20 estimated the number of distinct 'regions' that might exist to
be loQ3,so we're way off that mark! However, it does serve to illustrate

"As a first approximation I found the relationship between network size, TZ, and transient
length, La%,
(for rule &6cle53&, k.5) to be:

n e e x p [ - 11n ( k > ]
0.0126 3.0228
if a power law is assumed, or:

n m - -1- - - - - I n ( k )
0.2368 0.9952
if an exponential law is assumed.
""The only way one region of space can 'know' about the conditions in another is if
there has been time for some kind of influence to go back and forth between them-and
the maximum speed of any influence is that of light. Hence, the early Universe consisted
of los3 regions which were entirely separate, or 'causally disconnected"'2.
138

Figure 7. A section of the space-time diagram for CA rule &6cle53a8, -17, k 5 , and
the complete attractor field which contains 54 attractors (21 are qualitatively distinct -
only these are shown)

that CA may be ‘complex’ enough to support the ‘existence’ of complex

entities and maybe even ‘Everything‘. A simple 2-state CA of only 600
cells can have transients as long as 1061 steps containing no repeated con-
figurations. This does not confirm that such a system could contain exotic
quasi-entities such as humans, but Hansen and Crutchfield (1997) have at
least confirmed that simple hierarchies of interacting quasi-particles can be
uncovered using appropriate filters. Chou and Fteggia (1997) have even
found that constructing CAs that support self-replicating structures is a
surprisingly straightforward exercise. It would be interesting to perform
the same calculation for CAs with cells that can take on more than 2 states
to see if network sizes of log3 can be obtained, although there is no reason
to assume this estimate to be correct in the first place.

References
1. Bedeau, M. (1997) “Weak emergence,” Philosoph~calperspectives, 11: 375-
399.
2. Chown, M. (2000) “Beforethe Big Bang,” New Scientist, 166(2241): 24.
3. Corning, P. (2000) “The reemergence of ‘emergence’: A venerable concept
in search of a theory,” C o r n p ~ e ~7(6):
t ~ , 18-30.
 139

5 10 15 20 25 30

Figure 8. Maximum transient length versus network size for CA rule &6cle53a8, n=5-
25, k 5

4. Corning, P. (2000) “The emergence of ‘emergence ’: Now what?” Emergence,

4(3): 54-71.
5 . Crutchfield, J. P. (1994) “Is anything ever new? Considering emergence,”
in G. A. Cowan, D. Pines, and D. Meltzer (eds.), Complexity: Metaphors,
Models, and Reality (pp. 515-537). Cambridge, MA: Perseus Books.
6. Dennett, D. C. (1991) “Real patterns,” The Journal of Philosophy, 88: 27-51.
7. Emmeche, C., Koppe, S. and Stjernfelt, F. (2000) “Levels, emergence, and
three versions of downward causation,” In P. B. Andersen, C. Emmeche,
N. 0. Finnemann, and P. V. Christiansen (eds.), Downward causation (pp.
13-34). Aarhus, Netherlands: Aarhus University Press.
8. Feyerabend, P. (1975) Against Method, London, UK: NLB.
9. Feyerabend, P. (1999) Conquest of Abundance: A Tale of Abstraction Versus
the Richness of Being, edited by Bert Terpstra, The University of Chicago
Press: Chicago.
10. Goldstein, J. (1999) “Emergence as a construct: History and issues,” Emer-
gence, l(1): 49-72.
11. Hansen, J. E. and Crutchfield, J. P. (1997) “Computational mechanics of
cellular automata: An example,” Physica D, 103( 1-4): 169-189.
12. Hofstadter, D. R. (1979) Godel, Escher, Bach: A n Eternal Golden Braid,
New York, NY: Basic Books.
13. Mingers, J., and Gill, A. eds. (1997) Multimethdology: The Theory and Prac-
tice of Combining Management Science Methodologies, Chichester, UK: John
Wiley & Sons.
14. Poundstone, W. (1985) The Recursive Universe: Cosmic Complexity and the
Limits of Scientzfic Knowledge, New York, NY: William Morrow.
15. Richardson, K. A. (2004) “The problematisation of existence: Towards a phi-
140

losophy of complexity, Nonlinear Dynamics, Psychology and the Life Sci-

”

ences, 8(1): 17-40.

16. Ronald, E. M. A., Sipper, A., and Capcarrre, M. S. (1999) “Testing for
emergence in artificial life,” in Advances in Artificial Lzfe: 5th European
Conference, E C A L 99, volume 1674 of Lecture Notes in Computer Science,
Floreano, D., Nicoud, J. -D., and Mondad, F. (eds.), pp. 13-20, Springer:
Berlin.
17. Standish, R. K. (2001) “On complexity and emergence,” Complexity Interna-
tional, 9, available at http://journal-ci.csse.monash.edu.au//vo109/standi09/
18. Vennix, J. A. M. (1996) Group Model Building: Facilitating Team Learning
Using Systems Dynamics, Chichester, UK: John Wiley & Sons.
19. Wuensche, A. (1999) “Classifying cellular automata automatically: Finding
gliders, filtering, and relating space-time patterns, attractor basins, and the
Z parameter,“ Compledty, 4(3): 47-66.
 TRUTH IN COMPLEX ADAPTIVE SYSTEMS MODELS
SHOULD BE BASED ON PROOF BY CONSTRUCTIVE
VERIFICATION

DAVID SHIPWORTH
Faculty of Science
University of Reading

It is argued that the truth status of emergent properties of complex adaptive

systems models should be based on an epistemology of proof by constructive veri-
fication and therefore on the ontological axioms of a non-realist logical system such
as constructivism or intuitionism.
‘Emergent’ properties of complex adaptive systems (CAS) models create particu-
lar epistemological and ontological challenges. These challenges bear directly on
current debates in the philosophy of mathematics and in theoretical computer sci-
ence. CAS research, with its emphasis on computer simulation, is heavily reliant
on models which explore the entailments of Formal Axiomatic Systems (FAS). The
incompleteness results of Godel, the incomputability results of Turing, and the Al-
gorithmic Information Theory results of Chaitin, undermine a realist (platonic)
truth model of emergent properties. These same findings support the hegemony
of epistemology over ontology and point t o alternative truth models such as intu-
itionism, constructivism and quasi-empiricism.

1. The epistemology of complex adaptive systems

This discussion of epistemology starts with a discussion of the nature of
formal systems as these are the principle epistemic mechanisms operating
within model theory which itself is the dominant epistemology of com-
plex adaptive systems research. In this context ‘model theory’ is used in
the sense of the epistemology of mathematical This discussion
leads from an analysis of the limitations of formal systems to the choice of
ontological logical axioms based on these limitations.

1.1. Formal systems

Mathematical modeling involves mapping certain selected observables of
the system being modeled onto mathematical systems and in turn onto
the formal systems of programming languages (Figure 1 below). The fact

141
142

that the mathematical structures into which the selected observables are
encoded are formal systems is of particular importance in light of the known
incompleteness of such systems. The mathematician David Hilbert, at the
beginning of the 20th century, presented a challenge to the mathematical
community. That challenge was to demonstrate the consistency of the
axioms of mathematics. This challenge became known as the ‘formalist
program’ or f o ~ m a l i s r n . ~
It was well known by the end of the lgth century that mathematical
terms had both a syntactic and a semantic content [2] and it was this, the
formalists argued, that gave rise to some well documented apparent contra-
dictions within mathematics, such as the Russell para do^.^ The syntactic
content of mathematics is that part of mathematics that involves the ap-
plication of rules to strings of symbols. Certain of these strings of symbols
are taken as the axioms of the system; other symbol strings are created
through the application of the rules to these axioms to create new symbol
strings. This aspect of mathematics is purely logical and the symbol strings
mean nothing. It is this aspect of mathematics that was being described by
Bertrand Russell when he claimed that ‘pure mathematics is the subject in
which we do not know what we are talking about, or whether what we are
saying is t r ~ e ’ . ~ J ’ . ~ ~ ~
The semantic component of mathematics arises when meaning is at-
tached to these symbols. This meaning can be as simple as letting these
symbols represent number^.^ Without the semantic component of mathe-
matics, mathematical modeling is impossible, as it requires the encoding
of the observables of the natural system into the language of mathematics.
Likewise, it is impossible to conceive of any applied CAS model without
semantic content. It was this ‘meaning’ component of mathematics which
Hilbert and the other formalists thought gave rise to mathematical contra-
dict ions.
Hilbert could not tolerate any such contradictions and sought to elim-
inate them through elimination of the semantic content of mathematics.
Hilbert argued all semantic functions of terms could be captured by syntac-
tic rules. As Rosen [2, p.71 notes, ‘In the Formalist view, then, Mathemat-
ics ... is a game of pattern generation of producing new patterns (theorems)
from given ones (axioms) =cording to definite rules (production rules)’. In
this view mathematics is not ‘about’ anything, it is simply an exercise in
symbol manipulation according to a set of rules.
This view of mathematics is highly amenable to computation. Com-
putation is the process of applying predetermined transformation rules to
 143

a set of inputs to generate a set of outputs. The process is determinis-

tic, predictable, replicable and has of itself no semantic content. To cite
Rosen [2], computers, and their theoretical archetypes Turing machines, are
‘...purely syntactic engines’. As such, they have enormous appeal within
the formalist program. Indeed Turing was attempting to verify one of the
central tenants of formalism, that the truth of all mathematical statements
was decidable through a finite number of applications of the rules of the
formal system, when he developed the Turing m a ~ h i n e . ~
Unfortunately for Hilbert, Turing, and others working on the formal-
ist program, the logician Kurt Godel, in 1930, proved the entire formalist
program was logically impossible. As Rosen [2, p.71 noted ‘Godel’s In-
completeness Theorem devastated the entire Formalist program’. Godel
proved that any given formal system was either logically inconsistent, or
incomplete. As the former option of inconsistency is anathema to all math-
ematicians, Godel’s proof is almost universally taken to demonstrate that
such formal systems are incomplete. As Casti [6, p.121 notes:

...Godel’s Incompleteness Theorem, ... tells us that no system of

deductive inference is capable of answering all questions about num-
bers that can be stated using the language of the system. In short,
every suficiently powerful, consistent logical system is incomplete.

Rosen notes that Godel’s Incompleteness Theorem has the following

implications. That ‘...Number Theory is already too rich in entailments
(or equivalently that machines are too poor in entailments) for there to
be any kind of convergence between them. ...’ and that ‘...Mathematics,
like natural language itself, cannot be cut off from all referents and remain
Mat hemat ~ c s . ’ ~ + ~
This sets profound theoretical limits on the power of mathematical mod-
els. It means that no model of a system can, even in theory, ever capture
all that can be said about the system that it models.
Formal Axiomatic System (FAS) is defined by Chaitin as:

[A Formal Axiomatic System] consists of a finite set of axaoms and

of a finite set of rules of inference for deducing the consequences of
those ~ x i o m s ~ + ~

Chaitin also notes that:

The sets of axioms that mathematicians normally use are fairly

concise, otherwise no one would believe in them. In practice,
144

there’s this vast world of mathematical truth out there-an infi-

nite amount of information-but any given set of axioms only cap-
tures a tiny, finite amount of this information. That, in a nutshell,
is why Godel incompleteness is natural and inevitable rather than
mysterious and complicated.s~p.171

As noted by Chaitin [4, p.111, ‘Theorems are found by writing all the
possible grammatical statements in the system and testing them to deter-
mine which ones are in accord with the rules of inference and are therefore
valid proofs.’ That is, a statement can be grammatically correct and yet still
be false. Indeed, only a small proportion of grammatically correct state-
ments of any given formal system are provable as theorems of that system.
Hilbert, in developing the mathematics of formal systems as part of the
formalist program, devised a means of constructing all provable theorems
within any given formal system.
An alternate way to view the structure of a Formal Axiomatic System is
as a mathematical graph ( GFAS)(a set of ‘vertices’ (points) joined by a set
of ‘edges’ (lines)). The vertex set ( V F A S of
) such a graph would correspond
to the union of the set of axioms of the FAS, and the set of all syntactically
correct statements provably true through application of the FAS’s rules of
inference to the set of axioms of the FAS and subsequently derived theo-
rems. The edge set ( EFAS)would be defined by any two vertices connected
by application of the FAS’s rules of inference to the set of axioms of the
FAS and subsequently derived theorems. Here we follow the logic used by
Godel in the incompleteness theorem and equate ‘truth’ with ‘proof’ and
note that in the context of a FAS, proof means something that is provable
through execution of a properly constituted algorithm on a Universal Tur-
ing Machine. By definition, this graph G F A would ~ have one connected
component containing all vertices. It is to be remembered however that
FAS are literally meaningless-i.e. they are devoid of semantic content.
The use of FAS to model any meaningful system fundamentally changes
this graph. Chaitin [9, 41 studied the arithmetic of real numbers (a seman-
tic system) through analysis of real number Diophantine equations. He
found that there are an infinite number of essentially random but provably
true arithmetic facts that cannot be derived from any other facts through
application of the rules of any FAS model of the arithmetic of real numbers.
Following from above, we define the graph of the semantic system of
arithmetic G s e r n - ~as
, the vertex set V s e r n -of
~ all syntactically correct
and provably true statements within the semantic system of arithmetic. We
 145

define the edge set E s e r n - as

~ the union of the edge sets ( E F A S )of all legit-
imately constructed FAS models of arithmetic. We know from Chaitin that
each individual FAS model of arithmetic defines a connected component on
this graph of arithmetic. Further, differently defined formal systems will
have, as provably correct theorems of that system, different sets of ver-
tices on the graph G s e r n - ~ That
. is, different formal systems will, through
defining different edges on the graph, link different vertices of the graph into
connected components. However Chaitin [9, 41, through his construction of
the universal Diophantine equation (R), has demonstrated that the graph
of arithmetic also contains an infinite number of disconnected vertices that
are arithmetically provably true, but which are impossible to prove true
) any FAS model ( GFAS)of arithmetic.
(i.e. unreachable on G s e r n - ~by
Chaitin notes with respect to the Diophantine equations which consti-
tute these disconnected vertices:

Mathematical reasoning is essentially helpless in such a case, since

there are no logical interconnections between the Diophantine equa-
tions generated in this way. ..Randomness, uncertainty and unpre-
dictability occur even in the elementary branches of number theory
that deal with Diophantine equations4

This discovery is central to the epistemology of complex systems in

which the permutation spaces defined by the building blocks of the systems
and the rules of inference governing their combination into aggregates is
approaching the complexity of that defined by number theory. Indeed as
Morowitz [lo, p.195-61 notes:

This is one of the core elements of the series of studies designated

as complexity-the combinatorics of a set of simple constructs op-
erating jointly produces an array of possibilities too large to deal
with, sometimes transcomputational.

This is interesting in the light of Chaitin’s [4, p.181 comment:

Although Godel’s original theorem seemed to apply only to unusual

mathematical propositions that were not likely to be of interest in
practice, algorithmic information theory has shown that incom-
pleteness and randomness are natural and pervasive.

The argument is that if a comparatively simple and well-defined system

such as Diophantine equations can never be captured by the construction
146

of any formal system there is every reason to suppose that the same will
apply to more complex systems.
This has two implications that are central to this argument. Firstly, as
noted above, any mathematical model executed on a digital computer is by
definition a formal axiomatic system. This FAS model (syntactic) defines
an edge set ( E F A S )on the vertex set of all provably true statements about
the semantic system Vs,, it is modelling. The graph created by this vertex
set and edge set contains a connected component, (the set of entailments of
the model), and an indeterminate number of disconnected vertices. Each of
these vertices represents a ‘fact’ about the system being modelled but one
that is unprovable by that model. A model with different axioms, and/or
rules of inference, defines a different FAS and a different set of truths. Again
the graph thus created has a connected component and an indeterminate
number of disconnected vertices. Note that while any number of models
can be defined, we know from Godel and Chaitin that none of them can
connect all the vertices on the graph. This is precisely the point made by
Rosen [2, p.8-91 when he notes that ‘The relation of Number Theory to any
formalization of it is the relation of the complex to the simple...We shall
argue that a formalization constitutes a (simple) model of that which it
formalizes’. Rosen [2, p.91 goes on to note that:

For our purposes, the point of doing all this is that we are going
to identify this category of simple models of a system of entailment
with ‘what we can know’ about it. ... When we do this, all of
epistemology gets wrapped up in that category of models. ‘What we
can know’ about such a system thus becomes a question of how good
we are at navigating in this category of models.
Casti reaches the same conclusion. ‘The best we can hope for is to illu-
minate various facets of natural systems using different - and inequivalent
- formal d e s c r i p t i ~ n s ’ . ~ ~ ” ~ ~ ~ ~ ~ p . ~ * ~
It should be noted that the formal results of incompleteness of Godel
and Chaitin pertain to the relationship between mathematical systems such
as arithmetic, and formal systems models of them. This is illustrated in
Figure 1 in the lower half of the diagram. In the quotes above however, both
Rosen, and more explicitly Casti, suggest a parallel if less formally defined
incompleteness relationship exists between natural systems and mathemati-
cal model of them (the relationship on the left hand side of Figure 1).While
this is intuitively reasonable, extrapolating from the relationship between a
formal system and the mathematical system it models, to a mathematical
 147

Natural
System f
% f
Computer
Simulation
5’
1

f$
Mathematical
System
7 ~
Formal
System

Figure 1. Modelling relation (synthesising elements from Casti [6] and Traub [ll])

system and the natural system it models is contentious and ultimately an

argument from analogy. It is not central to the argument of this paper.

1.2. Mathematical Realism us. Intuitionism

Chaitin’s construction of the universal Diophantine equation, raises another
issue, and one which bears directly on the ontology of complex adaptive sys-
tems models. This point revolves around the question of what constitutes
a valid proof. What constitutes a valid proof also determines what we ac-
cept as being true for, as Snapper [12, p.541 points out ‘...we prove that
something is t r u e ’ . What constitutes a valid proof is in turn, dependent on
the logic system subscribed to. For classical mathematics and conventional
science this is the classical logic laid down by the Greeks. Classical logic is
founded on the three laws of Aristotelian logic: the law of identity, the law
of non-contradiction and the law of the excluded middle.3+’.22 These are
ontological laws, they are not proved in any sense, they are effectively the
axioms of classical logic. Of these ontological propositions, the ‘law of the
excluded middle’ is not accepted by mathematical intuitionists. The argu-
ment here is that it should not be accepted by complex adaptive systems
modellers either.
The law of the excluded middle states that any declarative s e n t e n c e
that is any grammatically correct sentence in a formal or natural language
which makes an assertion about a fact-has to be either true or false. It
has to be either true or false because it either does, or does not, correspond
to the fact about which it is making an assertion. This correspondence is
148

wholly independent of any verification of the proposition of the declarative

sentence, or of the individual making the declaration. As Snapper notes,
classical or Platonic truth is ‘verification Subscribing to the
law of the excluded middle imposes a Platonic character to truth. That
is, the truth or falsity of the sentence exists independently of time and
independently of any observer. In the Platonic notion of truth, a sentence
is true if it corresponds to the fact about which it is making the assertion.
This is therefore also known as the correspondence theory of truth.
Mathematical intuitionists, such as Brouwer, Heyting and Dummett,
do not subscribe to the law of the excluded middle.13 If, ontologically, one
rejects the law of the excluded middle, then it is impossible to say anything
about the truth or falsity of a declarative sentence before its truth or falsity
has been verified in some epistemologically acceptable manner.
Just as classical mathematics is based on an epistemology of classical
logic, intuitionistic mathematics is based on its own system of intuitionistic
logic formalized by Heyting.14 In intuitionistic mathematics, intuitionistic
logic defines what constitutes a valid proof, i.e. it defines its epistemology.
Intuitionistic mathematics also gives hegemony to epistemology over on-
tology. While in classical logic truth (ontology) logically pre-exists, and is
independent of, proof (epistemology), in intuitionistic mathematics proof
(epistemology) logically pre-exists, and is independent of, truth (ontology).
Here proof defines truth rather than simply resolving its status. In in-
tuitionism, prior to a properly constructed verification of a sentence’s as-
sertion, the sentence has no truth value ascribed to it. It is neither true
nor false but simply of indeterminate truth status. The ontology of intu-
itionistic truth is termed ‘constructivist’ and truth is established through
a process of ‘constructive verification’. It is strictly inductive in nature,
rejecting some classical proofs derived from deductions based on prior on-
tological assumptions of binary [true; false] truth states. It is a pragmatic
theory of truth that is based on the work of the philosophical pragmatists
William James, Charles Peirce and John Dewey.12
Chaitin notes (with some irony) both the role Godel and computation
have played in strengthening the arguments for such an inductive, construc-
tivist epistemology.

‘...measuring the complexity (algorithmic information content) of

m i o m s and showing that Godel incompleteness is natural and ubiq-
uitous, deepens the arguments that forced Godel, in spite of himself,
in spite of his deepest instincts about the nature of mathematics, to
 149

believe in inductive mathematics. f53P.4

‘So in the end it wasn’t Godel, it wasn’t Turing, and it wasn’t my

results that are making mathematics go in an experimental math-
ematics direction, in a quasi-empirical direction. The reason that
mathematicians are changing their working habits is the computer.
I think it’s an excellent joke! (It’s also funny that of the three old
schools of mathematical philosophy, logicist, formalist, and intu-
itionist, the most neglected was Brouwer, who had a constructivist
attitude years before the computer gave a tremendous impulse to
constructivism.) ’16~P.26

There are both pragmatic and philosophical reasons not to subscribe

to the law of the excluded middle. Many of the philosophical arguments
are presented in the collective works of the mathematical intuitionists and
presented in edited volumes on the philosophy of mathematics such as Be-
nacerraf and Putnam [13] and Hart [17]. While these reasons are persuasive,
they are not discussed here. Likewise, the arguments of the intuitionists for
rejecting the proof mechanisms and class creation mechanisms of classical
mathematics are not addressed.
The relevant issue is, given what we know of the character of complex
adaptive systems, and the means at our disposal to study such systems,
whether the truth status of complex adaptive systems models should be
based on constructive verification or be verification free.

2. Characteristics of CAS theory and practice

There are two dominant characteristics of complex adaptive systems the-
ory and practice which are relevant to this decision, namely our meth-
ods of studying them using computer experimentation, and their appar-
ent non-conformity to the dominant scientific reductionist ontology of uni-
directional causality. These issues are addressed in turn.

2.1. The epistemology of computer experimentation

Until fairly recently scientists could be catagorised as either experi-
mentalists or theoreticians. Over the past thirty years or so a new
type of scientist has emerged who does not fall easily into either of
these two categories as he deals with neither real systems nor the-
ory in the strictest sense. He is the computational scientist. The
150

computational scientist uses the computer as his main apparatus to

perform computer ‘experiments’ or computer simulations.

This quote by Jain [18, p.l] introduces one of the central aspects of the
epistemology of complex adaptive systems theory, that is, the increasing
role played by computer simulation and experimentation as a mechanism
for understanding certain classes of systems.
Indeed Coveney and Highfield [3, p 2771 note that, in the CAS field
of artificial life ‘The most important contemporary Alife system, Tierra,
has an evolutionary dynamic of such great complexity that the only way
to investigate its behaviour is to perform experiments on it within a com-
puter.. .’.
The relationship between complex adaptive systems theory and com-
puter simulation is well expressed by Holland [19, p.951.

Because CAS are so intricate, computer-based models, with their

well defined, manipulateable mechanisms, provide a crucial inter-
mediate step in the search for CAS laws. Such models, where they
mimic relevant CAS phenomena, put CAS data in a rigorous for-
mat, thereby facilitating the descriptions of patterns and laws.

Another aspect central to the importance of simulation in the complex

adaptive systems field is that simulation allows for the exploration of the
behaviour of systems that could, for practical or ethical reasons, never be
experimented on. This is particularly so for many ecological or biophysi-
cal systems for which experimentation is either impossible or would yield
results which are not generalisable. Complex adaptive systems theorists ar-
gue that their models can illustrate the mechanisms driving the behaviour
of whole classes of such systems, thus allowing for the development of ‘qual-
itative’ understandings of such s y ~ t e m s . ~ ~ ~ P . ~ ~
Chaos theory has shown that there are classes of systems that are gov-
erned by deterministic laws, but which are intrinsically unpredictable. This
is the study of deterministic chaos, which has been well documented in the
popular scientific press.20 The discovery of the existence of such systems
raises profound epistemological questions.
If we accept that there are classes of systems that are deterministic but
the behaviour of which is intrinsically unpredictable. If we also accept that
the identification of emergent properties of classes of deterministic systems
such as fitness landscapes or cellular automata is likewise a priori impos-
sible from an analysis of their component parts. Then the fields of chaos
 151

theory and complex adaptive systems theory to which these observations

are central, must adopt an epistemology which accepts as ‘true’ those sys-
tems properties which can be constructively demonstrated to exist. That is,
a property of a system exists where it can be constructively demonstrated
to exist through an appropriately constituted computer experiment, and
where its existence can be verified by reference and replication of that ex-
periment.
The statement ‘system X exhibits property Y’ is neither true nor false
until it has been constructively demonstrated to be either true or false by
an appropriately constituted computer experiment. Such a method of proof
is termed ‘proof by constructive verification’ and is as has been discussed
the hallmark of the intuitionistic philosophy of mathematics.

2.2. The ontological rejection of uni-directional reductionist

causality
The term reductionism is used in two distinct ways in the description of
knowledge systems. The most widely understood use of the term reduction-
ism describes systems of knowledge that only permit objects to be described
using a restricted set of descriptors and relations. The arguments outlined
above about the incompleteness of formal axiomatic systems shows that
complex adaptive systems modeling is reductionist in this sense.
A second and more formal definition of reductionism is a descriptor of
the ontological position that all properties of a system observable at one
hierarchical level are explicable through analysis of the system’s compo-
nents at a lower hierarchical level. This is the reductionism of Newton
and Descartes, that is as dominant an ontology of ‘conventional’ science as
adherence to the law of the excluded middle is of ‘conventional’ mathemat-
ics. Conventional science assumes that things can always be completely
explained by looking at their components. This is reductionist in the sense
that the characteristics and behaviour of anything can be reduced to the
characteristics and behaviour of its parts. Here causation is uni-directional
and always operates from smaller to larger scales.
Morowitz [lo] argues that an important aspect of the property of
emergence in complex adaptive systems is that it is fundamentally anti-
reductionist. In complex adaptive systems displaying emergent properties,
the rules governing the behaviour of elements at one hierarchical level are
frequently explained by (emergent) properties only observable at a higher
hierarchical level. CAS epistemologists such as Morowitz [lo] and Rosen [2]
152

argue that this, while scientifically unconventional, is still a perfectly valid

form of science. As Morowitz [lo, p.1961 points out

Under this mode of operation the successful theories may gener-

ate emergent properties not directly derivable from or perhaps even
imagined from the simple constructs but b y running the program.
The emergence is one of the desirable features of complexity theory.

In complex adaptive systems it is argued that it is possible to explain

the existence and behaviour of elements of a system in terms of the function
that system itself serves with larger systems. Here causation can operate
in both directions (smaller H larger) across all hierarchical levels of system
scale.
Subscription to the concept of bi-directional causality differentiates com-
plex adaptive systems theory from the conventional scientific ontology of
uni-directional causality. Consequently, to the extent to which reduction-
ism is defined as subscription to the ontology of uni-directional causality,
complex adaptive systems theory is non-reductionist.

3. Conclusion: Truth in complex adaptive systems models

should be based on proof by constructive verification
We have seen from the arguments on the use of the computer as an ex-
perimental tool that the epistemology of complex adaptive systems accords
with a system of proof by constructive verification. Further, if we accept
that bi-directional causality across scales is a defining ontological feature
of complex adaptive systems, then identification of emergent properties of
classes of such systems is a priori impossible from an analysis of their com-
ponent parts. It therefore follows that the existence of such properties can
only be demonstrated through adoption of an epistemology which accepts
as ‘true’ those system properties which can be constructively demonstrated
to exist. That is, emergent properties of complex systems can only be said
to exist if we can construct systems that display those emergent proper-
ties. Because, as Morowitz notes, such properties frequently cannot even
be imagined, it is not possible to postulate their existence through the for-
mulation of declarative sentences within the language of the formal systems
of complex adaptive systems models. Because it is not possible to formu-
late appropriate declarative sentences which proclaim the existence of such
emergent properties, it is not possible to ascribe truth value to them on
the basis of their correspondence with ‘facts’ which, likewise, cannot be
 153

imagined.
Indeed it is even questionable as to whether such qualitatively different
system properties operating at higher hierarchical levels can be expressed
within the limited vocabulary of the formal systems of CAS models. This
is because emergent properties are robust statistical aggregates of what are
frequently transcomputational sets of the combinatorics of deterministic
relations between elements of the system at lower hierarchical levels. For
this reason, the detection of emergent properties through the analysis of
individual deterministic relations is not possible.
For these reasons, it seems unreasonable, to ascribe truth value to a
priori unimaginable and qualitatively different properties of systems that
are both structured and limited by our choice of formal axiomatic system.
We therefore see that both the epistemological practice of CAS re-
searchers, and one of the defining ontological assumptions of CAS theory,
lie in opposition to the dominant deductive ‘realist’ ontology of classical
mathematics based on an ontological acceptance of the law of the excluded
middle. Rather, CAS ontology and epistemological practice lie in accord
with the constructivist epistemology of intuitionism and other inductive
logical frameworks. Thus, it is argued that the truth status of emergent
properties of complex adaptive systems models should be based on strictly
inductive logics and on proof by constructive verification.
Accepting this leads to very real implications for building CAS models
in practice. It becomes necessary to ask the question whether there are
computer hardware or software environments that operate in a way that
either uses, or entails, principles of mathematical logic based on adherence
to the law of the excluded middle that the strictly constructive proofs of
intuitionistic mathematicians would reject? The answer is yes. The most
notable example of this is the use of ‘autoepistemic logics’ in multi-agent
CAS models.
Autoepistemic logics are a class of non-monotonic logics and were devel-
oped by Moore [21] and Przymusinski [22]. Autoepistemic logic introduces
an additional syntactical element to those of standard propositional logic.
This additional element is the modal operator 0 representing knowledge.
This is used to indicate knowledge of the truth status of a declarative sen-
tence (S) and is used as follows: US indicates that the truth status of S
is known; 04indicates that the truth status of 4 is known; and TOS
indicates that the truth status of S is not known.
This operator is used, in conjunction with an implicit acceptance of the
law of the excluded middle, to support reasoning based on knowledge of
154

facts. This supports the reasoning that 10s -4,i.e. S is assumed to be

false unless it is known to be true. Thus autoepistemic logics, while not
strictly classical, are nevertheless predicated on an acceptance of the law
of the excluded middle and contain deductive elements unacceptable t o the
strictly constructive proofs of intuitionists.
Autoepistemic logics have been used for multi-agent modeling in Pro-
log and other languages. An example used in the field of organizational
modeling is the SDML language developed by Moss and o t h e r ~ . ~T~h?e ~ *
logic underpinning this agent based simulation environment enables deduc-
tive logic through adherence t o the law of the excluded middle and is thus
unacceptable from a strictly inductive, intuitionistic perspective.
It is argued therefore, that autoepistemic logics (or other non-monotonic
logics based on the closed world assumption) are incompatible with the epis-
temological practice of CAS researchers and on bi-directional causality, one
of the defining ontological assumptions of CAS theory. Emergent proper-
ties of CAS models based on such logics should therefore be regarded as
philosophically invalid from a coherent complex systems perspective.

References
1. J. Casti, Reality Rules: 1 & 2-Picturing the world in mathematics-
‘Vol. 1-The Fundamentals’ & ‘Vol. 2-The Frontier’, New York, Wiley-
Interscience, John Wiley & Sons (1992).
2. R. Rosen, ’What can we know?’, in Beyond Belief: Randomness, Prediction
and Explanation in Science, eds J. Casti and A. Karlqvist, CRC Press, p.310
(1991).
3. P. Coveney and R. Highfield, h n t i e r s of Complexity: The search for order
in a chaotic world, London, Faber and Faber (1995).
4. G. Chaitin, Information Randomness €9’ Incompleteness: Papers on algo-
rithmic information theory, Znd edn, Series in computer science - Vol. 8,
Singapore, World Scientific (1990).
5. J. Casti, ‘Chaos, Godel and Truth’, in Beyond Belief: Randomness, Predic-
tion and Explanation i n Science, eds J. Casti and A. Karlqvist, CRC Press
(1991).
6. J. Casti, ‘The outer limits: In search of the ‘unknowable’in science’ in Bound-
aries and Barriers: On the limits to Scientific Knowledge, eds J. Casti and
A. Karlqvist, Reading Massachusetts, Addison-Wesley (1996).
7. G. Chaitin, ‘Thoughts on the Riemann Hypothesis.’ The Mathematical In-
telligencer, Vo1.26, No.1, pp.4-7 (2004).
8. G. Chaitin, ‘Computers, Paradoxes, and the Foundations of Mathematics’,
American Scientists, Vol. 90, pp.164-171, (2002).
9. G. Chaitin, Algorithmic Information Theory, Cambridge, Cambridge Univer-
sity Press (1987).
 155

10. H. Morowitz, ‘Complexity and Epistemology’ in Boundaries and Barriers:

On the limits to Scientific Knowledge, ed. J. Casti, p.188-199, Reading Mas-
sachusetts, Addison-Wesley (1996).
11. J. Traub, ‘On Reality and Models’, in eds J. Casti and A. Karlqvist, Bound-
aries and Barriers: On the limits to Scientific Knowledge, Reading Mas-
sachusetts, Addison-Wesley, p.238-255 (1996).
12. E. Snapper, ‘What do we do when we do mathematics?’, The Mathematical
Zntelligencer, Vol. 10, No. 4, pp.53-58 (1988).
13. P. Benacerraf and H. Putnam, (eds) Philosophy of Mathematics: Selected
Readings, Znd edn, Cambridge, Cambridge University Press (1983).
14. A. Heyting, Intuitionism, an introduction, 2nd edn. Amsterdam, North Hol-
land (1966).
15. G. Chaitin, ‘Two philosophical applications of algorithmic information the-
ory’, Springer Lecture Notes in Computer Science (Proceedings DMTCS’03),
eds C. Calude, M. Dinneen and V. Vajnovszki, Springer, Vol. 2731, pp.1-10
(2003).
16. G. Chaitin, Randomness in arithmetic and the decline and fall of reduction-
ism in pure mathematics, IBM Research Report No. 27, IBM (1992).
17. W. Hart, (ed) The Philosophy of Mathematics, Oxford, Oxford University
Press ( 1996).
18. S. Jain, Monte Carlo Simulations of Disordered Systems, Singapore, World
Scientific (1992).
19. J. Holland, Hidden Order: How adaptation builds complexity, Reading Mas-
sachusetts, Addison-Wesley Publishing company (1995).
20. J. Gleick, Chaos: Making of a new science, New York, Viking (1987).
21. R. Moore, ‘Semantical considerations on nonmonotonic logic.’ Artificial Zn-
telligence, Vol. 25, pp.75-94 (1985).
22. T. Przymusinski, ‘On the declarative semantics of deductive databases and
logic programs’, in Foundations of deductive databases and logic program-
ming, ed J. Minker, San Francisco, Morgan Kaufmann Publishers, pp.193-216
(1988).
23. S. Moss, H. Gaylard, et al. ‘SDML: A Multi-Agent Language for Organiza-
tional Modelling’ Computational €9 Mathematical Organization Theory, Vol.
4,NO. 1, pp 43-75 (1998).
24. 0. Teran, B. Edmonds, et al. ‘Mapping the Envelope of Social Simulation
Trajectories’, Proceedings of The Second Workshop on Multi Agent Based
Simulation, Boston, Springer Verlag (2000).
 COMPLEXITY AS AN EPISTEMIC REVOLUTION:
CONSIDERATIONS ON THE NEW SCIENCE IN THE
CONTEXT OF WESTERN INTELLECTUAL HISTORY

DAMIAN POPOLO
Durham University, UK

The paper will seek t o present a genesis of complexity in European philosophical

thought. Following the works of Gleick and Hobsbawn it is possible to discern the
emergence of complexity’s ethos in the ‘age of revolutions’. Gleick, for example,
has noted the influence of natural philosophy on the first precursors of chaos the-
ory. Natural philosophy itself was deeply anchored in German romanticism. The
paper will thus seek to present, following Foucault, an ‘Archaeology of Complexity’
which considers Foucault’s definition of epistemes as evolving modes of thought. In
particular, the paper will seek to use Foucault’s description of the modern episteme
in order t o detect the novelty inherent within the ethos of Complexity.

Cause and effect: such a duality probably never exists, - in truth we

are confronted by a continuum out of which we isolate a couple of pieces,
just as we perceive motion only as isolated points and then infer it without
ever actually seeing it. The suddenness with which many effects stand out
misleads us; actually, it is sudden only for us. A n intellect that could see
cause and effect as a continuum and a flux and not, as we do, in terms
of arbitrary division and dismemberment - would repudiate the concept of
cause and effect and deny all conditionality.
Friedrich Nietzsche, The Gay Science
All is flux, nothing stays still.
Heraclitus, from Diogenes Laertius, Lives of Eminent Philosophers
Greek philosopher (540 B C - 480 BC)

1. Introduction
There has been some debate recently surrounding the question of whether
Complexity could be seen as a manifestation of ‘post-modern’ science’.
Some scholars have sought to underline the similarities between Complexity
science and post-structuralist p h i l o s ~ p h y ~ ySome
~ > ~ .others have perceived

156
 157

Complexity as being simply the manifestation of what Lakatos would call

a ‘progressive paradigm’. In this view, the emergence of notions such as
‘deterministic chaos’ suggest that Complexity is nothing more than a more
sophisticated tool for the understanding of patterns that just happen to ap-
pear as being completely random - but that remain patterns nonetheless a .
Others have suggested that Complexity, in a number of fields, has changed
many paradigmatic assumptions to such a point that it can now claim to
be a legitimate scientific revolution in the Kunhian sense6.
However, any attempt to contextualise and define Complexity should
also have a historical backing for its arguments, otherwise, words such as
‘modernity’ and ‘post-modernity’ run the serious risk of being nothing but
convenient terms deprived of any meaning. In this context, Dillon has
reminded us of the Nietzschean belief according to which only things which
do not have a history can be defined?. Complexity, as a set of ‘research
principles’, does have a history - a history which enables us to capture its
ethos rather than formulating its essentialist definition. This paper argues
that Foucault’s enquiry into forms of knowledge ( episternesb) could and
should provide a context to capture such ethos.
The Foucauldian framework presented here has the advantages of giv-
ing a coherent context for the understanding of modernity’s ethos, without
necessarily ‘defining’ it, while at the same time allowing us to consider as
‘post-modern’ (strictly in this Foucauldian sense) any construct of knowl-
edge that does not adhere to modernity’s conceptualisation of t i m e , for
modernity is conveniently described as the ‘age of (a precise form of lin-
ear) History’. According to Foucault, the modern emergence of time is
related to the need of placing contingency upon the objects of enquiry’.
Complexity Science - understood as a set of knowledge practices which, as
in Prigogine’s caseg, postulate the central importance of time irreversibil-
ity - could be understood as an epistemic revolution. Moreover, questions
surrounding the way in which notions of temporality should be handled in
research and philosophical enquiry have been asked throughout the history
of Western thought.
Thus, Foucault - through a precise analysis of the modern episteme -
allows us to put some order in the debate by characterising Complexity not

aByrne, for instance, considers Complexity exclusively for its potential in quantitative
social science, and believes that the new science represents a serious challenge to ‘post-
modernism’. See
bFoucault describes an episteme as the ensemble of the underlying rules that characterise
the general configuration of knowledge within a precise historical context8.
158

simply as that which looks for indeterminist, as opposed to deterministic,

scientific laws; not simply as a field of knowledge that calls for interdisci-
plinary inquiry and recognises the non-linear character and the intercon-
nectedness of all things; but as a set of practices which depart in their quest
for knowledge on the ‘post-modern’ assumption that the arrow of time ex-
ists, and that time is irreversibly a constitutive element in the formative
processes of things and not simply a convenient and contingent parameter.
Foucault’s rigorous analysis of the modern episteme allows for a detailed
‘Archaeology of Complexity’ to take place. Through these epistemic lenses,
this paper will seek to identify the genesis of Complexity in the history
Western thought, before addressing the question of whether Complexity’s
ethos is indeed surpassing the rules of the modern episteme. Additionally,
this research will allow us to consider to what extent Complexity represents
the reemergence of questions and ideas that have emerged in the European
intellectual past, ideas which were precisely discarded by the triumph of the
modern episteme.
This paper will thus commence with a Foucauldian description of the
modern episteme. This will give a firm ground to the concept of moder-
nity, which will allow subsequent sections to deploy the word in a more
rigorous manner. Second, the paper will make a short enquiry into modern
Continental philosophical history in order to develop a genesis of Complex-
ity. Finally, it will conclude with a few examples of how contemporary
Complexity thinking is indeed re-deploying arguments present in Europe’s
intellectual heritage which - seen through epistemic lenses - can indeed be
considered as post-modern.

2. The Modern Episteme

Foucault defines the era of the modern episteme as the ‘age of History’*
P.217. While in the Classical age elements were classified according to their
identities and differences (that is, their representations), that is by the spa-
tial position they occupied in the ordered tables constituted by taximonia
and mathesis‘; the collapse of representation obliges elements to be classi-
fied according to their proximity in the temporal succession of events. We
assist therefore to the substitution of the principle of the Classical tables
- Order, with the principle inherent in organic structures - History. As

Foucault puts it: ‘History gives place to analogical organic structures [the

CFora description of taximonia and mathesis consider chapter 3 in

 159

modern episteme], just as Order opened the way to successive identities

and differences [the classical e p i ~ t e m e ] P.219.
’~ This sudden mutation is
responsible for a double fragmentation of knowledge: not only modes of en-
quiry will differ according to how they deal with the concept of temporality,
but emerging empirical ‘sciences’ or disciplines will be formed around the
perceived ‘history’ of the elements that compose them. This second event
leads to the birth of organic structures as fields of knowledge, fields that
will be centred on historical transcendental notions.
The Classical configuration of knowledge did not require a concept of
temporality, as Gutting puts it: ‘...there was no important distinction be-
tween analytic (a priori) and synthetic (a posteriori) knowledge. An anal-
ysis of representations in terms of identities and differences was at the
same time a connection (synthesis) of them in the ordered tables that ex-
press their essential reality’”. Nonetheless, with the decline of the Clas-
sical episteme, representation cannot be regarded as the sole foundation
of knowledge. Thus occurs a crucial schism. Foucault explains that once
the representational foundations of thought fade, analytic disciplines are
founded on epistemological grounds that fundamentally differ from the ba-
sis upon which synthetic disciplines are based. The result is that ‘on the
one hand we have a field of a priori sciences, pure formal sciences, deduc-
tive sciences based on logic and mathematics, and on the other hand we
see the separate formation of a domain of a posteriori sciences, empirical
sciences, which deploy the deductive forms only in fragments and in strictly
localised r e g i ~ n s ’ ~ P . ~ ~ ~ .
Philosophy, especially after Kant, is given the task to resolve the conflict
between the emerging transcendentalisms and the world of representation.
Representation, in this sense, is ‘essentially the relation between a sub-
ject and the object it thinks and experience^'^^*.^^^. Now this is how an
embryonic conceptualisation of temporality is introduced within the gen-
eral frameworks of knowledge. For this relation can be analysed either
in terms of the conditions that precede and therefore ground the possi-
bility of any representation or in terms of how the experiencing subject
stands in relation to already represented objects (metaphysics). The first
solution corresponds to the creation of a transcendental philosophy of the
experiencing subject, while the second leads to a transcendental philosophy
of the experienced objects (metaphysics). It is this metaphysical solution
that characterises modern, pre-Complexity scientific epistemology. Both
solutions respectively correlate to the appearance of analytic and synthetic
sciences. Finally, both solutions equally assume that it is necessary to con-
160

nect representations in terms of which our experience occurs to either a

subject or an object that lies outside that experience but grounds its pos-
sibility. Philosophical criticism will therefore question representation from
the a priori requirement of an experiencing subject, while metaphysics will
seek to understand representations in terms of the experienced objects. In
both cases, representation is not a causal source of explanation in itself.
Foucault shows how the sciences of economics, biology and philology
adopt a particular position within the modern organisation of knowledge.
These fields clearly enter the category of synthetic sciences that seek to un-
derstand the world of representations through transcendental philosophies
of objects (metaphysics). As Foucault put it, ‘Labour, life, and language ap-
pear as so many ‘transcendentals’which make possible the objective knowl-
edge of living beings, of the laws production, and of the forms of language.
In their being, they are outside knowledge, but by that very fact they are
conditions of knowledge. . . This corresponds to the point enunciated
above which explains that a transcendental philosophy of objects connects
representations in terms of which our experience occurs to an object that
lies outside our experience but grounds its possibility. Nonetheless, this
philosophy necessitates a crucial analytic of finitude and a corresponding
notion of linear temporality to operate.
Let us take the analysis of the discipline of economics to illustrate the
point. Foucault argues that in the Classical age only purely representa-
tional sources of value were commonly considered. The value of things was
established through the notion of exchange: this was a purely represen-
tational system, for the value of an object was always represented in the
value of another. Things did not really have a proper value as such. That
is why in the Classical age economic ‘history’ did not exist: ‘value is always
related to the system of exchange, in which any changes in value are always
correlated with changes in other values. Since the system always involves
a finite number of correlated values, changes can only be ~ y ~ l i ~ a l ’ ’ ~ ~ ~ ’ ~
Foucault sees Smith as the first step towards modernity. Smith intro-
duced the idea that labour, and not exchange, was the source of value.
Nonetheless, Smith considered that this value still necessitated a repre-
sentational system of exchange to operate. Indeed, Smith’s thesis is that
a thing can purchase another object that embodies a relative quantity of
labour: the value of labour is determined by exchange. It is Ricardo who
operates the decisive break from the Classical era. After Ricardo, labour
is considered to be the sole source of value, regardless of how much this
labour can represent in the system of exchange. F’rom now on, ‘a theory of
 161

production must always precede a theory of c i r c ~ l a t i o n ’ ~ pBut

. ~ ~the
~ . tran-
scendental notion of labour, the object through which the subject economic
‘man’ will be analysed, necessitates two fundamental notions of finitude.
Firstly, the notion of anthropological finitude ensures that the subject
‘man’ is studied through the object labour. While the technocrats analysed
how men stood in relation to representations, ‘modern economics deals with
factors that causes men to form such representations - for example, with the
bareness of nature that forces us to work, with the bodily deficiencies and
the external threats that limit our ability to produce. . .modern economics
is ultimately based. . . on an ‘anthropology’ of human f i n i t ~ d e ’ l ~ p A .~~~.
problem, however, remains. How do we regard ‘man’ (or any other object
of metaphysical enquiry) simply as a constituted subject, how can we stop
this man from building alternative forms of production, how can we make
sure that a transcendental philosophy of the object does not become a
transcendental philosophy of the subject? This is precisely the problem
that the realm of philosophy must try to resolve, and that is how ‘the
study of man, precisely as a constituting subject, is the central concern
of modern philosophical reflection.. . ’10P.199. What it is usually referred
to in the social sciences as the ‘structure and agency debate’ and in the
natural sciences as ‘paradigmatic assumptions’ are issues inherent within
the modern episteme. These issues are about setting firm parameters of
enquiry.
Historical finitude is crucial for the placement of paradigmatic contin-
gency upon the subjects of enquiry. Ricardo demonstrates that there is
a historical precedent in all forms of production and capital accumulation
that cannot be transcended. The culmination of this historical contingency
is expressed in Marx’s maxim - men produce their own history, but not
in the conditions of their choosing. Furthermore, precisely because histor-
ical finitude is meant to place contingency, the emerging notion of history
must necessarily be unidirectional. That is how Foucault can argue that
‘we witness the birth of this grand linear and homogeneous series which is
the process of production’ that introduces ‘the possibility of a continuous
historical time’ which Ricardo, for instance, thinks ‘in terms of a persis-
tently slower m i s of time, and eventually.. . a total suspension of history’
8PP.253-63 (emphasis added). As Prigogine showed, time reversibility is
what mainly characterised modern scientific enquiry.
If a quote could summarise the analytic of finitude within the mod-
ern episteme it would have to be the following: ‘What is essential is that
at the beginning of the nineteenth century a new arrangement of knowl-
162

edge was constituted, which accommodated simultaneously the historicity

of economics (in relations to the forms of production), the finitude of hu-
man existence (in relation to scarcity and labour), and fulfilment to an
end to History.. .History, anthropology, and the suspension of development
are all linked together in accordance with a figure that defines one of the
major networks of nineteenth century thought.. . Finitude, with its truth,
is posited in time; and time is therefore jinite’8pp.262-73 (emphasis added).
The developments within the other two fields studied by Foucault, biology
and philology, also follow similar patterns of linear historicism. Similarly,
Prigogine’s description of modern physics reflects similar phenomena.
If we apply Foucault’s definition of the modern organisation of knowl-
edge, arguments such as Popper’s and Kuhn’s make us realise that what we
often brand as ‘science’ is centred on presuppositions, or conjectures, that
should be regarded as a transcendental philosophy of the object, for the
subjects of scientific enquiry are usually examined by how they stand in re-
lation to the basic presuppositions of the paradigm, and not vice-versa. In
other words, the paradigm places contingency upon the object of research,
and - as Prigogine argues - this contingency is often elaborated in terms
of time reversibility. Kuhn demonstrates that those elements that do not
comply with paradigmatic assumptions are often d i ~ c a r d e d ~ pThe
. ~ ~study
.
of reality seems to follow from metaphysical assumptions. DuprC, in a book
with a very pertinent title ( T h e Disorder of Things), seeks to illustrate the
‘dependence of modern science on metaphysics’ and argues that ‘science
itself cannot progress without powerful assumptions about the world it is
trying to investigate. . .without.. . a prior metaphysics’ll. While witnessing
scientific enquiry in action, Latour also noted the power of paradigmatic
assumptions12, both at the level of the analytic of finitude and histori-
cal linearity. Although the argument cannot be articulated in any length
here, ‘scientific’ theoretical assumptions are not that different from the a
posteriori transcendental notions that Foucault argues cause the necessary
linearity of history.

3. Continental Philosophy and the Collapse of the Modern

Episteme
The limits of the modern episteme have however been challenged by philoso-
phers who were keen to conceptualise temporality not as an element of
contingency, but as the cause of perpetual and dynamic change. As stated
above, Foucault explains that as the era of representation withers, paving
 163

the way for the modern episteme, the realm of philosophy finds itself di-
vided into three distinct areas of enquiry: ‘The criticism - positivism -
metaphysics triangle of the object was constitutive of European thought
from the beginning of the nineteenth century to B e r g ~ o n ’ ~ PBergson?
,~~~.
Why Bergson? Are we to understand that Henri Bergson was the first
thinker to push Western philosophy beyond the modern organisation of
knowledge?
Henri Bergson is one of the most important figures within the spiri-
tualist tradition, a tradition that advocated the presence of a distinctive
philosophical experience. Bergson followed with a lot of interest the de-
velopments that were taking place within the philosophy of science. More
precisely, PoincarC’s initial doubts concerning the infallible objectivity of
science, and his disciple’s (Le Roy) reinforcement of these doubts, must have
played an essential role in the constitution of Bergson’s thought. At the
core of Bergson’s philosophy of science lies the conviction that the scientific
method adopts a cinematographical view of temporality, which implies, as
Gutting puts it, ‘that science views reality not as a continuous flux (the du-
ration that in fact is) but as a series of instantaneous ‘snapshots’ extracted
from this
Science’s cinematographical view of duration is due to the fact that it
is primarily concerned with action. As thought that is primarily concerned
with practice, science must abstract from that concrete reality that we
experience, in which temporality is not simply another form of space, but
a ‘wholly qualitative multiplicity, an absolute heterogeneity of elements
which pass into the ~ther’~~J’.~’. For Bergson, in the real continuum of
duration there are no distinct elements that precede or follow real points in
‘time’. In this context, it becomes meaningless to speak of an a priori or
an a posteriori: Bergson envisages a notion of temporality as a ‘continuous
flux of novelty in which nothing is ever fixed, complete, or separate. In
this flux, anything that we can say exists ‘now’ also incorporates into a
qualitative whole everything we can say is ‘past’, a whole that is itself
being incorporated into the new synthesis of the ‘future”13.
The distinction between the synthetic and the analytic disappears in
the flux of time, for it is precisely this continuous temporal vortex that r e p
resents the formation of things (their synthesis) and their intrinsic (their
essence - in Kantian, the ‘thing in itself’) reality. This is the main postu-
late of what has been referred to as Bergson’s ‘superior Empiricism’. And
again, it is precisely this refusal to deal with transcendentalisms that char-
acterises Bergson’s drive for an immanent reality that can be experienced
164

while refusing to be cut into bits and abstracted. Clearly, the emergence
of such ontology revolutionises the basis of the modern episteme. And it is
precisely by demolishing the modern episteme ’s understanding of all pos-
sible paths and conditions for the attainment of knowledge that Bergson
issues a challenge to Kant’s Copernican revolution in philosophy and in
science:
unlike the transcendental procedure of Kant, [Bergson’s philosophy]
does not refer to the conditions of all possible experience; rather, it is mov-
ing toward ‘the articulations of the real’ in which conditions are neither
general and abstract nor are they broader than the conditioned.. . Bergson
insists upon the need to provide a genesis of the human intellect”14.
Having established the fact that science fails to tackle the issue of real
temporality, Bergson argues that philosophy might have been expected to
occupy this empty ground. However, this was not to be. Modern philoso-
phy, Bergson argues, has not challenged the view of time as ‘nothing more
than a fourth spatial dimension, which could readily be viewed as having
no creative efficacy, as merely the vehicle for the automatic unrolling of a
nornologically determined s e q ~ e n c e ’ ~ ~ PThis
. ~ ’ . modern scientific vision of
time, as Prigogine maintains, is all but dead.
Now we can start to understand how Bergson goes beyond the mod-
ern episteme. His conceptualisation of temporality refuses the a priori / a
posteriori distinction upon which the modern organisation of knowledge is
based. His refusal of transcendentalism, coupled with his insistence on the
realm of immanence, has produced, amongst many, challenges for the mod-
ern notions of abstraction, temporality, empiricism, science and freedom.
Bergson lies at the heart of Deleuzian philosophy14, whose conceptualisa-
tion of time is highly relevant to the ethos of Complexity, as Manuel De
Landa has shown3. Prigogine makes of Bergson a cornerstone in his theo-
rizing on Complexity and quantum physics.
Another timely reminder of Complexity’s origins can be found in the
works of Eric Hobsbawm. Hobsbawm rightly asserts that the principles
of Complexity did not ‘appear’ but reappeared under the misleading tag
of ‘chaos theory’. The reappearance of such ideas was possible thanks to
the increasing calculating powers of computers. According to Hobsbawm
the reemergence of ‘Complexity’ has profound implications for the concept
of causality. Such an approach does have the potential to undermine sev-
eral ‘political, economic and social’ assurnption~l~. Crucially, Hobsbawm
identifies the genesis of Complexity in the truly European ‘epistemic civil
war’, characterised by two contrasting interpretations of the role of rea-
 165

son after Enlightenment. Such an account of contrasting ‘modernities’ is

also present in Negri’s and Hardt’s Empire“. The following quote is of
fundamental importance to critically assess the genesis of Complexity:
Developments within ‘chaos theory’ in the 1970s and 1980s are not with-
out commonalities with the emergence, in the beginning of the lgth century,
of a ‘romantic scientific school’, of which Germany was the primary focus
(Naturalphilosophie); that reacted to the dominant ‘classical’ science, es-
sentially represented by practices in France and in Great Britain. It can
be interestingly observed that two eminent pioneers of the new science
(Feigenbaum and Libchaber) were in fact inspired by Goethe’s - radically
anti-Newtonian - theory of colours, and also by Goethe’s treatise ‘On the
Transformation of Plants’, which can be considered as a precursor of anti-
Darwinian and anti-evolutionist theories17.
Hobsbawm retrieves this crucial information on Goethe’s fundamental
influence on the two precursors of Complexity in Gleick’s seminal book18.
The genesis of Complexity is therefore to be located in a moment of dis-
location between two opposing visions of ‘scientific’ knowledge, ultimately
based upon two different visions regarding the role of reason after the En-
lightenment. What we are referring to in here, in the guise of Complex-
ity, is the re-introduction of German romanticism in modern science. It
is tempting to interpret recent, Complexity-related developments within
the epistemology of science as the re-introduction of romanticism’s ethos,
should we say, ‘through the back door’. As such, the future of Complexity
should be seen in terms of the ‘longue duree’ as one of the possible paths
to knowledge opened by the European intellectual heritage.
It is in fact impossible not to consider the intrinsically German nature
of these events. As Hobsbawn puts it:
. . . it was in Germany, where classical liberalism was weak, that a rival
scientific ideology to the classical was most popular. This was Natural
Philosophy.. . it was speculative and intuitive. It sought expressions of the
world spirit, or life, as the mysterious organic union of all things with
each other, and good many other things which resisted precise quantitative
measurement of Cartesian claritylg.
More importantly, it was Natural Philosophy that provided the intellec-
tual bedrock to ‘Goethian Physics’, which in turn considerably influenced
the thought of thinkers such as Nietzsche, Husserl and Heideggerd. Cru-
cially these thinkers put an alternative conceptualisation of time a t the

dConsider, for example 2o

166

centre of their attempt to demolish modern metaphysics (consider, for in-

stance, Being in Time).
Heidegger related the triumph of mechanistic, metaphysical epistemol-
ogy to his contemporary political situation by correlating political events
to ‘the Forgetfulness of Being’ and to technology:
The wilful construction of a world empire to last four millennia shows
[historical linearity]. . .a preference for quantity over quality that is alien
to genuine creators like the Greeks. Empire-building stems not primarily
form dictators and authoritarian states, but from the metaphysical essence
of (emphasis added).
This ‘verdict’ on Nazism was ‘delivered in the summer of 1941, when
Hitler’s power was at its height’21. Similarly, when Europe was on the
verge of the Holocaust, Husserl considered that it was appropriate to relate
the socio-political environment of his Germany to the ‘Crisis of European
Science’22.
Nietzsche first individuated this fatal crisis. That great lover of Goethe’s
Natural Philosophy, Nietzsche, is the kernel that unites French post-
structuralisme and the intrinsically German experience of modern ‘ratio-
nalistic thought’. The crisis, as Antimo Negri argues in an important study
of Nietzsche’s ideas on the implications of ‘classical’ science and the im-
pact of such ideas on German philosophy as a whole23; is the belief that
science’s ‘modern’ purpose is to provide metaphysical certainty. Drawing
from this idea, Negri also examines how Heidegger argues that “the funda-
mental nature of contemporary human existence, which seeks for certainty
and security everywhere, is being consolidated under the powers of the
convictions professed by the principles of reason”, and ironically, Heideg-
ger states that “Leibniz, who discovered the principles of sufficient reason,
is also the inventor of life insurance”. But what kind of certainty can
‘open-ended’ scientific research, always marching towards the - ultimately
unreachable - goal of rediscovering the why and how of all things; bring to
human existence? Heidegger fears that such impossible research may end
up ‘exhausting man’. Here, we are talking about a fear that can only be
dissipated if men, including men of science, do not suffer from the exhaus-
tion of starting always again from the beginning. It can only be dissipated
if man has the courage to start again there where he though he had fin-
ished: Husserl maintains that the future of European culture depends on
this??23p.26 (my translation).

eThe influence of Nietzsche on Foucault, Derrida and Deleuze is undeniable. See l3

 167

The quest for certainty and securityf is ultimately the biggest danger
facing European civilisation. It would be clear, following Foucault’s analy-
sis of the modern episteme, that transcendental philosophies of the object
- metaphysics - necessitate two crucial elements in order to function: a

concept of analytic finitude and a concept of linear temporality. Heideg-

ger’s conviction according to which the metaphysical nature of his time is
what forced a strict philosophy of the limit and of historical linearity to
take place is entirely compatible with Foucault’s description of the mod-
ern episteme. Metaphysics is necessarily a philosophy of the limit - the
contingency which in scientific epistemology translates into paradigmatic
thought.
The necessary securing of limits is entirely encapsulated in Leibniz’s
thought: ‘if one is to establish the elements of human knowledge some fixed
point is required, on which we can safely rest and from which we can set
out without fear24P.13.It is with this quote in mind that Nietzsche exhorted
men of science to build their house on the Vesuvius - only through the
acceptance of temporality as the principal creative and destructive force in
nature will modern human beings ‘always start again from the beginning’,
without fear.

4. Complexity and the Modern Episteme

There are at least two ways in which the scientific developments elaborated
by Prigogine, a flagship of the loosely defined ‘Complexity Science’, can
be broadly understood within a Bergsonian framework. Firstly, Prigogine
quickly acknowledges the cinematographical nature of pre-complexity sci-
ence and secondly, he indicates how this precise view is caused by the notion
of linear.
Prigogine articulates the idea that contemporary science uses a static
notion of time principally by making reference to Einstein and Hawking.
Einstein famously declared that ‘time is an illusion’. More recently, “in his
Brief History of Time, Hawking introduces ‘imaginary time’ to eliminate
any distinction between space and time”gp.58. Thus Classical and Modern
physics have maintained a spatialised vision of time that is basically static,
and, even today, scientists assume as a ‘matter of faith that as far as the
fundamental description of nature is concerned, there is no arrow of time’g.
Prigogine argues that ‘Nature involves both time-reversible and time-

‘For a comprehensive argument, see 24

168

irreversible p r o c e s ~ e s ’ ~ pReversible
.~~. processes deny temporality as a con-
stitutive apparatus of the process. Examples of these processes can be
found in Newton’s formulation of classical physics and in Schrodinger’s
basic equation of quantum mechanics. In both cases, equations are invari-
ant with respect to time inversion. Contrarily, time irreversible processes
break time symmetry. In these processes temporality does affect how the
general rules of motion will impact the system in a precise temporal con-
text. More importantly, time irreversibility produces entropy. An example
of time irreversible processes is the second law of thermodynamics. How-
ever, Prigogine argues that time reversibility is produced firstly because we
accept to reduce the analysis to an elementary level, and secondly because
we abstract: ‘Reversible processes correspond to idealizations: We have
to ignore friction to make the pendulum work r e ~ e r s i b l y ’ ~In. Foucault’s
words, these metaphysical transcendentalisms adopt reductionism because
they ‘must deploy the deductive forms only in fragments and in strictly
localised regions’. Once Prigogine dismisses the idea that entropy might be
caused by insufficient data or faulty examination, the ideas that follow from
his arguments suggest that, if we bear Foucault in mind, time reversibility
is a particular cause of a transcendental philosophy of objects, that is, a
metaphysical system that ignores elements which do not happen to coexist
with the basic premises of a paradigm. This causes the need to discard
incompatible elements (precisely such as the second law) on the grounds of
humanity’s imperfect observation capacities or on the inadequacy of its in-
struments (in other words, limits). However, according to Prigogine: ‘The
results presented thus far show that the attempts to trivialize thermody-
namics.. .are necessarily doomed to failure. T h e arrow of t i m e plays a n
essential role in the f o r m a t i o n of structures in both the physical sciences
and b a ~ l o g y ’ ~ *(emphasis
.~~ added).
Time irreversibility becomes undeniable once, on the one had, we adopt
a more i m m a n e n t approach to nature, and on the other, we look at popu-
lations and not at single elements that compose them. The apparently con-
tradictory pulls towards immanence and connectivity are resolved through
the continuum of time and this notion of empiricismg.
Prigogine is happy to demonstrate that the results of his own research
concord with Bergson’s conceptualisation of temporality:
I’m certainly not the first one to have felt that the spatialization of time

g For a more detailed account of Prigogine’s empiricism and its significance for Complex-
ity’s ethos see 25
 169

is incompatible with both the evolving universe, which we observe around

us, and our own human experience. This was the starting point for the
French philosopher Henri Bergson, for whom ‘time is invention or nothing
at all’. . . I mentioned one of Bergson’s later articles, “The Possible and
the Real”. . . where he expressed his feeling that human existence consists
of ‘the continual creation of unpredictable novelty’, concluding that time
proves that there is indetermination in n a t ~ r e ’ ~ P . ~ ~ .
Contemporary cognitive research is equally finding an inspiration in
the thought of Bergson. For example, the dynamical hypothesis of experi-
ence elaborated by Robbins relies heavily on Bergson’s concept of contin-
uous time. As Robbins explains, ‘the dynamical hypothesis can be placed
within the framework created with the theories of Henri Bergson.. . From
this emerges a basis for supporting experience, semantics and the funda-
mental intuition behind representations - representative thought’26.
However Robbins goes beyond Bergson, his work has approached Hei-
degger’s theories of being in time and related these insights to the concept of
emergence inherent in Complexity scienceh. Robbins takes note of Heideg-
ger’s thought, which contributed in questioning ‘the distinction between
a conscious, reflective, knowing ‘subject’ and a separate ‘ o b j e ~ t ” ~ ~ p . ~ ~ ~
Heidegger does indeed see ‘representations as a derivative phenomenon,
which occurs only when there is a breaking down of concernful action.
Knowledge lies in the being that situates us in the world, not in reflective
representation’26.
Moreover, it is equally important to note that Heidegger was inspired
by Bergson, and to realise that the essential insight on the role of time is
what links these authors and turns the modern episteme on its head:
Heidegger was certainly aware of Bergson. Cassirer was straightforward,
noting, “it is the lasting achievement of the Bergsonian metaphysic that it
reversed the ontological relation assumed between being and time”. The
relationship of subject and object in terms of time constitutes the funda-
mental framework within which ‘situatedness’ truly lies.26Pp.320
In conclusion, Robbins locates the struggle between different theories of
experience in a confusion over real and abstract space and abstract time,
he can thus affirm that ‘Bergson’s theory.. . can explain the reconstruction
of the image of a past experience via an event occurring in the environ-
ment; it can at least logically support the phenomenal experience of the
memory image26P.337’,and can finally conclude by stating that Bergson

hSee references to Gleick in 26

170

offers a viable theory of conscious perception, a basis without which any

theory of cognition and memory has hitherto only been tentative. It truly
acknowledges time. In fact, with respect to the problems of a scale of time,
it raised questions 100 years ago that still have not dawned upon repre-
sentationalism. It relies, integrally, upon dynamics, yet makes clear why
dynamics must enlarge its vision of time before it can support a theory of
,ind.26p.333
Thus, he who Foucault describes as the end point of the modern epis-
teme, precisely by virtue of his opposition to the two axis of modernity -
namely a varying analytic of finitude and linear historicism - has become
an inspiration for truly ‘post-modern’ research (in the Foucauldian sense)
at the cutting edge of contemporary physics and cognitive science

5. Conclusion
This paper sought to present Foucault’s ‘Archaeology’as an adequate con-
ceptual means for charting the origins of Complexity. Following Foucault,
the paper has 1) presented the emergence of linear temporality as an es-
sential feature of the modern episteme, specifically related to birth of tran-
scendental philosophies of the object (modern metaphysics); 2) indicated
how figures present in the European philosophical tradition challenged the
basic understanding of time inherent in the modern episteme, making spe-
cial references to Bergson and Heidegger and 3) outlined how these figures,
precisely through their alternative conceptualisation of time, are becoming
increasingly influential in Complexity-related research. There is no doubt
that the question of how time should be conceptualised is becoming increas-
ingly topical in contemporary scientific research. Moreover, such questions
directly tackle the issue of how such conceptualisation of time as an ir-
reversible feature of dynamical (as opposed to static) processes should be
understood. It appears that an accurate understanding of change - more
precisely irreversible change - has become a primary objective in many
research agendas. The acceptance of such irreversibility (or, as Prigogine
puts it, the ‘End of Certainty’) radically undermines the nature of meta-
physical knowledge, and clearly constitutes one of the pillars of Complexity
Science. The message inherent to Complexity’s ethos is beautifully encap
sulated in the simple concluding remark of Etienne Klein’s exploration of
the philosophy of time in contemporary physics: ‘We must learn to love
irre~ersibility’~~.
 171

References
1. G. MorCoI, ”What Is Complexity Science? Postmodernist or Postpositivist?,”
Emergence 3, no. 1 (2002).
2. P. Cilliers, Complexity and Postmodernism: Understanding Complex Systems
(London ; New York: Routledge, 1998).
3. M. De Landa, Intensive Science and Virtual Philosophy, Transversals. (Lon-
don ; New York: Continuum, 2002)
4. W. Rasch and C. Wolfe, eds., Observing Complexity: Systems Theory and the
Postmodern Episteme (Minneapolis: University of Minnesota press, 2000).
5. D. S. Byme, Complexity Theory and the Social Sciences : A n Introduction
(London: Routledge, 1998).
6. Thomas Kuhn, The Structure of Scientific Revolutions (Chicago: University
of Chicago Press, 1970).
7. M. Dillon, ” Poststructuralism, Complexity and Poetics,” Theory, Culture €4
Society 17, no. 5 (2000).
8. Michel Foucault, The Order of Things: An Archaeology of the Human Sci-
ences, World of Man (London: Routledge, 1989).
9. Ilya Prigogine and Isabelle Stengers, The End of Certainty: Time, Chaos,
and the New Laws of Nature, 1st Free Press ed. (New York: Free Press, 1997).
10. Gary Gutting, Michel Foucault’s Archaeology of Scientific Reason, Modern
European Philosophy. (Cambridge England ; New York: Cambridge Univer-
sity Press, 1989) 184.
11. John Dupr6, The Disorder of Things: Metaphysical Foundations of the Dis-
unity of Science (London: Harvard University Press, 1995) 1-2.
12. B. Latour, Science i n Action : How to Follow Scientists and Engineers
through Society (Milton Keynes: Open University Press, 1987).
13. Gary Gutting, French Philosophy in the Twentieth Century (Ca,mbridge,
U.K. ; New York: Cambridge University Press, 2001) 51.
14. K. A. Pearson, Philosophy and the Adventure of the Virtual. Bergson and the
Time of Life (London: Routledge, 2002) 11-13.
15. E. J . Hobsbawm, Age of Extremes : The Short Twentieth Century, 1914-1991
(London: Abacus, 1995).
16. Michael Hardt and Antonio Negri, Empire (Cambridge, Mass. ; London:
Harvard University Press, 2000).
17. Hobsbawm, E. L ’age des extremes (Bruxelles: Editions Complexes, 1999),
756 - my translation.
18. J. Gleick, Chaos (London: Vintage, 1998).
19. E. J . Hobsbawm, The Age of Revolution : Europe, 1789-1848 (London: Aba-
cus, 1997) 335.
20. Harold Bloom, The Anxiety of Influence : A Theory of Poetry, 2nd ed. (New
York; Oxford: Oxford University Press, 1997) 50.
21. M. Inwood, Heidegger: A Very Short Introduction (Oxford: Oxford Univer-
sity Press, 2000) 6.
22. Edmund Husserl, The Crisis of European Sciences and Transcendental Phe-
nomenology; an Introduction to Phenomenological Philosophy, Northwestern
172

University Studies i n Phenomenology B Existential Philosophy (Evanston,:

Northwestern University Press, 1970).
23. Antimo Negri, Nietzsche: La Scienza Sul Vesuvio, 1. ed. (Roma: Laterza,
1994).
24. Michael Dillon, Politics of Security : Towards a Political Philosophy of Con-
tinental Thought (London ; New York: Routledge, 1996).
25. D. Popolo, "French Philosophy, Complexity, and Scientific Epistemology:
Moving Beyond the Modern "Episteme"," Emergence 5, no. 1 (2003).
26. S. Robbins, "Semantics Experience and Time," Cognitive Systems Research
3 (2002): 304.
27. Klein, E. Les Tactiques d e Chronos, Champs : Flammirion 2004.
 METAPHORS AND METHOD: EPISTEMOLOGICAL
CONSIDERATIONS ON COMPLEXITY SCIENCE

RODRIGO M. Z E I D A N ~AND MARIAG . DERENGOWSKI F O N S E C A ~

‘Escola de GestGo e Negdcios , UNIGRANRIO
Rua da Lapa, 86, Centro, Rio de Janeiro-RJ, CEP: 20021-180
Email: rodrigo-teidan@hotmail.com
Instituto de Economia, Universidade Federal do Rio de Janeiro
Av. Pasteur 250, Urca, 22290-240, Rio de Janeiro-RJ
Email: mderengowski@gmail.com

Any scientific development entails many epistemological considerations. In this

paper we consider two stages of scientific development - the metaphorical and
empirical levels. When applied to transposition of complexity science t o other
disciplines, in special economics, we conclude that complexity is still, in many
aspects, at the metaphorical level. This implies that transposed concepts should
be scrutinized to verify its validity towards the development of complexity science.

1. Introduction
A presentation at a complexity conference sometime in the early 2000’s:
its main purpose is to relate the philosophical implications of the fractal
gravitational singularity concept. Fractal gravitational singularity? What
does it mean? A gravitational singularity is a precisely defined concept
(such as a point of infinite space-time curvature) and a fractal is also a
clearly defined concept (a fractal is a geometric object which is rough or
irregular on all scales of length)”. However a fractal singularity makes no
sense from a scientific point of view, not even as a metaphor.
This kind of misinterpretation has been the subject of popular science
debate, and it shows how easily the scientific method can be ignored when
multidisciplinary conceptual transpositionb occur. The question is then:

aMost concepts relating t o complexity theory are rigorously defined in physics or biology
bTransposition is here defined as a metaphor for the process of spreading scientific con-
cepts and models from one discipline into another. Carruthers’ calls the recipient of
transposing concepts emerging sciences, but this may cause confusion because of the
diffusion of the term emergence.

173
174

does complexity theory has a sound epistemology? Although the answer

should be trivial, when transposing complexity theory to other areas such
as economics, or sociology, there are many epistemological implications
regarding the usage of the scientific method.
Carruthers' summarizes the nature of modern science in three main
questions, shown below.

Table 1. The nature of modern science - discussion points2.

How is science done? Dynamics of scientific Psychology of scien-
ideas tific creativity.
By whom? Consensus as the ref- Science as truth
erence frame
Where? Advocacy Science as process
Who pays? The role of experi- Peer opinion
ment
Disciplinary The yearning for The 'anxiety of influ-
structure truth ence'
Decision making The emergence of The discipline of
dogma quantitative science
Science policy New ideas and their Synthesis-its hazards
fate
The social context The role of personal Big science as the
contact emergence of an exec-
utive class
Big science/little sci- The role of tribalism The institutional con-
ence I I text
The tendency of spe- Pressure to confor-
cialization mity
The yearning for gen- Is anybody listening?
erality and synthesis
The powerful influ- The significance of
ence of aesthetics heroes
Competition

This paper is related to the dynamics of scientific ideas and is an ex-

tension from Zeidan and Fonseca2, where it was argued that, regarding the
transposition of complexity into economics, nothing prevents complexity
 175

studies from trying to incorporate the orthodox economic theory language

while developing new models and insights into economics processes. Again
economics is used as a case study for more general analysis, in this case
as a recipient of transposing concepts from original complex adaptative
disciplines such as biology.
What we propose here is to analyze the two epistemological levels of
scientific evolution of multidisciplinar complexity theory, the first one a
metaphorical level and the second an empirical level, both related to the
dynamics of scientific ideas and the scientific method.

2. The Metaphorical Level

Complexity as a metaphor represents many powerful insights for many dis-
ciplines. Carrut hers' groups emergence disciplines as computation, evolu-
tion, cognition, cosmology, particle physics and mathematics, biology and
the new medicine, economics, political systems, and cultural dynamics.
Those areas would be where complexity would thrive. For instance,
that is true of economic processes, where the idea of complex adaptative
systems to economic interactions is a very persuasive one [e.g. Refs. 3,
4,51. Moreover, there are many studies where biological agent models are
transposed to economic agent models [e.g. Refs. 6, 71. Complex agent-
based dynamics models the world's behavior through the use of "agents".
In this context, an agent is a computer program that acts according to a set
of rules, e.g., in a self-interested manner as it deals with numerous other
agents inside a specific economic environment - this arrangement can be
set-up to mimic almost any interactive system, including consumer theory.
These developments are then the result of the fact that economics sys-
tems can be analyzed as complex adaptative systems, since there are many
non-linear dynamics, feedback processes, hierarchies etc. More important,
economics systems are agent-based, and thus regular complex adaptative
systems originating in biology are easily transposed into economics complex
adaptative systems.
The metaphorical level is then the level where ideas and concepts from
complex adaptative systems cross over from discipline to discipline. This
leads to many new insights, and incorporating complexity concepts leads to
many adaptations to fit these concepts into the specific context of the disci-
pline. It seems that there is where the fractal singularity concept emerged,
as a new concept for transposing ideas into another area. However, this is a
crucial part of the scientific development of crossing-over models, since this
176

is where it runs into scientific terminology being used without the slightest
knowledge of its real meanings.
The metaphorical conceptual nonsense created in transposing complex-
ity into other disciplines has a powerful justification in Kuhn's work - much
cited in philosophical complexity works [e.g. Refs. 9, 10, and probably
a good percentage of philosophical complexity papers presented at confer-
ences around the world]. The biggest problem with Kuhn's view is that, up
to the point of the paradigm shift, proponents of the older and newer theo-
ries cannot be speaking the same scientific language, and, in fact, alienate
each other from their respective views of the world. This problem perme-
ates complexity studies in economics, where most economists cannot begin
to cope with the new theory because of its inherent incommensurability2.
However, Kuhn's paradigm shift and incommensurability in no way im-
plies an anything goes relativistic view of the dynamics of science. New
paradigmatic science still would have to have consensus as the reference
frame, and internal coherence. This would be derived from new concepts
and ideas, and new meanings of old ones - but those new meanings would
still have to make sense inside a reference frame.
What is happening is that to deconstruct what is now referred as the old
Cartesian world-viewc internal coherence in many philosophical complexity
works have been lost, even in the metaphorical level.
The metaphorical level is inherently theoretical and can lead to en-
demic insights in the process of transposition. For instance, Goldsteing
proposes a self-transcending construction of emergence new formalism that
is a very powerful insight and is born out of philosophical theory, even if
relating some discussion to biology theoretical work. The main problem
with the metaphorical level in the process of the dynamics of complexity
theory is that its lack of rigor can lead to the hubbubs and nonsensical
notions. Moreover, many implication are derived from it without any sem-
blance of a rigorous demonstration - for instance, complexity was used as
tool for raising ethical Issues Concerning the Use of Animals in Scientific
Experimentation".
The metaphorical level is where many epistemological problems can
arise. The fact that it is a speculative stage on the dynamics of scien-
tific nature means that many propositions, implications and conjectures

=This view of cartesianism as anathema to complexity is permeating many works in

complexity philosophy, even if proper definition of cartesianism seems to be lost to. It
seems that cartesianism and newtonian science means only non-complexity science.
 177

will be raised when transposing concepts and notions to other disciplines.

By the very loose nature of this speculation many nonsensical ideas will
Aoat around, without the necessary justification and demonstration of its
importance. Moreover, without rigorous screening many implications will
leak through the next stage, and thus will slow the dynamics of the scientific
development of complexity.

3. Empirical Level - The Method

This is the level where applications of the theory occur. Not only applica-
tion in terms of objective results, but developments posterior to the consoli-
dation of metaphorical concepts. It involves the yearning for generality and
synthesis, and the development, in natural sciences, of quantitative testable
results. In social sciences like economics, it involves models that lead to
policy prescription and/or better comprehension of economic phenomena.
Supposedly at this level the transposition would have implemented con-
sensus as the reference frame and new dogma would emerge. However, at
this stage in complexity theory there is still search for the new dogma,
although many concepts have spread out as reference frame. But to this
day the lack of the institutional context permeates many transpositions of
complexity into social science disciplines, and hence Zeidan and Fonseca2
showed how this lack of institutional context leads to developments that
disregard almost completely the present orthodoxy in economics.
The allure of the transposition process is the fact that in the originat-
ing disciplines, like biology, complexity has yielded consistent results, to
the point of a paradigmatic shift in evolutionary theory, for instance. The
role of experiment is then much more important, and formalization through
models has to lead to good empirical results. At this point in the complex-
ity social sciences, however, empirical research has not yielded good results,
and thus complexity is stuck in a metaphorical state, when regarding many
discipline transposition by this point. To take an example, in consumer the-
ory, the similarities between the agent-based model7 and regular orthodox
consumer theory is presented in table 22.
The table represents the metaphorical transposition of an agent-based
model to answer the same questions as orthodox consumer theory. On the
empirical level, however, agent-based models are resolved through simula-
tions. In current state, however, simulations have not yielded consistent
results, and thus agent-based models can not offer better results than reg-
ular orthodox theory - yet. Santa Fe instate, for instance, is most known
178

Table 2. Metaphor and Method in Complexity Economics.

Orthodox Theory Heterodox Theorv’
Two bundles of goods (A and B) Individual agents can be repre-
are related through preference: sented by a string such as: A =
A(B, AiB, or A B (indifferent) 3 x is either 0,1, or *
~ 1 x 2...~xn.
(indifferent)
For any group of bundles an in- From a number of strings a
difference curve is derived. schemata is developed.
For any indifference curve there is For any schema H there is a fit-
a utility function f(x) associated ness function f(x) associated with
with it. it.
~

There is a probability p associ- There is a probability p associ-

ated with the selection of a bun- ated with the selection of a string.
dle.
The selection is done through The selection is done through sat-
maximization. isficing.

for using computer simulations to study complex systems, including envi-

ronmental, economics, and social systems. Some results are encouraging,
but most are still tentative, with no replacement of orthodox theory - again
yet.

4. The Role of Metaphor

When dividing the epistemology of science in two dividing lines, it should
clear the role of each line and its connection to the scietific development
process. While in the empirical level results can be adequately verified, and
thus it is where science has its best screening process, the metaphorical level
is mainly speculative and is not verifiable per si, implying a non-screening
process and a loose part of the scientific development process.
The role of metaphor resides in the insights and theoretical inquiries
that lead to better theories. It was in metaphor that complexity science
reached a multitude of scientific areas - in economics, for instance, the idea
of complex adaptative systems is very attractive to deal with economic
processes. The main problem is translating this inquiries into models and
methods for working out sustainable results. Although there is no adequate
test for metaphors as in models, the role of metaphors can be analyzed by
reverting to its basic core language. In this respect Kuhnian’s ideas of in-
 179

commensurability are an important part of how the evolution of scientific

knowledge takes place in the metaphor stage. In that respect, metaphor
occurs as part of the paradigm shift that leads to new emerging theories.
In that process the language is fluid to the extent that concepts are not
interchangeable between competing theories. This can lead to a relativis-
tic point of view where fractal singularities are acceptable terms for new
theories.
This interpretation of Kuhn’s point of view is very pervasive in social
sciences circles dealing with complexity. Kuhn is often cited as the basis
for new concepts as fractal singularities. However, Kuhn’a analysis is not
a relativistic analysis and it leads to scientific evolution. Extreme views of
Kuhn’s incommensurability concept seems as wrong as fractal singularities.
In this way the role of metaphor has to strike a balance between novelty
concept and transpositions and scientific background that will eventually
lead to model. How to strike this balance depends on the scientific com-
munity, and can lead to better thory or even to the burial of valid theories.
But the main point is that metaphors should be constrained by its com-
mensurability, not with the theories that it is trying to replace, but with
itself. Metaphor should have internal consistency, and it seems that in
much of transposition of complexity into other scientific are= this internal
consistency is being lost.

5. Final Comments
Separating the two levels of the scientific development of complexity can be
most useful to verify how transposition of complexity into other disciplines
is evolving. Moreover, it leads to better comprehension of how concepts
and ideas are being incorporated into other disciplines, with better chance
to envision the hubbubs and nonsense being created in the process. In-
stead of a pure Kuhnian approach, where incommensurability is the norm,
by having a metaphorical and an empirical level, some commensurability
is possible, especially when regarding inter-disciplinary commensurability
- i.e. ensuring that transposed concepts and ideas are compatible with its
origin, and also make sense with the recipient discipline, and, more impor-
tantly, internal commensurability. Not doing so leads to fractal singularities
and even more outrageous concepts.
180

References
1. Carruthers, P. (1988). Emerging syntheses in modern science. European Jour-
nal of Physics, 9: 110-116.
2. Zeidan, R. M., Fonseca, M. G. D. (2004) Epistemological considerations
on agent-based models in evolutionary consumer choice theory. Emergence:
Complexity and Organization. Boston, MA: , v.6, n.3, p.32 - 39.
3. Arthur, W. B. (1995) Complexity in Economic and Financial Markets, Com-
plexity, v.1, n.1, April.
4. Anderson P.W., Arrow K.J., Pines D. eds. (1988) The economy as an evolving
complex system; Addison-Wesley Pub.Co, Reading Ma.
5. Cohendet P., Llerena P, Stahn H., Umbauer G. eds. (1998) The Economics
of Networks, Interactions and Behaviours Springer, Berlin.
6. Holland, J. H. & Miller J. H. (1991) Artificial Adaptive Agents in Economic
Theory, American Economic Review, May, pp. 365-370
7. Goldberg, D. E. (1989) Genetic Algorithms: in Search Optimization and
Machine Learning, Reading, MA, Addison-Wesley.
8. Sokal, A. & Briemont, J. (1998) Intellectual Impostors. London: Profile
Books.
9. Goldstein, J. (2005). Impetus without Drive or Teleology: The Self-
transcending Construction of Emergence. paper presented at the Complexity,
Science and Society Conference. University of Liverpool.
10. Waldrop, M. Mitchell. (1992) Complexity: The Emerging Science at the Edge
of Order and Chaos. New York: Simon & Schuster.
11. Robinson, F. (2005) The Relevance of Complexity Science to the Ethical
Issues Concerning the Use of Animals in Scientific Experimentation: the
Matter of Constraints. Paper presented at the CSS Conference - Liverpool.
SOME PROBLEMS FOR AN ONTOLOGY OF COMPLEXITY

MICHAEL McGUIRE
Department of Applied Social Sciences
London Metropolitan University

Ontology, like the broad philosophical tradition of metaphysics of which

it is part, stands ‘beyond physics’ since its propositions are not exclusively
subject to empirical refutations. But neither is each discipline fully au-
tonomous from the other. Where possible, ontology ought to avoid direct
contradiction with physical evidence. Similarly, many fundamental assump-
tions and justifications within science have a basis that cannot be estab-
lished by the experimental method. These notoriously include standard
principles of scientific methodology such as the faith in inductive reason-
ing, indispensability arguments, the value of simplicity, and of course the
range of entities to which is ontologically committed. Whilst it is con-
testable as to whether the range of mathematical techniques and physical
observations that are associated with ‘complexity theory’ (henceforth ‘CT’)
really form a distinct branch of science in the way say, that quantum me-
chanics does, they must certainly aspire to the status of scientific theories.
And, whether one is willing to engage with this fact or not, such aspirations
entail a set of underlying ontological commitments. Complexity theorists
can of course proceed to work in an instrumental fashion, without bothering
to interrogate their foundational ontological assumptions. But if such as-
sumptions turn out to be incoherent, then C T itself may ultimately inherit
those problems. In this paper I intend to press this question and to indicate
some potential problems for CT when it is wedded to certain traditional
ontologies - in particular those involving the substance/particular and uni-
versal/property dichotomies. A highly plausible alternative - an ontology
of patterns a also has its own set of problems. If the difficulties of this more
radical ontology are also insuperable then one possible conclusion about the
apparent lack of theoretical unity to CT is precisely because it rests upon

aFor which I have pressed the case in my [l,21

181
182

suspect foundations. In what follows then I will outline some reasons for
thinking that standard ontologies cannot meet the muster, before moving
on to outline a few of the questions for the alternative that I advocate.

1. Ontology
Two (broad) senses of ontology are operative in contemporary analytic
philosophy.
(i) ‘Classical’or philosophical ontology - what Aristotle called the “sci-
ence of being qua being’’.b The approach deals with questions
about objects, existence, properties] relations, parts, boundaries]
measures, causality, possibility and so on. The development of for-
mal techniques in ontology has increasingly enabled such questions
to be addressed in more rigorous terms rather than by what D.C.
Williams termed ‘speculative cosmology’ [3]
(ii) ‘Technical’ Ontology. An approach that has roots in formal ontol-
ogy but is mostly applicable to developments in computer science,
particularly artificial intelligence (AI) research. Here the focus is
less upon real world objects, but upon ways of specifying what is
required to exist for some conceptualisation to function correctly.
In A1 for example knowledge based systems need to be able to rep-
resent the range of assumptions they deploy to operate. Specifying
an ontology will help the programmer attain this as well provid-
ing a shared vocabulary that can operate across other systems in a
coherent and consistent manner‘
In this paper it is the former approach that I will draw upon for I intend
to ask what is the most general kind of thing that theories of complexity
could take to act as the ontic vehicle for the phenomena they purport to
identify. Given the discontinuities in content between such theories this
will be, by necessity, an exercise in what Strawson calls ‘descriptive’rather
than ‘revisionary’ metaphysics [4]. But if the ontological commitments
of CT could be given some satisfactory metaphysical grounding, then the
radical nature of some of its propositions may in turn suggest that serious
revision to the traditional cast of metaphysical candidates ontologists are
familiar with may be necessary.

bAristotle uses this term in Metaphysics (Books, G, E & K )

=See Tom Gruber’s useful discussion of this kind of ontology at http://www-
ksl.stanford.edu/kst/what-is-an-ontology.html
 183

2. What should ontology do for complexity theory?

Whilst it cannot directly arbitrate upon empirical questions, a coherent
ontology ought to offer a variety of clarifications and conceptual syntheses
for CT. Insights into causal relations in complex processes; specifications of
the conditions for ‘self-organisation’; a better understanding of part-whole
relations, in particular the thorny question of emergence, and so on. Above
all, it should offer a taxonomy of the different classes of entities posited
by C T for its laws and generalisations to function consistently and effec-
tively. As a branch of metaphysics, ontology is permitted to organise the
natural kinds of science into broader more general classes in order to dis-
cover otherwise unspecifiable relations between them. Indeed, it can ask
not just how many different kinds of entity there are, it can also put this
question foundationally - how many basic kinds of entities there are.d Since
philosophical predilections, like those in science, are for simplicity it is usu-
ally held to be desirable to be sparse with the number of such basic kinds,
and to permit only those kinds thought to be explanatorily necessary any
kind of status. Of course natural science (or at least that branch prac-
tised by physicists) has been marked from its inception by the prejudice
that just one set of fundamental ontological kinds - those located at the
atomic level - suffice to support its propositions. There has never been a
satisfactory rationale provided for this prejudice and with the discovery of
subatomic classes of entity, it is one that has had to be seriously qualified.
Whilst reductionism remains the unstated prejudice for most scientists, the
faith in a unique level of kinds to which all other phenomena reduces ex-
haustively has become increasingly less plausible. Whilst there may be
supemeniencee between explanatory levels, or a nomological “marching in
step” many properties in the world such as those involving biological or
mental phenomena do not appear to be exhaustively reducible (i.e. expli-
cable) in terms of some unique fundamental level. Philosophers of science
such as Nancy Cartwright now argue strongly against the idea of a seam-
less hierarchy of reducing levels, advocating instead a ‘patchwork’ of laws,
of discontinuities that undermine the traditional deference to fixed nomo-

dIt is worth noting that a categorisation of basic metaphysical kinds does not rule out or
reduce away large scale, or ‘common sized’ objects. It merely suggests that they might
also be subject to unification at deeper levels of abstraction
eLoosely, the idea of supervenience holds that all non physical phenomena is realised
by physical phenomena without reducing to it. More precisely, a set of properties P1
supervenes on a set of properties P2 if and only if there can be no changes or differences
in P1 without there being changes or differences in P2. See [5, 61
184

logical reducibility (see her [7, 81). Such views sit well with many of the
underlying assumptions of theories of complexity. But how does this cash
out in ontological terms? Are the ideas of holism and emergence, of ‘infi-
nite’ or irreducible ontological kinds it hints at as equally questionable as
the commitments of the scientific reductionist?

3. What ought complexity theory to ask of ontology?

Complexity theory contains a number of mysteries that have tended to make
more orthodox scientists uncomfortable with the substance of many of its
claims. Do chemical reactions such as the Belousov-Zhabotinsky reagent
really indicate a kind of self determination aspect at the sub-organic level?
Can attractors be regarded as ‘genuine objects’ in the way electrons of-
ten are, or are they convenient theoretical posits that do no more than
to unify tendencies within dynamical systems? Is the apparent ubiquity
in nature of power laws like Zipf’s law something that is ontologically sig-
nificant, or is it no more than a ragbag of mathematical coincidences and
pseudo-continuities? Indeed is there even really any such thing as a ‘com-
plexity science’ or must we accept, as some critics argue that “there is no
such theory, and it even may be that there are no such structures as they
are implied to exist by Prigogine, Haken, and their collaborators.”f Of the
many questions of this kind surely one of the most mysterious for any pu-
tative ontology of complexity is how it ought to represent the relationship
between order and chaos. C T would seem to have an urgent requirement
for an adequate set of conceptions here, not least because of one of its
governing assumptions - that many important structures arise at or near
some kind of ‘boundary’ between order and chaos. But not only is there
more than one conception of order which is used, such concepts tend to
be formulated it in negative terms - as that which is non-random or fore-
grounded from the ambiently chaotic. A variety of further mysteries then
arise; how do ‘ordered’ objects differ from ‘random’ objects? Can there be
a random object? Can there be absolute randomness? Why should order
be so fundamental? And so on. This central dichotomy between ordered
and ‘disordered’ phenomena appears to require some urgent conceptual re-
finements. Ontology ought to provide us with at least some kind of insights
here. But when we come to examine how order/disorder distinctions might
be handled within traditional ontologies, we find all of them wanting in

f[9, p. 4471
 185

some respect.

4. Some ‘ontologies’ of complexity

Since the ontology of complexity has not been widely discussed the meta-
physical preferences of complexity theorists are not immediately obvious.
But it does not take much reflection to see where the predominant sympa-
thies lie. Thus, we hear of; “generic low-energy properties determined by
a higher organizing principle and nothing else” [lo]; or how “the validity
of a kinetic equation” can imply “ergodic properties” [ll, p. 431; or of
“collection(s) of molecules which have the property that the last step in
the formation of each molecule is catalyzed by some molecule in the sys-
tem” [12, p. 2851. Tacit appeals by complexity scientists to the category of
properties, together with their close association with the idea of emergence,
immediately suggest that the traditional philosophical distinction between
particulars and properties (or its close cousin substances and universals) is
at work here. So what more precisely is a property and how does it relate
to the category of particular?
Properties have been conceptualised in various ways. One idea has
them as sets - specifically sets of particulars, though this idea quickly runs
into technical problems which are not relevant here [13, p.211. A more
natural view sees them in terms of things which are ‘had’ by a particular,
features or characteristics such as ‘redness’ ‘roundness’ and so on. But
there is a further, important ingredient of the distinction between properties
and particulars which relates to how these objects can be distributed in
space and time. Properties can be repeated, particulars cannot. Thus, one
approach to an analysis of the distinction might focus upon location and
repeatability as follows:

A particular P is an object which has a unique location. That is,

it can exist at only one place in space-time

By contrast:

A property K is an object which can have more than one location.

That is, it can exist at multiple places in space-time
Thus a single hydrogen atom might count as particular since it will be
uniquely located, whilst various characteristics it has such as ‘mass’, or
‘charge.’’ will count as properties since other atoms may have them at the
same time. A more elaborate version of the particular/property distinction
186

is the distinction between substances and universals. A substance counts as

a particular since it will be uniquely located. But it involves further ideas
such as Locke’s idea of them as a ‘substratum’ or property bearer [14, pp.
39-50]. Aristotle offered a still more subtle and complex view of substances.
In the Metaphysics he defined substances as “impredicables” objects which
cannot be predicated of other objects. Substances here serve as summum
genera - paradigm subjects of scientific investigation.
Like properties, universals have the capacity for multiple location but
as with the particular/substance continuity they too carry slightly more
metaphysical baggage than their more sparse cousin. Plat0 and Aristotle
famously disagreed about universals and so set off an argument that has
been running in philosophy ever since. On Plato’s view, true universals were
‘forms’existing in Platonic heaven that could only be physically manifested
in imperfect ways. For Aristotle, all universals were ‘immanent’ - located
materially - a further basis for the greater palatability of an Aristotelian
ontology to natural scientists than a Platonic 0ne.g But whilst properties
can be contrasted with particulars in terms of their locative character they
need not be taken to really exist in the way their more tangible partners
usually are. Instead, as Quine has repeatedly argued (see for example
his [16, p.1221) their role may be purely semantic - indicative of a range
of linguistic proclivities amongst certain language users. Nonetheless the
presence in language of predicates which indicate recurrence - (“is red”, “is
complex”’), has been taken by many to imply some deep ontological fact
- the existence of properties or their more robust metaphysical correlate,
universals. Properties are often taken to be a more neutral basis for the
concept of a universal - they may even be seen as a way of “solving the
problem of universals” [13, p. 751. Where particulars and properties are
taken to exist in conjunction we have a ‘two’ category ontology. Whether
either is held to exist in isolation a more economical ‘one’category ontology
emerges. Some of the basic possibilities & associated terminology are set
out in Table 1.
The substance/particular, property/universal ontology appears to enjoy
the widest acceptance amongst scientists in general as well as with complex-
ity scientists, so it is here where I will focus my main objections. However
there are of course other kinds of ontologies a complexity theorist might
appeal to. A Whiteheadian ontology of processes or events for example is

gThis is the medieval distinction between universalia ante rem and universalia in rebus.
See [15]for more on this
 187

Table 1.
Particulars
One Category On-
tology jects
part icularised of properties
things with particu-
larised features
Two Category On- All objects are combinations of particulars and
tology properties

an obvious metaphysical resource for complexity science [17]. The appeal of

this lies in the attractions of the dynamical - a ‘Heraclitean’ universe of flux.
But note that processes can be analysed in terms of particulars and their
properties - for example, processes might be construed as particulars that
have temporal stages or parts.. Objects are thereby seen as four, rather
than three dimensional - extended in time as what are sometimes referred
to as continuants (objects with temporal parts) as opposed to occurrents,
objects which are fully present a t each moment. We shall return to this
issue shortly.
A further potential ontological category for complexity science that
might be considered is the idea of Facts or States of Affairs. Both were fa-
mously invoked by Wittgenstein, though they have been less popular within
philosophy than ontologies based on the particular/property dichotomy. A
fact is a specific kind of states of affairs - those which exist in actuality
and which can be the referents of true propositions. Facts have a range of
their own problems which I will not discuss here, but as a species of states
of affairs one can assume that any problem had by the latter will affect
the former.h A state of affairs is normally defined as the complex that is
made up of the conjunction of a property and particular and is ontologically
implied by the use of propositions of the logical form P ( z ) . But therein
lies the problem for the putative category. Either it reduces t o particulars
and properties - in which case it must inherit their problems, or it is an
irreducible complex. But if it does have an existence ‘over and above’ the
particular and property that makes it up then the failure t o understand

such problem is technical - the so-called ‘slingshot’ argument which, it is held,

implies t h a t any ontology of facts would ultimately be committed t o a very unlikely
entity - the one ‘great fact’. See [18].
188

how it is distinct makes it a rather mysterious entity. Alternatively, as a

complex simpliciter, it may just turn out to be a less well developed version
of the ontological category I am about to discuss.’ Similar reflections ap-
ply to a further move that might be made within this particular species of
ontological candidates - that of taking information itself as a kind of onto-
logical category. For not only are there problems arising from the different
interpretations of information available. There is also the fact that it too
appears to be largely subsumed by the candidate I shall argue for shortly.

5. Nominalism and Complexity

Advocates of a one category involving only particulars - nominalists - have
included amongst their illustrious ranks thinkers as diverse as Leibniz,
Hume and Quine. It is not only the desire for ontological simplicity that
drives the interest in this approach. There is also the requirement to re-
spect empirical constraints and to assume the existence only of that which
is observable. By occupying distinct (and finite) portions of space-time
particulars fit these requirements in a way which multiply located proper-
ties seem not to. But nominalism, whilst appealing in the abstract, works
less well in the context of scientific practice - not least because of the com-
mitment of scientists t o mathematical properties and entities. In spite of
elegant attempts to explain away the role of mathematical claims in science‘
the majority verdict appears to be that pure nominalism is an extremely
difficult position t o maintain. The same reflections would appear to hold
where complexity theories are interpreted in nominalist terms. Here, a
purely nominalist ontology would appear to fail for two fairly clear rea-
sons. First, realist interpretations of complexity would appear to involve
commitments t o many kinds of phenomena that require multiple location -
for example attractors, scale phenomena, or dissipative chemical reactions.
The dynamical systems approach surprises us precisely because, as Andy
Clark argues it uses the “same basic approach to tackle many superficially
very different kinds of real world phenomena” [2l, p.1151. This ‘same a p
proach’ assumes ontological continuities which underpin the behavioural
modelling of distinct systems - not the distinct behaviours in distinct sys-
tems that the nominalist must advocate if they are to be consistent. Second,
and following from this, nominalists must deny the reality of relations since

‘See [19] for a discussion of states of affairs and a more developed account of whether
universals and particulars from irreducible complexes
JSeefor example [20] which attempts to develop an approach to ‘science without numbers’
 189

these are in effect ‘dyadic’ properties (properties that obtain in virtue of a

connection between more than entity. For example the relation x > y can
be thought of as a dyadic property of x, or of y). But most complexity
sciences have instances of phenomena which appear to imply the very op-
posite - an intimately relational universe, where discrete elements appear
to suddenly operate in close harmony and ceoperation to produce higher
order phenomena.
The more developed idea of particulars - as substances not only falls
foul of these problems of multiple location but has some problems of its
own. For example, a common distinction here - between primary & sec-
ondary substances does not appear to be a comfortable one within CT. A
primary substance has traditionally been identified with specific, or indi-
vidual objects, whilst a secondary substance is read as a more general type
of the object concerned - a kind. Thus, Napoleon would count as an in-
stance of a primary substance and ‘man’ as a secondary substance. But the
kind of modularity within complex systems that Bechtel and Richardson
identify [22] would appear to make room for graduated classes of complex
objects. That is, a continuum of overlapping objects of varying degrees of
complexity - not objects which are forced to accommodate absolute dis-
tinctions between general and primary kinds. Worse, as we saw earlier,
primary substance in the Aristotelian sense is understood as an ‘impredica-
ble’ - something which can be predicated of anything else ( e g one cannot
say “John is a George Bush” since GB is a primary substance). But this
seems to be an ontological commitment C T ought to avoid for it seems
plausible that it should allow for complexity to go ’all the way down’. That
is, C T ought to allow for infinite nesting of objects within objects, not
metaphysical full stops. And it appears to be a further prerequisite for CT,
not only that its basic objects can generate new basic objects, but that
basic objects can pass out of existence altogether. This again is not easily
made consistent with a substance ontology.
A recent nominalist compromise has been to appeal to the category of a
trope. Tropes avoid the problems to do with the recurrence of features across
multiple locations by ‘particularising’ properties. Thus, trope theorists can
say that any objects properties are made up of bundles of tropes TI . . . T,
where each T is a particularised property - e.g. the tropes Redl . . . Red,
or Shape1 . . . Shape,. Tropes have found some support amongst analytic
ontologists, perhaps because they permit one to have ones nominalist cake
and eat it. But there are many more philosophers who have found tropes
190

to be much too mysterious an entity to fully commit to.k And an interpre-

tation of complex objects and processes in terms of tropes does not seem
very intuitively promising. Every complex object or process would have to
consist of infinite numbers of tropes which would need to ‘come into being’
and then simply disappear as new complex objects emerge. And since there
can be no trope heaven, there could be no explanation of the hows, whens
or wheres of any of these manifestations. We would simply have to accept
that tropes appear in the location they do and then disappear. No general
laws could be invoked that allow for recurring phenomena to underpin this.
For C T then, a commitment to tropes would seem to be more of an act of
philosophical faith than a clarification of its underlying ontology.

6. Problems with the alternatives to nominalism

Alternatives to a nominalist ontology face their own kinds of objection.
There is not space here to run through every kind of possible ontological
solution or each of their specific problems. But as an illustration we can
pick out two classes of problems which appear to pose serious problems for
a seamless application of standard ontologies to CT. Whilst these are not
conclusive they seem to me to require so much ‘fixing’ that what results
lacks in elegance. All the more so if, as I will suggest, there is an ontological
candidate which not only avoids these problems, but which accommodates
certain requirements of complexity theories far more neatly. The first class
of problems relates to problems to do with explanation. The second class
relates to tensions between fundamental ontological requirements of com-
plexity theories and the suitability of the available candidates.
As an example of the first class of problems we might consider expla-
nations of certain kinds of change and how these are dealt with in CT. It
is the kinds of explanation of these changes offered by complexity theories
which contributes a great deal to the idea that they have a distinct scientific
character from more orthodox approaches. For example one claim that is
frequently made is that, as result of increases in the magnitude in control
parameter for some system such as temperature, ‘spontaneous’ manifesta-
tions of completely new kinds of behaviours or characteristics can occur.
These so called ‘phase transitions’ do not appear to be straightforward kinds
of changes which are gradual and long term. Rather these are sudden, and
dramatic and frequently require new kinds of laws or explanations.

kSee [23] for a defence of tropes. See [24] for reasons to question them
 191

Finding ways to accommodate explanations of change and generation in

terms of universals has been a widely discussed problem in philosophy. A
prima facie problem of contradiction appears to emerge when the predicates
that refer to properties relate to properties had by an object at different
times. Since it appears natural to think of any object as being wholly
present at different temporal points, we appear to be committed to the
belief that an object can be both p and not p . For example a piece of paper
may be white, but later it may have been painted black. But then we seem
to be committed to saying that the paper, which is wholly present through-
out its existence both has whiteness and does not have whiteness. Such a
contradiction is clearly untenable with our common sense view of the world.
One obvious response to what goes wrong here [25, p.2041 - the so-called
‘perdurantist’ view - is to say that objects are not wholly present at indi-
vidual times - that is, they do not endure, they rather perdure. This entails
a view of objects as four dimensional with a succession of temporal ‘parts’
or slices that constitute their worm-like extensions through time. But for
this to work one must not only commit oneself to the counterintuitive view
that nothing is ever wholly present, but to the vagaries of time indexed
properties - ‘the property of being P at t’. The idea of temporal parts and
time indexed properties has many critics, but in any case it does not appear
to deal with what would be required for CT. For temporal parts may suc-
ceed each other as one time indexed property is replaced by another. But
to explain phase change in C T by saying that one property is replaced by
another seems to distort the force of the scientific explanation here - which
is to say that the properties undergo a kind of orderly mutation into differ-
ent properties. Not only that, there are transparently identifiable control
parameters that make this possible. Since the same universal or property
couldn’t span the mutation (different properties by definition are involved),
and since the idea of simple replacement of one property by another is not
a very accurate illustration of what is happening here, properties seem to
be weak ontological vehicles for what is, after all, one of the most impor-
tant forms of explanation offered by complexity theories. The category of
process is a direct response to these kinds of problems. In fact Heracli-
tus’s claim that ‘everything flows’ was a specific rebuttal to the atomists
perception of elements of reality where objects do not change but merely
exchange properties. Processes do better than properties here - though it
is less clear how well they deal with the idea of spontaneous phase change.
Since processes just are four dimensional entities, change is fundamental
to them. But C T makes specific claims about certain kinds of changes.
192

Unless there is a further ontological distinction amongst processes to allow

for this, one change is much the same as another. And if there a further
ontological distinction is required it looks like any economy that might have
been gained by the appeal to processes over properties has been lost. Since
processes are a direct response to the problem of change it is not very sur-
prising that they do better than properties. However it is less clear how
well they deal with the second of the problems raised above.

7. Problems with the Order/Disorder Distinction

Amongst the most serious kinds of objection faced by standard ontologies
when it comes to accommodating the research programme of CT must be
those centring on their weaknesses in accounting for the key distinction
between order & randomness. Universals standardly depend upon tacit
concepts of order in lots of ways - not least the fact that they must be
conceived of as occurring in orderly complexes - i.e. ‘attached’ in regular
ways to one or more particulars. As Wittgenstein pointed out, the orderly
conjunctions of universals and particulars indicates something significant
“. . . universals and particulars may appear to be mutually independent but
this independence is a form of connection with states of affairs - a form of
dependence”. Indeed, there is even a suggestion of metaphysical priority
for: “it is essential to objects that they should be possible constituents of
states of affairs” [26, 2.011 & 2.0121.
But the question of how they do this has never received any satisfac-
tory answer. Indeed, the role of ordering (processes) is usually ignored or
brushed to one side. One motivation for this wilful denial to address some-
thing so fundamental may lie in the disaster that results from one famous
attempt to explain the orderly connection between particulars and their
properties. On this (Platonic) line an additional universal is introduced as
a way of providing the regular link between particulars and properties. But
this only serves to lead, inexorably, to a logical and metaphysical regress
(for what then ‘binds’ the binding universal, other than a further one?
And so on).’ Aristotelian style solutions avoid regress by simply positing
the binding between a universal and particular as something primitive and
inexplicable. The problem then is any concept of how this orderly connec-
tion holds must remain to unexplained. As its stands then accounts of order
on this line appear to be missing, incoherent or to lead to inconsistency.

‘The so-called ‘Third M a n Regress’ [27]

 193

If the tactic of ignoring order or providing a metaphysical basis for it

fails then what of attempts to analyse it away in other terms? In general
specific attempts to analyse order in terms of properties seem to lead to
further failings. For example, one way this might be essayed is in terms of
a second order property - a property of a property which holds in virtue
of other things being the case. On this line order is conceptualised in
terms of a property that exists only where other properties do. But not
only have second order properties tended to be regarded as an unnecessary
or mysterious kind of commitment,”’. it is not clear whether they could
offer the kind of conceptions of order that is consistent with the claims
of complexity scientists. CT appears to permit different degrees of order
- scalar conceptions rather than the discrete ‘pockets’ which second order
properties (or third order, and so on) would seem to represent. So second
order properties fail twice here. First in the failure to illuminate notions
of “degrees” of order. Second, because a commitment to second order
properties as an analysis of order would further commit CT to a hierarchy
involving properties of properties, of properties and so on. Again the threat
of a (vicious) infinite regress raises its head.
Perhaps then it would be better to deal with order by treating it as a first
order property simpliciter. But this too has some undesirable outcomes.
For presumably randomness could be a lack of the property of ‘order’.
Random phenomena plays a key explanatory and causal role in CT - for
example in discussions of entropy production. To allow causal efficacy to
arise as a result of ‘lack’ of another thing sits uneasily with our normal
understanding of causation. So if order could be a first order property,
so too would randomness need to be - if only for the reason that this
would facilitate the distinction of random things from orderly things. But
then suppose there were such a property of “randomness”. This would
imply the existence of some method of being able to distinguish a random
property from an orderly property. But then we would be saying that
there is potentially decidable method for deciding something is random.
Yet we know from [31] that there is no such method - indicating that a
binary string has the property of randomness for example is not a thing
that be definitively demonstrated. To do so would require a program of
less complexity than the random string it outputs. But then the string
would have a minimal program and would have some pattern. To claim

“‘For some objections to second order properties see [28]. For some arguments in favour
see (291. (301 discusses the idea of a second order logic
194

that the randomness of a binary string is demonstrable leads inevitably to

the contradiction of saying it is not random.
At this point it becomes clearer why our other candidate ontological
category - that of processes, runs into problems. Whilst processes appear
to deal with change more naturally than the’contortions we are forced into
with properties, similar problems seem to arise when it comes to applying
the distinction between order and randomness to them. For we similarly
ask - what distinguishes a random from a non random process? Either we
get the unilluminating response “because one process is orderly and one is
not”. Or, by attempting to definitively identify a process as ‘random’ we
run the risk of encountering the same kind of contradictory outcomes as
before.

8. Further, More General Objections

In each case where a property is used to explain order there seems to be
something extra required - order (however it be explained) and the entity
that is ordered. Further more general kinds of objection to this picture that
arise suggest a further kind of strategy might be worth considering.
(i) Objections relating to economy. If ‘order’ appears to be something
that is an additional requirement for properties, either in ontological
or analytic terms, might it not be more economical to drop the
commitment to properties, processes or whatever and to investigate
whether order itself could be made to do the ontological legwork?
(ii) Objections relating to reduction. It appears that there may be
reasons for suggesting that traditional categories can in any case
be themselves subsumed into this putative category of order. These
reasons revolve around arguments that appear to suggest that the
very conceivability of properties, particulars and so on depends upon
a prior conception of order.
An indication of this potential conceptual reducibility can be seen if we
examine again the way considered earlier in which universals, particulars
or indeed processes are distinguished. As we saw, one powerful way in
which this can be done is in terms of location and more particularly, the
way that the identity conditions for properties, particulars and the like are
fixed in terms of differing characteristics of location. An extended definition
incorporating identity might go as follows:
A particular P is the same P if it is identifiable as the same P
 195

across unique locations at different times

A property Py is the same Py if it is identifiable as the same Py

across multiple locations at the same time

A Process Po is the same Po if it is identifiable as the same Po

across multiple locations at different times

In other words, we obtain the outlines of what Kant called a ‘transcen-

dental’ argument for a priority of order, or pattern over these Categories.”
That is, an argument which demonstrates the plausibility of accepting the
existence of something by showing that it is a necessary condition for posit-
ing the existence of other things. Here it relates to how certain kind of prior
order needs to be in position if we are to invoke the existence of properties
or particulars. And there are other ways too in which this apparent con-
ceptual priority of order might be illustrated. For example, set theory - a
foundational tool amongst analytic philosophers for the conceptualisation of
phenomena in both the physical and the mathematic contexts can be shown
to be lacking when it comes to analyse order itself. [33] indicated how the
‘Wiener-Kuratowski’ procedure, which attempts to analyse ordered pairs
in purely set-theoretic terms requires a further element - that of choice,
one which in effect, masks its own implicit commitment to the priority of
0rder.O There is not enough space here to consider many other similar ex-
amples, but the implications are clear enough - the apparent metaphysical
priority of order over more developed ontological categories indicate that,
for CT, this may be the best direction for developing its own ontology.

9. Patterns
If venerable and long serving range of metaphysical candidates such as these
are found wanting as the referents of the language of complexity, then a neg-

“See (321 for a discussion of this form of argumentation

OThe Wiener-Kuratowski procedure states that an ordered pair < x,y > can be construed
in terms of a class such that < q y >=def. {{x}, {x,y}}. Any pair < x,y > can be
introduced if the following condition holds: < x,y >=< w, z > iff (z= w&y = z). Since
{{a},{a,b}} and {{a,b}, {a}} denote the same class it would seem that ordered pairs
(and hence order) has been analysed in terms of an unordered entity - the class. But
as Hochberg points out if one “uses sets like {{a},{a,b}} and {{b}, {b, a}} one chooses
which of { { a } , { a , b } } and { { b } , { b , a } } is to represent < a , b > and which represents
< b, a >. This shows that in the construal of < a,b > as {{a}, {a,b}} one implicitly
takes a and b in and ordering for one takes the element in the unit set as the first
element”. Russell dismissed the whole procedure as a trick.
196

ative conclusion might be that is the language of complexity itself which is

wanting. But another, more positive conclusion seems possible, one that
draws upon these suggestions of some kind of priority of order. Such prior-
ity suggests that there might be a further powerful and plausible alternative
ontological candidate in waiting - a reification of order in all its different
forms itself. This category of ‘pattern’ (to use a more convenient term) not
only sits very neatly with fundamental conceptions in C T - specifically a
definition of complexity itself, it also offers a certain kind of objectivity in
apprehending the notoriously subjective divide between what can be con-
sidered to be ‘ordered’ and what is not. For there is a definition of pattern
that both allows for a degree of observer relativity - patterns that exist only
in the eyes of beholders (faces in Rorschach inkblots for example) as well
as permitting only a certain class of patterns, the ‘real’ patterns, to have
scientific relevance. The definition, taken from the idea of computational
complexity - states that a pattern in some binary sequence can be defined
if it has a compression - a program shorter in length than the sequence
itself that can be used to generate the sequence. And the real pattern in a
sequence is only identified where a ‘minimal program’ for it has been iden-
tified - the shortest possible sequence which can be inputted to a Turing
Machine so as to produce that sequence as an 0utput.P There are many
kinds of lhring machine, but since the definition relativises all definitions
of length to the Universal Turing machine - that Turing machine which can
simulate any member of the class of Turing Machines there is no ambigu-
ity in what counts as the shortest program for produce the sequence and
hence defining the pattern there. There are too many qualifications and
developments of this basic idea to go into it in detail hereq but it proved
tempting enough for Daniel Dennett to make use of it in his [36] as a way
of better defending his views about mental representation and the idea of

PSee[34] and [35] for the earliest discussions of the compression criterion for pattern and
the idea of algorithmic complexity
qIn my (forthcoming) I consider a way in which computational complexity conceptions
of pattern can be augmented by a complementary, more naturalised conception of them.
This relates to the idea of symmetry and the deeper notion that goes with this - the
notion of invariance (through transformations of various kinds). This idea provides for
a continuity between overtly mathematical or geometrical patterns and the regularities
in physical objects. On the one hand there is invariance and stability in the face of
transformation through mathematical operations, on the other stability in the face of
transformations through physical processes such as time, causal effect and change. A
plausible bi-conditional holds, one that relates the computational complexity conception
of pattern t o this symmetry, or invariance notion
 197

an intentional state. Such states are, in a nutshell, grounded in real pat-

terns. But Dennett showed little interest in further developing this brief
foray into ontology and the idea of pattern as an ontological category of
interest has not been developed much further within analytic philosophy.
Aside from an eccentric paper from the 1940’s and some discussions within
the philosophy of mathematics, little else has been said.’ This seems odd,
not least because patterns appear to play a central theoretical role in many
other ways that simply grounding intentional, or mental states. We can see
oblique references throughout philosophical history to them in the form of
causal regularities; as dispositions and capacities; in the appeal to laws, the
salience of structures and structural thinking in the philosophy of science,
and so on. In some cases key forms of scientific reasoning such as induction,
or the best way to connect statistical observations into hypotheses (the so-
called ‘curve-fitting’ problem) can be shown to have specific derivations
from complexity definitionss

10. Patterns and Complexity

This widespread (but tacit) use of patterns in philosophy and science in-
dicates that, at the very least, there is something here worthy of further
investigation. And for CT patterns may not just be an interesting poten-
tial metaphysical resource - they may be indispensable to its practice. The
very foundation of CT lies in the task of explaining how elements as di-
verse as molecules, plants, or economic actors act collectively to to produce
complex patterns of behaviours. To this can be added a range of other de-
sirable features which indicate their usefulness for complexity science. For
example:
(i) Definitions and identity criteria for other ontological candidates
such as properties are frequently unclear. Patterns by contrast can
draw upon resources for definition and identification from algorith-
mic and invariance notions, resources which offer impressive levels
of clarity.

‘The discussion is 1371. 1381 argued that mathematical objects can be dispensed with in
terms of structures or patterns such as ( N >) where each number functions as no more
than ‘position’ in this pattern. It is not a view he has developed in any great ontological
detail
“Solomonoff 139) develops a model of induction which involves parallel conceptions to
Chaitin & Kolmogorov’s ideas about compressibility. I401 indicates a way in which
Solomonoff’s idea can be extended into the “MinimumDescription Length Principle” as
a way of solving the curve-fitting problem
198

(ii) As we saw, there is a case for arguing that patterns subsume many
traditional metaphysical categories. The necessity for depending
upon some form of regularity or continuity in order to identify and
reidentify anything at all provides both justification for priority as
well as suggesting something very basic underpin this need for or-
dering.
(iii) Patterns serve as precisely the kind of entities that can provide a
grounding for the apparently universal capacity within science, and
especially complexity science to distinguish between more or less
ordered phenomena.
(iv) Even better, a multiplicity of patterns allows complexity science
to justify its rejection of reductionism and so permit a plurality of
explanatory levels and processes.
(v) Patterns provide a way of grounding the apparent ubiquity of regu-
larity in the world given its extremely low probability. Continuities
in object identity, regular causal connectivity or the tendency of
complex systems to resist the pull of entropy all acquire an ontolog-
ical grounding they would otherwise lack.
(vi) By extension, patterns not only offer important grounding concep
tions into the concepts of ‘Laws of Nature’ and of theory change
in science (the development of increasing sphiosticated compressors
of the world. As Ray Solomonoff’s work indicated [39] they may
illuminate some of the mechanisms for inductive reasoning and its
successes (the search for minimal programs in the identification of
natural phenomena).
(vii) Patterns seem also to offer insights into numerous other mysterious,
but desirable principles of scientific reasoning - in particular the
effectiveness of Occams razor - the appeal to simplicity (or shortest
compressor).

11. The problems with patterns

In addition to the difficulties faced by other metaphysical candidates, the
variety of such a list seems to me to be a further reason why CT should
take the category of pattern highly seriously. But of course, life is never
simple. For whilst patterns certainly seem to have some good credentials
to be the referents of claims within the sciences of complexity, this does not
mean that there are not also number of problems which would need to be
addressed if they are to function in effective ontological terms. I will list a
 199

few of the more significant ones these before concluding.

(i) There appear to be a range of problems involving the nature of

the relationship between a pattern and the elements which go to
provide its ontological foundation. Prominent here is the issue of
foundational elements - especially should a one category ontology
which involves only patterns be thought desirable. For without
particulars, or other traditional ‘substantial’ entities to play the
role, it must be the case that patterns serve as the constituents of
patterns. For many philosophers this would lead to an undesirable
outcome - the lack of a categorical base. And since patterns seem
always to be patterns of other things we would seem to end up
lacking in any ‘material’ like stuff in world. Instead we seem only
to have a chain of dependencies - something like a series of brackets
within brackets, within brackets - a series that never reaches any
substantial kind of thing.
(ii) There are also problems of extending the computational complexity
line to patterns in, or of physical things. Since the algorithmic
definition of patterns requires clearly definable binary sequences for
it to work properly, using it to define patterns in physical objects
requires that binary sequences can be shown to adequately model
or simulate such objects. But this then invokes the problem of
what a ‘truthful’ or even ‘adequate’ representation really is. And
this remains one of the most controversial and difficult problems in
philosophy.
(iii) The problem of pattern identity. In the formal context pattern iden-
tity is fairly clear. P=P’ if both have the same minimal program.
Or, focussing upon patterns elements, one might say that P=P’ if
there is an isomorphism or 1:l mapping between the elements. In
less formal contexts, especially where there is certain ambiguity in
saying what the elements of some pattern actually are, identity con-
ditions are less clear. For which ‘elements’ would we select for the
mapping? And how 1:l could it really be? Yet without a proper
criterion for identity there could be no proper ontology of patterns.
(iv) The risk of tautology. Patterns may be a desirable way of accounting
for the ubiquity of order. But as orderly things there is the very
real risk that we end up saying no more than ‘order is understood
by way of order’.
(v) The problem of accounting for causality. It certainly seems plau-
200

sible, as we saw earlier, t o say that the regularities seen in causal

relations have a pattern t o them. But if causality is also understood
in purely physicalist terms as some kind of “bumping” or “impact-
ing” effect (cf. [41, p. 181) then the question remains as to what
could be doing the bumping and impacting

I do not think these problems are fatal, though they are certainly dif-
ficult problems that need to be addressed. My (forthcoming) addresses
some of them and it is to be hoped that additional lines of research can
deal with others. But if, in the end, an ontology of patterns turns out t o be
as unworkable for complexity science as traditional ontologies seem t o be,
then perhaps we would need t o look again at how scientifically robust the
patchwork of claims and observations that make it up can really be taken
to be.

References
1. McGuire M.R. 1999 On a General Theory of Pattern Doctoral Dissertation,
University of London
2. McGuire M.R (forthcoming) Patterns Kluwer Academic Press
3. Williams, D.C. 1953 “On the Elements of Being”, Rev. Metaphysics 7. 3-18,
171-192
4. Strawson, P. 1959 Individuals: A n Essay in Descriptive Metaphysics, Rout-
ledge
5. Davidson, D. 1970, “Mental Events”, in D. Davidson, Essays on Actions and
Events, Oxford, Clarendon Press., pp. 207-227
6. Kim, J. 1993 Supervenience and Mind: Selected Philosophical Essays ,Cam-
bridge: Cambridge University Press
7. Cartwright, N.1983 How the Laws of Physics Lie, Oxford: Clarendon Press
8. Cartwright, N 1989 Natures Capacities and their Measurement, Oxford:
Clarendon Press
9. Anderson, P & Stein, D 1987 ”Broken Symmetry, Emergent Properties, Dis-
sipative Structures, Life: Are They Related, in F. Eugene Yates (ed.), Self-
Organizing Systems: The Emergence of Order (NY: Plenum Press, 1987), p.
445-457
10. Laughlin, R.B. and Pines, D.2000 “The Theory of Everything,” Proceedings
of the National Academy of Sciences 97 (l),28-31
11. Prigogine, I. (et al) 1976 “Kinetic theory and ergodic properties” Proc. Nati.
Acad. Sci. USA Vol. 73, No. 6, pp. 1802-1805
12. Kaufmann, S. 1993 The Origins of Order: Self-organization and Selection
in Evolution, Oxford, OUP.
13. Oliver, A. 1996 “The Metaphysics of Properties”, Mind, 105, (pp. 1-75)
14. Ayers, M. 1991 Locke, Vol 2, Part I “Substance and mode”, Routledge, Lon-
don.
 20 1

15. Armstrong, D.M. 1978 Nominalism and Realism: Universals €9 Scientific

Realism, Vol I, Cambridge, Cambridge University Press
16. Quine, W.V.O. 1960 Word and Object, Cambridge, Mass: MIT Press
17. Whitehead, A.N. 1929 Process and Reality, New York: Macmillan
18. Neale, S. 1995, “The philosophical significance of Godel’s Slingshot.” Mind
Vol. 104, NO. 416, pp. 761-825.
19. Armstrong, D.M. 1997 A World of States of Aflairs, Cambridge, CUP
20. Field, H. 1980 Science Without Numbers: A Defence of Nominalism Prince-
ton, NJ: Princeton University Press
21. Clark, A. 1997 “Emergence and Explanation” from Being There: Putting
Brain Body and World Together Again, Cambridge MA: MIT Press, p.103-
128
22. Bechtel, W., Richardson, R. C. (1992). Emergent Phenomena and Complex
Systems. In A. Beckermann, H. Flohr, J. Kim (Eds.) Emergence or Reduc-
tion? Essays on the Prospects of Nonreductive Physicalism. (257-288). Berlin:
Walter de Gruyter
23. Simons, P. 1994 “Three trope theories of substance”, Philosophy and Phe-
nomenological Research, 54, pp. 553-575
24. Denkel, A. 1997 “On the compresence of tropes”, Philosophy and Phe-
nomenological Research, 57, pp. 599-606
25. Lewis, David. 1986. O n the Plurality of Worlds, Oxford: Blackwell.
26. Wittgenstein, L. 2001 Tractatus Logico-Philosophicus, (trans. David Pears
and Brian McGuinness), Routledge
27. Vlastos, G. 1954 “The Third Man Argument in the Parmenides” Philosoph-
ical Review, 63, 319-349.
28. Kim, J . 1998 Mind in a Physical World: A n Essay on the Mind-Body problem
and Mental Causation. Cambridge (Mass.): A Bradford Book, The MIT
Press
29. Gozzano, S. 2003 “Second Order Properties: Why Kim’s Reduction Does
Not W o r k , Logic and Philosophy of Science, Vol. 1 No. 1
30. Boolos, G. 1975 “On Second-Order Logic”, Journal of Philosophy, 72 pp.
509-527;
31. Chaitin, G 1975 “A theory of program size formally identical to information
theory”, Journal for the Association of Computing Machinery, 21, pp. 403-
424
32. Cassam,Q 1987, “Transcendental Arguments, Transcendental Synthesis and
Transcendental Idealism” Philosophical Quarterly
33. Hochberg, H. 1981 “The Wiener-Kuratowski Procedure and the Analysis of
Order”, Analysis, 41, pp.161-163.
34. Chaitin, G. 1966 “On the length of programs for computing finite binary
sequences’ Journal for the Association of Computing Machinery pp. 547-569
35. Kolmogorov, A. 1963 “On tables of random numbers”, Sankhya, A, 25, pp.
26-376
36. Dennett, D. 1991 “Real Patterns” Journal of Philosophy 88, pp. 27-51.
37. Bush W.T. 1940 “Concerning the concept of pattern” Journal of Philosophy,
37, pp.113-134
202

38. Resnik, M. 1997 Mathematics as a Science of Patterns, Oxford: Clarendon

Press
39. Solomonoff, R. 1964 “A formal theory of inductive inference”, Infomation
& Control, 7, pp 1-22 (Pt I) & pp. 224-254 (Pt 11)
40. Rissanen, J. 1978 “Modelling by the shortest data description”, Automatica,
14, pp. 465-71
41. Crane, T . 1995 “The mental causation debate” Proceedings of the Aristotelian
Society Supplementary Volume L X I X
 HOW TO LOVE THE BOMB
Trying to solve the prisoner’s dilemma with evolutionary game
theory

VASCO CASTELA
University of Manchester
School of Social Sciences - Philosophy

Economists traditionally see altruistic acts as irrational. However, in the Pris-

oner’s Dilemma, a rational player can do worse than a moral player. The rules of
the game imply that one cannot defend one’s best interest if one tries to. Game
theory has struggled t o explain how an agent could have access t o the strategically
best outcome without behaving irrationally, but with little success. Can a complex
systems approach do better?
Peter Danielson, using Evolutionary Game Theory, has avoided some of the as-
sumptions of Game Theory by using a complexity approach to reframe the prob-
lem, and offers a solution of sorts. According t o Danielson, the foundations of
altruism are ‘mechanisms of deterrence that rely on credible threat - we are nice
for fear of retaliation. He is both right and wrong. It will be argued that utili-
tarian, consequentialist principles must have been at work t o create the conditions
for altruistic acts t o be performed. It is wrong to expect, however, that the same
reasons are the reasons for action.
In order for a model of genuine altruism to be possible, an extra cog must be
inserted in the mechanism of causality in order t o distance moral action from its
strategic advantages. If emotions fulfill this role, we can tell a story in which it is
rational to act on altruistic motivations and materially advantageous to hold such
motivations. Moral sentiments can be seen as a tool designed by evolution t o help
optimize cooperation in a social environment. The proposed account integrates
the Humean theory of motivation with Robert Frank’s commitment model and
Aristotle’s views on moral education, keeping an adequate story of how it can be
in our material interest to be moral without having to renounce t o the existence
of genuine acts of altruism.

1. How to love the bomb - the MAD strategy

In most situations of strategic tension, the outbreak of actual conflict does
not imply the end of diplomacy. A war ordinarily involves a number of
battles and plenty of time and opportunity for negotiations, such as tem-
porary ceasefires and conditional or unconditional surrenders. This was
not the case in the Cold War, as the highly destructive power of nuclear

203
204

weapons meant there was a decisive advantage in being the one to strike
first. After an initial attack, in which the enemy’s defenses would be ren-
dered inoperative, victory was sure. This of course made giving the order
of striking first both very tempting and quite final, as once an attack had
been launched and the bombs were in the air, there was no turning back.
The instability of this situation led both American and Soviet govern-
ments to adopt a doctrine of nuclear deterrence known as Mutual Assured
Destruction (MAD). The idea was for each to eliminate the opponent’s
strategic advantage of striking first by stockpiling strike-back missiles that
could be launched within minutes of the initial attack or that would survive
the blasts by being placed in bunkers. If each state could provide evidence
of adequate retaliatory capability, this would give the other state a good
reason not to make a first move, as it was to expect heavy casualties in case
of war, regardless of whether it was the first to attack.
The situation seemed under control until an influential game theorist
claimed MAD was not entirely safe, as mutual destruction could not be
assured - which implied it could not be a reliable deterrent. After an initial
strike from the Russians, for instance, why would the US retaliate, if this
no longer served any rational purpose? If we consider a situation in which
Soviet missiles had been launched, this would mean American bombs had
already failed to act as a deterrent. Why use nuclear weapons, at that
point, for killing millions of Russian civilians, with no possible strategic
gain to be had? If retaliation was irrational, there would be no assurance of
mutual destruction, and the strategic advantage of striking first would once
again be in place. The solution the game theorist devised to this dangerous
situation was a doomsday device, a machine that would be programmed to
automatically retaliate in case of attack. The existence of such a machine
would bring the situation back to a MAD condition of equilibrium.
MAD was the basis for the plot in Stanley Kubrick’s hilarious dark
comedy, Dr. Strangelove or How I Learned To Stop Worrying And Love The
Bomb. In the film, the President of the US discovers, following the launch
of a nuclear strike by a rogue American general, that the Russians had built
a doomsday device to ensure the effectiveness of nuclear deterrence, only
had kept it a secret as they were saving the announcement for the upcoming
elect ions.
 205

2. The problem
MAD may not seem such a good idea now, but it was the best solution that
could be found to a puzzling problem: rational action does not always bring
about the most materially favourable state of While acting in the
interest of national safety, nations were paradoxically putting themselves
in increasingly great danger. Note that the notion of rationality we are
discussing is that of self-interest theory, underlying game theory and main-
stream economics (and cold war politics), according to which an action can
only be rational so long as it is in the interest of the person who performs
it. No form of giving can then ever be rational, unless it happens as a
part of a trade. Or, to use the terms of game theory, an agent's action is
said to be rational if it maximises his utility, according to a set of personal
preferences.
While this theory of rationality may seem to make rather pessimistic
assumptions regarding human nature, it does have the merit of acknowl-
edging the trivial fact that systems (states, companies, cells, humans, and
so on) must conform to the laws of physics. Systems must conserve their
resources and ensure their security if they are to survive. In a situation
where conflict seems likely, it is perhaps inevitable that states will care
about their national security above that of their opponents, if they are to
remain a state. The same is true regarding the actions of individuals.
We could explain altruistic behaviour by saying, like Hume, that when
we act unselfishly we are merely satisfying a natural instinct of benevo-
lence, which would enable us to claim that altruism is compatible with
self-interest.17 If we think like Hume, altruistic behaviour no longer seems
to be counter-intuitive. Such psychological explanations, probably of great
importance in a complete ethical theory, do not deal, however, with the very
real problem that the range of psychological reasons must be restricted by
material demands, if we want to have a system that survives. Regardless
of whether we prefer to count altruistic interests as rational or irrational,
we must provide an explanation for how they arise and remain viable.
The strategic advantage of striking first, during the cold war, before
MAD was put in place, was not simply a mistake of the analysts. It was
quite real. While both nations would be better off by not going to war,
striking first was the only safe way for each to avoid total annihilation.
Diplomatic negotiations, interestingly, could not help, as no promise could

"An interesting analysis of this and other interesting commitment issues involving the
MAD strategy can be found in 141.
206

be trusted in that situation. What could the penalty for lying be after the
trusting party had been destroyed? One of the most popular and systematic
approaches to try to solve the problem of trying to make peace between
rationality and cooperation has been to study the Prisoner’s Dilemma, a
strategic puzzle which closely mirrors the nature of the cold war situation.

3. The Prisoner’s Dilemma

The limitations of rational action were popularly studied in game theory in
the form of the Prisoner’s Dilemma (PD). The dilemma arises out of the
following situation. Consider that two individuals, partners in crime, are
arrested and each is interrogated in private. Each of them is told that all
charges against him will be dropped if he rats on his partner, who will as a
consequence get a longer prison sentence. Game Theory predicts that, as
each will realise that, regardless of what the other does, he will always get
a lower prison sentence by ratting on his partner, they will both do it, and
both will get iong prison sentences. In game theory terms, mutual defection
will be the dominating strategy. If they had both chosen to cooperate and
remained silent, however, they would both have been better off, each with
a relatively short sentence. The paradox is that, similarly to the cold war
situation, as each rationally chooses to do what is in his best interest (in
game theory terms, what awards the highest payoff), they both stand to
lose.
As Skyrms4 notes, however, from the standpoint of rational choice the-
ory, the subjects’ utility function is up to them. Game theory does not
say humans are selfish, it rather describes the behaviour of self-interested
agents. The distinction makes sense because game theory imposes no re-
strictions on what an agent’s interests can be. So if a person has a preference
for the happiness of others, game theory would recognize it is in his self-
interest to seek it. If we imagine a version of the PD in which the prisoners
are partners not in crime but in kindness, and the payoffs are cheques of
different amounts to the prisoners’ favourite charities, the results will be
exactly the same. Each kind player, by trying to maximise the amount
of the cheque to his charity, which he thinks will bring the greater good,
will act rationally by not cooperating. Acting so as to maximise their non-
selfish interests would still bring about the same suboptimal outcome and
they will both come out with lower cheques.
So we can see that the origin of the problem as defined by game theory
is not that players are selfish, but that they care to maximise their inter-
 207

ests, regardless of whether these are selfish interests or not. This suggests
the obvious solution: what if each would care about the other’s interests?
According to Peter Kollock, game theory’s assumption that agents make
decisions based on fixed payoffs is m i ~ t a k e n .For
~ Kollock, a subjective
sense of moral righteousness could affect them, making cooperation seem
more attractive. As an example, imagine that in the PD the agents would
prefer to help their friend rather than try to avoid prison time at all cost.
The payoff for cooperation would then be higher and cooperation would be
rational. However, as we have seen above regarding Hume, this solution
would solve the problem at the expense of taking the explanation of why
they would adopt this preference for granted.
In practice, economists label altruistic actions as irrational.’ This is
a fair assumption, or at least a good starting point, as it deals with the
problem of physical viability of a system. It makes sense to expect not
only that the agent will satisfy his interests, but also that these interests
will be self-serving. Any other claim will contradict the Darwinian model
of evolution. Evolution does not care about subjective moral principles. A
genetically inherited trait only gets selected if it directly contributes (in a
material sense) to the survival of the individual and, to quote Robert Frank
(1998:54), “We can’t eat moral sentiments”. So, in order for altruistic
behaviour to emerge and spread, it seems it must be advantageous in a
material way. How could this work? There have been a few interesting
attempts to try to explain how altruistic behaviour could have emerged
(See [3] or [l],for a detailed survey), but, as we will see, they all have their
limit at ions.

4. Naturalistic accounts of the emergence of altruism

The idea of altruism arising by group selection was first mentioned by Dar-
win, who claimed that a group in which individuals would self-sacrifice
for the common good would be superior to other groups in the fight for
survival. Altruism could then emerge by the normal process of natural se-
lection. The problem with this argument is that every element inside the
group would have an advantage in saving the energy costs of cooperation
by being a free-rider, exploiting the altruist. This implies that a mutant,
born with no instinct to cooperate, would have better survivability and so
would in time leave more offspring than a cooperator, therefore soon dom-
inating the population. Altruistic behaviour would not be, to use the term
coined by evolutionary biologist Maynard Smith, an “evolutionarily stable
208

strategy” .’
Kin selection was proposed by Hamilton and can be understood as a
variety of the group selection argument, but with much better support.6
Hamilton claims it would make sense for someone to sacrifice himself for his
family as the genes he shares with his kin would live on. This is an argument
easily understood today if we think in terms of Richard Dawkins’ notion of
the selfish gene.1° The phenotype (the body, a physical implementation of
the genotype) may be sacrificed in order to save the genotype (the DNA).
An instinct for sacrificing oneself for the sake of our genotype would be an
Evolutionarily Stable Strategy (ESS).
Kin selection, however, does not explain why some of us are willing
to risk our lives to save a stranger and surely, in today’s social life, most
of our interactions are not with kin. It could be argued that this is a
recent phenomenon in the history of human evolution, and that perhaps
our genetic programming has not had time to adjust to our new lifestyles.
Robert Frank, however, correctly argues that genetically a second cousin
is almost no different from a stranger, and so even if humans used to live
in small groups in recent natural history, kin selection could not by itself
explain the evolution of altruism.’
For Robert Trivers, reciprocity is the basis of altruistic behaviour.” If a
set of players was to play the PD a number of times, a failure to cooperate
could be punished in the next interaction.b The success of the Tit for
Tat strategy in Axelrod and Hamilton’s study of the PD lent considerable
empirical support to this claim.12
Reciprocity can also work indirectly through r e p ~ t a t i o n In
. ~ real-world
situations, behaviour that could appear altruistic at prima facie could con-
tribute to our establishing a reputation for being honest or kind, which
could give us an advantage in future interactions. In many cases, we have
an interest in keeping a promise because we would be punished if we did
not. However, we all know situations where we could stand to gain by
behaving selfishly but behave altruistically, not merely cooperating with

bTo be precise, the number of plays would have to be unknown to the players for co-
operation to be possible. If they both knew they would play 10 times, for instance, it
would be rational for both to defect on the 10th game, as there would be no possibility
of retaliation. This would mean there would be no possibility of retaliation on the 9th
game either, as both would know they would defect on the last game anyway. This would
then be true of all games, and so they would never cooperate. The versions of the puzzle
that are the hardest t o solve (and the most interesting for the purposes of this paper)
are either the single-shot PD or a variant in which the number.
 209

others for mutual advantage but going to some considerable trouble to help
a stranger, for instance, with no hope of material gain.
Reciprocity’s major drawback is that it fails to explain cooperation when
there is no possibility of retaliation. Even when humans play a PD-style
game a single time (also known as the single-shot PD), when not only
defection is the only rational choice but also it cannot entail retaliation,
results do not follow the predicted result of game theory. Many of us give
tips in restaurants when we are on holidays away from home. Why would
anyone pay for good service after he already got it, if he could simply leave
and save some money? Some of us would risk our lives to save someone
we do not even know. Reciprocity does not seem to be able to solve our
problem. Our aim should then be to explain how altruism is possible in
the single-shot PD, which reciprocity fails, not least because thinking of
altruism as being based purely on conscious seeking of material gain surely
contradicts our common sense understanding of what genuine altruism is.
For some of those not satisfied with the standard accounts of how al-
truistic behaviour emerges and who do not want to abandon game theory’s
concept of rationality (which has the merit of dealing with the issue of
physical viability of ethical systems), a complexity approach seemed an
interesting route. Skyrms, Danielson and Axelrod, among others, wanted
to see how adding an evolutionary dimension to traditional models would
change the nature of the PD, making cooperation a rational strategy.* 21
12

According to Jason Alexander, there are two main approaches to evolu-

tionary game theory: one in which strategies reproduce by natural selection
and another in which the mechanisms of selection of strategies are included
in the model.8 We will take no more than a brief look at an example of the
former type, which we will call type 1, as it takes for granted the mech-
anisms we want to understand. We will then discuss in more depth an
example of the latter type, which we will call type 2. We will see what
this approach achieves and how it fails to escape some of the rationalistic
pitfalls of traditional game theory.

5. Complex Systems and Evolutionary Game Theory

Even in the 21st. century, weather forecasting seems to be a fiddly bit
of science. Forecasts are famous for being all too often disastrously wrong.
Only short term prediction is reliable, and this is still on a probability basis.
The reason why it is hard to predict the weather is not so much that we do
210

not understand the mechanics involved in it. The main problem does not
lie in a lack of knowledge regarding intra-molecule interaction, but on the
fact that the behaviour of every single molecule in the atmosphere has a
role in the behaviour of the system as whole. As the climatic system cannot
be decomposed into large chunks of functional parts as, say, the engine of a
car, it resists traditional methods of analysis. The behaviour of the climate
is a result of the rich interactions between its elements - it is a complex
system. Human society, like the weather, is composed of a high number
of functional elements. Interactions between humans are complex and it
certainly seems that any model that ignores them will be overly reductive.
In weather forecasting, thousands of measurements of the temperature
and pressure of the atmosphere at different altitudes are obtained by satel-
lite and weather balloons and are inputted into computer simulations. In
the simulation, time can then be fast forwarded to yield predictions. The
more detailed the model is, the more accurate will the predictions be. Much
as even a perfect understanding of the interaction between two molecules
would not yield good weather forecasting, game theoretic analysis of the
interaction between two rational agents will not yield adequate models of
the mechanisms of altruism. Evolutionary Game Theory (EGT) introduces
the concept of population in game theory, eliminating a perhaps excessive
reduction of standard game theory that may be throwing away the baby
with the bath water.
When playing the PD in a game theory environment, the agent who
cooperates when the other defects will have the lowest payoff of all, as we
have seen. He will get the so-called “sucker’s payoff’. It is only mutual
cooperation that is beneficial for both, which is the reason why defection is
such an alluring strategy. EGT models have been instrumental in showing
that a cooperator will not necessarily do worse than a defector even when
playing with other defectors. Consider a typical type 1 model: there is a
virtual space, a grid in which a number of strategies, some “Cooperate” and
others “Defect”, placed at random, interact. The game theoretic concept
of utility is replaced with that of fitness (something similar to health points
which reflect the ability to reproduce more effectively), and strategies play
the one-shot PD repeatedly for fitness points. Note that repetition does not
change the fact that the game keeps the characteristics of the one-shot PD.
As long as the rules of the system make sure the players have no possibility
of retaliating (for instance by not giving agents ways of identifying and
remembering other players) they play each game as if it was the first and
only time they played.
 211

Let us run the model. We will notice that the strategy of Defect, free-
riding on Cooperate (taking advantage of its kindness) and not losing too
many fitness points against other Defect strategies, will do quite well. De-
fect will do better on average than Cooperate. So far this confirms the
predictions of game theory. Now consider we introduce an extra element of
complexity in the model. When strategies are very healthy, having enough
fitness points, we allow them to produce offspring. This is where EGT starts
yielding considerably different results from game theory. If we program the
rules of the model so that the offspring remain within a short distance from
the parents, as happens in many biological systems, we will notice that local
pockets of Cooperate will start to grow, as Cooperate will have a good prob-
ability of playing against other Cooperate strategies and get more fitness
points on average. In this scenario, and depending on the variables regu-
lating spatial distribution and the fitness points attribution matrix, Defect
may be driven into extinction, or keep only a small population, surviving
in relative isolation on the borders of populations of Cooperate they can
exploit. A moral player does better that a rational player, and so coopera-
tion can become an Evolutionarily Stable Strategy. (See [4]for a detailed
explanation of the mechanics of standard EGT models).
In order to achieve the results described above, however, careful tweak-
ing of the variables is essential. We do get the results we want, but at the
expense of rigging the experience, making sure the model is set-up so that
nothing else can happen. This does not detract from the fact that it is good
news that such a set-up is possible at all. There are however many more
possible scenarios in which cooperation cannot evolve, due to the natural
advantages of free-riding. The model described is also abstracting away
quite a few important things. The characters of the plot are excessively
flat, for instance. These agents always cooperate or always defect, regard-
less of context. Calling them strategies is more of a metaphor, as they are
equivalent to their strategies, which means there is no process of decision.
It is useful to have discovered that the spatial distribution of coopera-
tors and defectors can be a deciding factor in cooperation dominating the
population, but surely we do not always or mostly interact with our kin.
The argument that was valid against kin selection can be again brought
in here. There are, however, other more sophisticated applications of EGT
that deal to this issue, such as Paul Danielson’s, which we will now examine.
212

6. A solution to the Prisoner's Dilemma?

The EGT model we will look at is inspired in David Gauthier's naturalistic
account of For Gauthier, morality is grounded on a principle of
rational choice: we cooperate, respecting a contract that limits our freedom,
because it is in our interest to do so. For instance, we accept that we are
forbidden to steal as that agreement protects us from others who could steal
from us. We trade freedom for security - an argument famously associated
with Thomas Hobbes." Reason alone, however, does not seem to be enough
to solve the one-shot PD, as we have seen. So how can we use the merits
of rational choice theory to solve the problem?
Gauthier's solution is that agents should apply rational choice not only
to the maximisation of preference but also to the selection of behavioural
disposition^.'^ They can then avoid the sub-optimal mutual defection out-
come by acquiring a disposition to behave altruistically simply because it is
rational to do so. After these behavioural dispositions have been acquired,
the agent is merely serving his own interest in being altruistic. He now
needs nothing but reason in order to solve the PD.
Gauthier's account is apparently compatible with the demand based on
common-sense that the label of genuine altruism can only be applied to
behaviour that is disinterested. As the agent has a genuine interest in the
welfare of his partner in the PD, he is not motivated by a material/strategic
personal gain, even if he will in fact stand to gain from his generosity. The
agent's behaviour can be self-interested without being selfish. Gauthier's
account seems then to roughly match our demands of a reasonable account
of altruism. Gauthier has been attacked over the years, however, for failing
to explain how these dispositions would work. Even with a disposition to
cooperate, the rational player knows that it is in his best interest to defect,
and this is what he will still do. This is where EGT comes in.
Peter Danielson used EGT models in an attempt to improve on Gau-
thier's work on altruism.21 For Danielson, the problem of mutual defection
in the one-shot PD comes from the fact that, within game theory, the
strategy of the rational agent always involves straightforward maximisa-
tion. This means the only rational course of action is for the agent to
maximise utility at each point at which there is a decision to be made. The
straightforward maximiser cannot but choose defection in the one-shot PD
as, at the moment of decision, defection maximises utility.
Danielson was interested in producing an agent that would not maximise
utility at each decision point but one that would maximise average utility,
 213

as only this agent could solve the PD. In order to produce it, Danielson asks
us to consider not the PD, but a slightly modified version of it that shares
most of its structure. In what Danielson calls the Extended Prisoner’s
Dilemma (XPD), one of the agents makes his move first. According to
game theory, this does not change the payoff matrix or the sub-optimal
result of mutual defection. The first player will still be better-off defecting,
regardless of what the second player does. The second player will also still
be better-off defecting, again regardless of what the first player has decided.
However, now it would be rational for the first player to cooperate if he could
somehow be sure that the second player would respond to cooperation and
only to cooperation with cooperation. It would also be rational for the
second player to use this strategy if proof that he will use it is the only
thing that can convince the first player to cooperate.
Danielson creates a constrained maximiser that is able to show his moral
engine to his opponent as evidence of his intention to respond to cooperation
with cooperation (and to defection with defection). The agent, a computer
program, gives access to his own code to the other agent, which can run it
to test it. As Danielson runs an EGT model in which there are such moral
agents, these will soon dominate the population, even if they start off with
just a small proportion of elements. Again, a moral player does better than
a rational player, and this time no careful tweaking of initial conditions is
necessary.
We have seen before that no form of discussion or promise-making could
result in a commitment in the game theory analysis of the PD. That is be-
cause talk is cheap. Regardless of what promises the agent would make to
cooperate, he would have no reason to keep them because he knows that, as
he will not play again against the same opponent, he cannot suffer retalia-
tion for lying. Knowing this, his opponent would have no reason to take his
promises seriously, and would be better off ignoring them altogether. Both
will defect. In Danielson’s XPD, however, commitment becomes possible
because there is no need for promises. There is in fact no real commitment,
in the psychological sense. If I say I behave like a human being because
I am a human being and present as evidence my healthy human body, I
am not just promising to behaving like a human being, I am proving I will
behave like one, by necessity. Similarly, the agent that plays first knows
that the other player will respond to cooperation with cooperation in virtue
of its moral make-up. It will be irrational for the first player to defect, as
the retaliation is inevitable and automatic - the equivalent to the cold war’s
doomsday device. It seems we have got to love the bomb.
214

7. The role of morality

According to LaCasse and Ross, Danielson’s work promises to provide the
answer to what they believe is an essential question: “What is the role
of morality?”22 In their paper, Morality’s Last Chance, they start from a
position of moral scepticism, doubting that any role can be found, and argue
that Danielson fails to satisfy, actually giving them tools to strengthen
the case against the existence of morality. An analysis of LaCasse and
Ross’s critique of Danielson will provide us with interesting insights on the
advantages and limitations of the uses of EGT in research on altruism.
For LaCasse and Ross, morality is certainly not something that exists
independently from human beings, as in Plato. They do not find Rawlsian
arguments, inspired in Kant’s claim that moral behaviour is an inescapable
consequence of being rational, particularly compelling. For them, if there
is such as thing as morality at all, it requires a kind of justification that
could only be found in contractarianism, in the tradition of the work of
David Gauthier. The central argument in Danielson’s argument, which
is in Gauthier’s tradition, is that a moral player does better (materially
speaking) than a rational player. LaCasse and Ross believe that, if the
claim turns out to be true, there is then a role for morality.
In order to test Danielson’s claim, they proceed to analyse Danielson’s
XPD from a conventional game theory approach. For them, according to
game theory, as player 2 reveals his strategy before player 1 makes his move,
player 2 should actually be represented as playing first. Player 1’s actions
are indeed fully determined by player 2’s strategy. If we look at it this
way, a straightforward maximiser does not do worse than a moral agent.
Indeed they do exactly the same. It seems, then, claim LaCasse and Ross,
that once we change the PD into the XPD, give agents the ability to apply
rational choice in choosing their behavioural dispositions (strategies) and
reveal their strategies to others, there is no further role for morality, as
reason alone can solve the dilemma.

8. Discussion
If LaCasse and Ross are right, Danielson’s work is a complete waste of
time. But let us consider what kind of argument would convince them they
were wrong. What kind of argument would they accept? Hume said that
“ought” cannot be derived from “is”, to mean that normative statements
(what is morally good) cannot be derived from prudential considerations
(what is advantageous). LaCasse and Ross seem to agree with Hume, when
 215

they refuse to accept Danielson’s explanation for how altruism can exist.
His “is” arguments cannot bridge the gap t o “ought”. However, they seem
to demand such impossible bridging arguments, since they consider that
the failure t o provide them can help them to prove morality has no role.
LaCasse and Ross seem to expect a normative account of morality to
be based on prudential explanations. This is indeed an impossible mission.
Nevertheless, failing it has no implications for the possibility of a natural-
istic account of altruism. If Hume was right in saying that reason is but a
slave of passions, such an account cannot be grounded on rational reasons.
According to Hume’s account of action, reason only job is to find the means
to ends defined by passions (desires). Emotions, then, are more likely ac-
ceptable candidates for the causes behind altruistic action, when we ask
for a psychological explanation. And when we ask for the foundations of
such psychological explanations, we will have to look a t what evolutionary
pressures produced such emotions, so as to complement Hume’s account of
motivation.
Economist Robert Frank offers us an account of altruism that deals with
both the psychological and the evolutionary aspects of the problem. For
Frank, emotions, like Danielson’s or Gauthier’s behavioural dispositions,
act as enforcers of commitment. This needs not be seen as a deviation
from rationality if we accept Hume’s minimal notion of reason, according
t o which it is rational t o do whatever we desire. We have expressed some
reservations regarding the fact that Hume’s account in incomplete, but
Frank’s, however, includes the essential explanation for why moral senti-
ments do not compromise the organisms’ viability.
Holding moral sentiments, for Frank, may be an evolutionary advantage,
as this can make the agent a preferred target for cooperation. The agent’s
sense of justice implies that he may seek revenge at any cost, when he
considers he has been treated unfairly. This will act as a deterrent t o
potential offenders. These mechanisms can only work if it is somehow
visible to others that the agent is a moral agent (perhaps because he sounds
and looks honest). It must also be true that mimicking the appearance of a
moral agent is quite costly, from an evolutionary point of view, or we would
all pretend to be honest without being it, enjoying the advantages without
incurring the costs.
While it is not easy to provide empirical proof to support these claims,
Frank’s story succeeds in providing an account for situations in which one
is generous even when one knows the other may have no chance of returning
one’s generosity. Holding moral sentiments may mean that, a t times, one
216

will engage in costly self-sacrifice for someone else’s sake. However, the
theory will remain compatible with Darwinist evolution as long as holding
moral sentiments is advantageous on average.
To be fair to Hume, we should note that he does include a crucial
component of a kind of non-intentional consequentialism in his account of
how desires are formed. The story goes as follows. We desire t o help some
rather than others because we have come to love them. But for Hume
we come t o love those who have something that interest us: a quality we
admire, fortune, influential friends. This is not because we are planning
to take advantage of such people. The love is genuine. It is simply in
our nature to love those we admire. After this genuine love is formed, it
will suit our purposes to help such people, as they will be in a position to
return more than we have gave them - only this advantage is now accidental.
Interestingly, Hume’s account of how we come to love can help t o solve a
major problem in Frank’s theory, as we will see.
Robert Frank goes one step too far when he argues that the selfish
person who, after reading his book, now knows that being an altruist is
advantageous, will start acting altruistically. This is the mistake Gauthier
has made and Danielson was trying to avoid. Acts of genuine altruism do
involve sacrifices, and if we accept Frank’s account, it is only advantageous
to be an altruist on average. Each individual act of altruism remains some-
thing to be avoided by a selfish person, regardless of whether he agrees with
Frank’s account. So Frank still leaves us in need of an explanation for how
we can come t o be altruists.
Evolution has made it possible for us to feel the emotions that make
genuine altruism possible, but we are still not born natural altruists. How
can these emotions be used in the right way? How can they serve our in-
terest? Hume’s account of how we come to love others can be part of it.
We love those who have something to offer us. But here we have an inter-
esting case in which genuine sacrifice leads to material advantage, and that
cannot always be the case. Genuine altruism can involve straightforward
loss, like the sacrifice of one’s own life for the sake of someone else’s welfare.
A r i ~ t o t l eoffers
~ ~ an interesting account of moral education, according to
which we come t o know what is right, as children, by doing the right thing.
We simply go through the motions, mechanically at first, and then end up
enjoying it. This is indeed how we come to like music, or sports: by listen-
ing to music, by doing sports, not by theorising about it. It could be the
way how we come to enjoy moral action.
So where does this leave Danielson’s work on altruism? His model shows
 217

that while cooperation may seem irrational at a certain level of analysis, it

may actually become rational at another, once we understand the details
of the mechanism that produces it. It is true that Danielson did cheat.
He did change the rules of the game: the XPD is simply not the PD. In
the standard PD, the players cannot commit, by an impossibility defined
by the rules. EGT also corrupts the essential nature of the one-shot PD.
It is true that in the models described, players do not have a memory of
what is happening and therefore cannot retaliate, but as the system will
hold such a memory under the form of physical location and respective
proportions of cooperators and defectors, punishment can come under the
form of a defector being eventually surrounded by defectors as it reproduces.
What Danielson did was to reframe the question, transforming the problem.
Perhaps expressing the problem of altruism by looking for a traditional
game theoretic solution to the prisoner’s dilemma was like asking what the
evolutionary advantage of having half an eyeball could be (half an eyeball
that will fall off its socket), on the route to a full eye. These are not
problems evolution had to deal with. It could be fine, then, to rephrase
a question that made little sense. Danielson’s approach is promising in
the sense that it can reveal that some problems that apparently can only
be solved by genuine altruism can actually be shown by EGT analysis to
be solvable by some form of reciprocal altruism. Davidson has not dealt
satisfactorily, however, with the issue of genuine altruism.
Peter Danielson does not include an explanation for how us, humans,
could show our lines of code to others, in order to prove our intentions
to cooperate. Is it the case that only computers can behave morally? It
would be possible to improve the model by including some extra details,
perhaps giving the agents a neural network and some kind of simulated
emotions. This would make it possible to study scenarios that could explain
why humans are rarely very good liars, for instance. Perhaps evolution
never had to solve commitment problems. It is not hard to conceive of
an evolutionary story in which bad liars were produced as an accident, as
a compromise between the advantages of being able to lie convincingly,
pretending to be a cooperator, and the advantages of communicating one’s
emotions effectively and automatically, as infants. This could mean the
efforts of EGT must be combined with those of developmental psychology
and neuroscience.
218

9. Conclusions
One important lesson to learn is that models of complex systems are not
at all immune to rationalistic assumptions. These can indeed dangerously
creep up unseen, covered up by the apparent realism of this methodological
approach. This seemingly down to earth, pragmatic nature of Evolutionary
Game Theory cannot make us forget its models are still heavily theory
laden.
Danielson’s story is a story of how to love the bomb, as it claims al-
truistic actions can be made rational if they can rely on a mechanism of
deterrence. We have seen that being moral cannot compromise a system’s
physical viability. The merit of the Prisoner’s Dilemma puzzle is that it
highlights this issue, a problem the Humean account of motivation does
not deal with directly. Acting altruistically therefore needs to be materi-
ally advantageous, at least on average. In search for the elusive connection
between morality and rationality, Danielson has followed Gauthier’s ap-
proach of reducing altruism to rational action. Danielson demands a direct
connection between physical viability and action, one which makes genuine
altruism impossible. If adequate evidence that genuine acts of altruism are
an illusion and can be explained in terms of pure rationality could be pro-
vided, we should be ready to accept it. It seems, however, that those who
are happy about reducing altruism to rationality do not do it because they
have found such evidence, but rather because they are confused by the fact
that altruistic acts paradoxically seem to serve the individual who performs
them. Although their intuition might be telling them the same story those
who defend in genuine altruism believe in, they reject it possibly on the
grounds that we may be fooling ourselves. In fact, we often do. Perhaps
we like to think we are being kind when we are actually always trying to
further our interests. But why would we be any happier for thinking we are
morally good if we are not? If we accept this happiness is a result of purely
cultural reasons, we are still left with no explanation for why culture would
favour such self-delusion. Keeping genuine altruism and explaining why it
is possible offers us a much more satisfactory story of human nature.
An extra cog is needed in the mechanisms of causality; one that some-
times seems to be turning the wrong way, and emotions are a good candi-
date for this. Danielson’s other interesting work in applied ethics has done
much to establish EGT as a powerful tool for understanding and predicting
human behaviour. However, EGT, in its present stage, is unable to offer
an adequate account of altruistic behaviour. Danielson may be right in ex-
 219

pecting deterrence to be the reason (foundation) why altruism exists, but

he should not expect it to be the reason (motivation) for why we act altru-
istically. Perhaps it is simply the case that some of our interests are in fact
the welfare of others, and that it is rational to act on them. Hume’s psy-
chological story can now become plausible and is confirmed and completed
by Frank’s and Aristotle’s account.
It is not possible to solve the single-shot Prisoner’s Dilemma with game
theory or evolutionary game theory. Only genuine altruism can solve it.
Deterrence may have shaped emotions, the mechanisms that allow com-
mitment, but there are surely cases in which it is not out of fear or con-
siderations of possible material advantages that we are generous or kind.
This is not being sentimental. It is important t o realise that although the
moral person performs altruistic acts for moral reasons, not self-interest,
he is not a fool. He will profit from cooperation and will refuse t o be ex-
ploited. Attempts t o free-ride will be punished. This means there are no
exceptions to the advantages of being a moral person, as not only it serves
the person who performs the altruistic acts when he deals with other moral
people, but also with those whose behaviour is more often governed by
self-interest. Mechanisms of reciprocity such as fear of retaliation are but
one of the reasons why we cooperate, and moral sentiments are an essential
complement t o such mechanisms. There is space for altruism within ratio-
nality. We should then love the bomb, only not because it motivates us t o
cooperate, but rather because the strategic virtues of deterrence are part
of what makes us want to act morally.

Acknowledgments
I conducted the research presented in this paper with a PhD scholarship
from the Portuguese Foundation for Science and Technology (FCT). I would
also like to thank Peter Goldie, Federica Frabetti and Marina Hasselberg,
for a number of interesting discussions on the topic of altruism

References
1. R. Frank Passions within Reason: the strategic role of emotions. Norton
(1988).
2. C. Darwin The Descent of Man and Selection in Relation to Sex. New York:
Appleton (1871)
3. F. Heylighen, Evolution, Selfishness and Cooperation. In: Journal of Ideas
Vol 2, NO 4, pp 70-84 (1992)
220

4. B. Skyrms (Ed.) Evolution of the Social Contract Cambridge University Press

(1996)
5. P. Kollock, Transforming Social Dilemmas: Group Identity and cooperation.
In: B. Skyrms (Ed.) Evolution of the Social Contract Cambridge University
Press (1996)
6. W. D. Hamilton, The Genetical Evolution of Social Behaviour I and 11. In:
Journal of Theoretical Biology, 7: 1-16, 17-32 (1964)
7. J. Maynard Smith, Group Selection and Kin Selection. In: Nature, 201: 1145-
1147 (1964)
8. J. M. Alexander, Evolutionary Game Theory. In: Edward N. Zalta(ed.), /tex-
titThe Stanford Encyclopedia of Philosophy (Summer 2003 Edition) URL =
~http://plato.stanford.edu/archives/sum2OO3/entries/g~e-evolutionary/~.
9. M. Nowak and K. Sigmund, Evolution of Indirect Reciprocity by Image Scor-
ing / The Dynamics of Indirect Reciprocity. In: Nature 437:1291 (2005)
10. R. Dawkins The Selfish Gene. Oxford University Press (1976)
11. R.L. "rivers, The Evolution of Reciprocal Altruism. In: /textitQuarterly
Review of Biology, 46: 35-57 (1971)
12. R. Axelrod and W. D. Hamilton, The Evolution of Cooperation. In: Science,
211: 1390-96 (1981)
13. S. Okasha, Biological Altruism. In: Edward N. Zalta(ed.) The Stan-
ford Encyclopedia of Philosophy (Summer 2005 Edition) (2005) URL =
~http://plato.stanford.edu/archives/sum2OO5/entries/altruism-biological/~.
14. M. Osborne, An Introduction to Game Theory, Oxford University Press
(2003)
15. D. Gauthier, Morals by Agreement, Oxford University Press (1986)
16. T. Hobbes, Leviathan, Oxford World's Classics (1998)
17. D. Hume, A Treatise of Human Nature Oxford University Press (1978)
18. D. Hume, Enquiries Concerning Human Understanding and Concerning the
Principles of Morals Oxford University Press (1975)
19. D. Gauthier and R. Sugden (eds.), Rationality, Justice and the Social Con-
tract, Harvester Wheatsheaf (1993)
20. P. Danielson (ed.) Modelling Rationality, Morality and Evolution, Routledge
(1998)
21. P. Danielson, Artificial Morality, Routledge (1992)
22. C. LaCasse and D. Ross, Morality's last chance. In: P. Danielson (ed) Mod-
elling Rationality, Morality and Evolution, Routledge (1998)
23. J. Rawls, A theory of justice, Harvard University Press (1971)
24. Crisp, Roger (ed.), Nichomachean Ethics, Cambridge University Press (2000)
PHYSICAL COMPLEXITY AND COGNITIVE EVOLUTION*

PETER JEDLICKA
Institute of Clinical Neuroanatomy
J. W. Goethe University
fiankfurt/Main, Germany
E-mail: jedlicka@em.uni-frankfurt.de

Our intuition tells us that there is a general trend in the evolution of nature, a
trend towards greater complexity. However, there are several definitions of com-
plexity and hence it is difficult to argue for or against the validity of this intuition.
Christoph Adami has recently introduced a novel measure called physical com-
plexity that assigns low complexity to both ordered and random systems and high
complexity to those in between. Physical complexity measures the amount of in-
formation that an organism stores in its genome about the environment in which
it evolves. The theory of physical complexity predicts that evolution increases
the amount of ‘knowledge’ an organism accumulates about its niche. It might be
fruitful t o generalize Adami’s concept of complexity to the entire evolution (includ-
ing the evolution of man). Physical complexity fits nicely into the philosophical
framework of cognitive biology which considers biological evolution as a progressing
process of accumulation of knowledge (as a gradual increase of epistemic complex-
ity). According t o this paradigm, evolution is a cognitive ‘ratchet’ that pushes
the organisms unidirectionally towards higher complexity. Dynamic environment
continually creates problems t o be solved. To survive in the environment means t o
solve the problem, and the solution is an embodied knowledge. Cognitive biology
(as well as the theory of physical complexity) uses the concepts of information
and entropy and views the evolution from both the information-theoretical and
thermodynamical perspective. Concerning humans as conscious beings, it seems
necessary to postulate an emergence of a new kind of knowledge - a self-aware and
self-referential knowledge. Appearence of selfreflection in evolution indicates that
the human brain reached a new qualitative level in the epistemic complexity.

1. Introduction
Our intuition suggests that there is a general trend in the evolution of na-
ture, a trend towards greater complexity. According t o this view evolution
is a progressive process with Homo sapiens emerging a t the top of the life’s

*This work is supported by Graduiertenkolleg: Neuronale Plastizitat - Molekule, Struk-

turen, Funktionen (J. W. Goethe University)

221
222

hierarchy. An alternative opinion is that there is no trend in evolution and

we are just one of many leaves on the evolutionary tree. There are several
definitions of complexity and hence it is difficult to argue in a quantitative
way for or against the validity of each of these views. In my article, I will be
focusing on a novel measure of biological complexity (physical complexity)
that has been proposed by Christoph Adami.

2. Kolmogorov - Chaitin complexity and biology

One well-known definition of complexity is the Kolmogorov-Chaitin com-
plexity (introduced independently by Kolmogorov ( l ) , Chaitin (2) and
Solomonoff (3)). It represents an algorithmic measure of system’s random-
ness (4, 5). It defines the complexity of an object (or a process) by the size
of the smallest program for calculating it. For example, a long sequence
of digits with a regular pattern (e.g. 01010101...) can be compressed to a
much shorter description/program (“repetition of 01”) and hence it has a
small amount of complexity. In contrast, an infinite random sequence of
digits “19255324193625168147...” has no intrinsic structure and cannot be
compressed at all. The only way to express the sequence is to enumerate all
the digits that it consists of. Thus, Kolmogorov - Chaitin’s definition im-
plies that maximum complexity is ascribed to a completely random process
that is algorithmically incompressible (Fig. 1, c.f. Ref. 6).

complexity

0 randomness 1

Figure 1

However, although useful for the theory of computation, Kolmogorov-

Chaitin complexity does not satisfy our expectations inspired by biology
that most complex organisms are neither completely regular nor utterly
 223

random but lie between the two extremes of order and randomness (7).
What makes living systems complex is the interplay between order and
randomness (8). This principle has been supported by recent research on
biological networks (9). Complex networks of cellular biochemical path-
ways have neither regular nor random connectivity. They have a so called
scale-free structure (10). A scale-free network contains a small number of
hubs - major nodes with a very high number of links whereas most nodes in
the network have just a few links. “Scale-free” means that there is no well-
defined average number of connections to nodes of the network. In case of a
biochemical network, nodes and links represent molecules and their chemi-
cal reactions, respectively. In such a molecular network, hubs are important
molecules that participate in a large number of interactions (e.g. CAMP,
H20) in comparison to other molecules that take part in a few biochemical
signaling paths (11). An important consequence of the scale-free archi-
tecture is robustness against accidental failures of minor nodes/links and
vulnerability to disfunctions of major hubs. (Note that this may bring inter-
esting insights into the molecular pathophysiology of various diseases.) Sur-
prisingly, recent mathematical analysis of various scale-free networks (in-
cluding biochemical signaling networks) has revealed self-similarity (fractal
pattern) in their structure (12). Self-similarity is a typical property of sys-
tems that are on the edge of order and chaos. Such a critical state (with
a tendency to phase transitions) might be useful for optimizing the per-
formance of the system (13). To sum up, we need a measure which would
capture the complexity of dynamical systems that operate between order
and randomness.

3. Physical complexity
Christoph Adami (14) has recently introduced a novel measure called physi-
cal complexity that assigns low complexity to both ordered and random sys-
tems and high complexity to those in between (Fig. 2). Physical complexity
measures the amount of information that an organism stores in its genome
about the environment in which it evolves. This information can be used
to make predictions about the environment. In technical terms, physical
complexity is a shared (mutual) Kolmogorov complexity between a sequence
and a n environment (for mathematical equations see Ref. 14). Informa-
tion is not stored within a (genetic) sequence but rather in the correlations
between the sequence and what it describes. By contrast, Kolmogorov-
Chaitin complexity measures regularity/randomness within a sequence and
224

therefore fails to address its meaning in the environment.a Information is

a relative quantity - always about something, in case of the genome it is
about the niche that an organism lives in (16, 17). The theory of physical
complexity uses two central concepts of Shannon’s theory of information:
entropy and information. Entropy is a measure of potential information,
it determines how much information a sequence could hold. Entropy of a
sequence can be compared to the length of a tape, and information to the
length of a tape portion containing recordings (14). Physical complexity has
the advantage that it is a practical measure, because entropy of an ensem-
ble of genetic sequences can be measured by estimating the probabilities of
finding the genetic sequences in the environment.

maximum I

0 randomness 1

Figure 2 .

Evolution is a slow process and therefore obtaining exact data is dif-

ficult. How could we test the hypothesis of the evolutionary complexity
increase? An interesting option is to study digital evolution (18). Digital
organisms (digitalia) are self-replicating computer programs (sequences of
instructions) that mutate and compete for space and computer time (19).
Thus, digital evolution occurs since the three conditions of evolution are
met: replication, variation (mutation) and competition/selection (due to
differential fitness/replication rate). Digital organisms have a much shorter
generation time (seconds) and their physical complexity can be measured
exactly. In this case, physical complexity is the information in a digitalia

aInterestingly, Chaitin has proposed his own definition of life based on Shannon’s concept
of mutual information. See Ref. 15.
 225

program that is vital to organism’s survival (replication). Of course, digital

evolution is a highly simplified model of biological evolution and this cre-
ates the possibility of artifact conclusions or generalizations. On the other
hand, simplification is at the same time the greatest strength of simulated
evolution experiments since it allows us to find and see the forest (general
principles of evolution), not just the trees.
Interestingly, Adami’s experiments on digital organisms revealed a
steady trend toward increased physical complexity in their evolution within
a fixed environment (16). According to his theory and simulation data, evo-
lution increases the amount of ‘knowledge’ an organism (or a population
of organisms) accumulates about its niche. Since in this context entropy
is a measure of potential information, biological evolution leads to a de-
crease of entropy. Natural selection turns an e m p t y tape into a filled tape:
entropy into information. The information-filled part of the tape is impor-
tant for the survival of an organism in the environment. If the selective
advantage fixes a beneficial mutation within the population, the amount of
information (physical complexity) increases. Adami views natural selection
as a unidirectional filter (see below the discussion of ratchetting) that lets
information enter the genome, but prevents it from flowing out.
Adami’s simulations were done in a fixed environment. He discusses
that a rapidly changing environment as well as several other factors that
were not included in his experiments (high mutation rates, sexual recombi-
nation, co-evolution between species occupying different niches) may lead
to complexity declines. However, these factors are ambivalent - sometimes
they help, rather than hinder, the evolution of complexity (14). Therefore
he argues that there are good reasons to expect that the theory of phys-
ical complexity will reveal an overall trend towards higher complexity in
biological evolution.

4. Physical complexity and the concept of cognitive biology

It might be fruitful to generalize Adami’s concept of complexity (which
has been primarily thought to describe the evolution of genome) to the
entire evolution including the evolution of man (Fig. 3). Physical complex-
ity fits nicely into the philosophical framework of cognitive biology which
considers biological evolution as a progressing process of accumulation and
application of knowledge, i.e. as a gradual increase of epistemic complex-
i t y (20, 21). Cognitive biology provides a broader philosophical frame for
Adami’s approach since the central idea of his theory of physical complexity
226

is relating complexity to the system’s ‘knowledge’ (information) about its

environment. Physical complexity can be viewed as a special case of epis-
temic complexzty. This becomes clear if we look at many common features
of Adami’s ideas and cognitive biological ideas. According to the paradigm
of cognitive biology, evolution as a whole is a cognitive ‘ratchet’ that pushes
the organisms unidirectionally towards higher complexity. Epistemic ‘ratch-
etting’ operates at all hierarchical levels, from molecules to societies (20).
Dynamic environment continually creates problems to be solved (i.e. each
niche is a solution problem). To survive in the environment means to solve
the problem, and the solution is an embodied knowledge.

The Concept of Physical Complexity

w.krl.caltech.edu/-adami

1
generalization to the
entire evolution including
the evolution of man

The Paradigm of Cognitive Biology

w.fns.uniba.sk/-kbi/kovlablprincip.htrn

Figure 3.

‘Ratchetting’ is a general phenomenon that has been usually described

in the context of thermodynamics. Cognitive biology acknowledges that
progress in evolution has thermodynamical reasons (20). Both cognitive
biology and the theory of physical complexity use the concepts of infomna-
t i o n and entropy and strongly emphasize that there is a close connection be-
tween thermodynamical thinking and information-theoretical approaches.
By thermodynamical reasoning we can identify a ‘differentiation of a system
from environment’ as a dissipative ‘movement from thermodynamic equilib-
rium’ (22). Biological evolution creates organisms with an ever increasing
amount of embodied ‘knowledge’ and with an ever farther distance from
thermodynamical equilibrium (20). Living systems are far away from the
equilibrium because of the information stored in their genomes (17). John
 227

Polkinghorne (23) predicts that “by the end of the twenty first century,
information will have taken its place alongside energy as an indispensable
category for the understanding of nature.” The paradigm of cognitive bi-
ology points t o the same direction.

5. Self-referential cognition - a ‘Big Bang’ of complexity in

cognitive evolution?
Concerning humans as conscious beings, it seems necessary t o postulate
an emergence of a new kind of knowledge - a self-aware and self-referential
knowledge. We not only know, we know that we know. We not only
have informations but we possess informations about informations (meta-
informations) as well. Cognitive biology itself is an example of the self-
referential knowledge. It is a theory (knowledge) about accumulation of
knowledge. Appearence of selfreflection in evolution indicates that the hu-
man brain reached a new qualitative level in the epistemic complexity. One
may speak of a cognitive ‘Big Bang’. The expression of this cognitive ‘ex-
plosion’ that occured in human species may be found in the development of
science, art, culture and religion. Many writers have noticed this interest-
ing phenomenon. Gilbert Keith Chesterton wrote that “man is not merely
an evolution but rather a revolution.” Since the most primitive men drew
pictures and the most intelligent monkeys don’t, “art is the signature of
man” (24).
What are the mechanisms of self-awareness? Can we describe the emer-
gence of self-referential knowledge in the mathematical language of em-
pirical science? Is it possible to extend the theory of physical complexity
and formalize the meta-knowledge which is characteristic for our species?
Interestingly, Godel’s theorem shows a basic limitation of formal methods
due to self-reference (25). His results demonstrate that it is impossible
to achieve a complete knowledge of a (nontrivial) formal system with the
means available within the system. To fully understand the formal system
one must go outside the system (otherwise one falls into inconsistencies).
It is interesting to apply this principle to human cognition. According t o
the computationalism, our mind basically functions as a certain (unknown)
computational algorithm. However, a Godelian line of thinking would sug-
gest that we are not able to find this out because we cannot step out of the
algorithm. If this reasoning is valid then there are two possibilities: either
we are no mechanical machines or we are machines but we cannot know
it because of the limitation imposed by computational selfreference. It is
228

the limitation of an observer in “Cartesian prison” (26). In a paraphrase

of Stephen Hawking’s words (27): we don’t see the mind from the outside.
Instead, we and our models, are both part of the mind we are describing.
Thus, a theory of human cognition is self-referencing, like in Godel’s theo-
rem. One might therefore expect it to be either inconsistent, or incomplete.
So it seems to be improbable that a complete and consistent formalization
of human epistemic complexity is possible (28).
Based on Godelian and other arguments, some authors argue that the
evolutionary leap from ‘pure’ cognition to self-referential cognition might
have been governed by some novel noncomputational principles. (It should
be noted that Godelian arguments are highly controversial and the inter-
esting debate about their validity continues. See e.g. Ref. 29, 30, 31.) To
account for the emergence of consciousness, new physical principles (32)
or known physical (quantum) principles operating in the brain in a special
way have been suggested. An interesting quantum brain proposal was pub-
lished by Jeffrey Satinover (33) that combines findings in computational
neuroscience, nonlinear dynamics and quantum physics. Since indetermin-
ism observed in quantum events is sometimes interpreted as a fundamental
time asymmetry (34), it is tempting to speculate that if quantum brain
hypotheses contain some truth then a deeper link might connect thermody-
namic, cosmological and epistemic ‘ratchetting’ processes (for introduction
to quantum ‘ratchetting’ see e.g. Ref. 35).
If it is not possible to entirely reduce emergent human consciousness
to neuronal computation (28), then an interesting philosophical question
arises, namely what kind of emergence is responsible for it. Mark Bedau
distinguishes nominal, weak and strong emergence (36, see also Ref. 37). If
a new kind of causation powers (so called ‘downward causation’) has been
brought about by the emergence of selfreferent cognition, then we can call
it a strong emergence, according to Bedau’s definitions. The existence of
downward causation belongs to intensely discussed topics in the philosophy
of science. I a recent article, George Ellis (38) describes a hierarchical view
of the complexity of the universe with the autonomously effective mind at
the top of it: “the higher levels in the hierarchy of complexity have au-
tonomous causal powers that are functionally independent of lower-level
processes. Top-down causation takes place as well as bottom-up action,
with higher-level contexts determining the outcome of lower-level function-
ing ...” These words resemble the Satinover’s hypothesis (33) of a hierarchy
of nested networks (similar to Hofstadter’s “tangled hierarchy”, Ref. 39).
An attractive speculation is that the strong emergence might be related to
 229

quantum phenomena since quantum theory postulates that quantum events

cannot be reduced t o some form of bottom-level microscopic law governing
their outcomes (41, 40). In addition to information, the downward causa-
tion (or - in Aristotelian terms - the inside-out causation, Ref. 42) seems t o
be a very interesting topic for discussion and research in complexity science
in the next years.

Acknowledgements
T h e author thanks Dr. Stephan W. Schwarzacher for valuable discussions
and for reading the manuscript

References
1. A.N. Kolmogorov, Three approaches to the quantitative definition of infor-
mation. Problems Information Transmission 1:l-7 (1965)
2. G.J. Chaitin, On the length ofprograms for computingfinite binary sequences.
J Assoc Comput Mach 13:547-569, (1966)
3. R.J. Solomonoff, A formal theory of inductive inference, part 1 and 2. Inform
Contr, pp. 1-22, 224-254 (1964)
4. G.J. Chaitin, Computers, paradoxes and the foundations of mathematics. Sci
Am 90:164-171 (2002a)
5. G.J. Chaitin, On the intelligibility of the universe and the notions of sim-
plicity, complexity and irreducibility. In: Grenzen und Grenzberschreitun-
gen, XIX. Deutscher Kongress fur Philosophie (Hogrebe W, Bromand J, ed)
Akademie Verlag, Berlin, pp. 517-534 (2002b)
6. J.P. Crutchfield, When Evolution is Revolution Origins of Innovation. In:
Evolutionary Dynamics Exploring the Interplay of Selection, Neutrality, Ac-
cident, and Function (Crutchfield JP, Schuster P, ed) Santa Fe Institute Se-
ries in the Science of Complexity, Oxford University Press, New York (2002b)
7. J.P. Crutchfield, What lies between order and chaos? In: Art and Complexity
(Casti J, ed) Oxford University Press (2002a)
8. P. Coveney, R. Highfield, Frontiers of Complexity: The Search for Order in
a Chaotic World, Ballantine Books (1996)
9. A.L. Barabasi, Z.N. Oltvai, Network biology: understanding the cell’s func-
tional organization. Nat Rev Genet 5:lOl-113 (2004), www.nd.edu/ networks
10. A.L. Barabasi, E. Bonabeau, Scale-free networks. Sci Am 288:6(r69 (2002)
11. D. Bray, Molecular networks: the top-down view. Science 301:1864-1865
(2003)
12. C. Song, S. Havlin, H.A. Makse, Self-similarity of complex networks. Nature
4331392-395 (2005)
13. S.H. Strogatz, Complex systems: Romanesque networks. Nature 433:365-366
(2005)
14. C. Adami, What is complexity? Bioessays 24:1085-1094 (2002)
230

15. G.J. Chaitin, Toward a mathematical definition of “life”. In: The M m i m u m

Entropy Formalism (Levine RD and Tribus M, ed), pp. 477-498, MIT Press
(1979)
16. C. Adami, C. Ofria, T.C. Collier, Evolution of biological complezity. Proc
Natl Acad Sci U S A 97:4463-4468 (2000)
17. C. Adami, Information theory in molecular biology. arxiv.org/abs/q-
bio.BM/0405004 (2004)
18. The Digital Life Laboratory at the California Institute of Technology:
http://dllab.caltech.edu
19. C.O. Wilke, C. Adami, The biology of digital organisms. Trends Ecol Evol
17~528-532 (2002)
20. L. Kovac, Fundamental principles of cognitive biology. Evol and Cognition
6~51-69 (2002)
21. H. Kuhn, Origin of life and physics: Diversified macrostructure - induce-
ment to form information-carrying and knowledge-accumulating systems. J
Res Develop 32: 37-46 (1988)
22. 1. Prigogine, The End of Certainty. Free Press (1997)
23. 3. Polkinghorne, The person, the soul, and genetic engineering. J Med Ethics
30~593-597 (2004)
24. G.K. Chesterton, The everlasting man. Ignatius Press (1993, orig. 1925)
25. K. Godel, Uber formal unentscheidbare Satze der Principia Mathematica und
venuandter Systeme I. Monatshefte fur Mathematik und Physik 38:173-198
(1935)
26. K. Svozil, On self-reference and self-description. In: Functional Models of
Cognition, (Carsetti A, ed) Kluwer, Dordrecht, p. 18S197 (2000)
27. S. Hawking, Godel and the end of physics. Lecture given on Texas A&M’s
College Station campus. www.damtp.cam.ac.uk/strtst/dirac/hawking/
28. P. Jedlicka, Neuroethics, reductionism and dualism. Trends Cogn Sci 9:172
(2005)
29. B. Giolito, Monographica: John Lucas against mechanism. Etica & Politica,
(5) (2003) http://www.univ.trieste.it/ etica/2003-l/index.html
30. M. Redhead, Mathematics and the mind. Brit J Phil Sci 55:731-737 (2004)
31. S. Bringsjord, A. Shilliday, J. Taylor, B. Bram van Heuveln, J. Baumes,
K. ROSS, Y. Yang, A new Giidelian argument for hypercomputing
minds based on the busy beaver problem. J Comput Appl Math (2005)
http: //kryten.mm.rpi.edu/laihyper 11.pdf
32. R. Penrose, Shadows of the Mind: A Search for the Missing Science of Con-
sciousness, Oxford University Press (1994)
33. J. Satinover, Quantum Brain, John Wiley & Sons (2001)
34. D. Zeh, The Physical Basis of The Direction of Time. -.time-direction.de
(2004)
35. M. Brooks, Quantum clockwork. New Scientist 165:28 (2000)
36. M. Bedau, Downward causation and the autonomy of weak emergence. Prin-
cipia 6:5-50 (2003)
37. P. Humphreys, How Properties Emerge. Philosophy of Science 64:l-17 (1997)
38. G.F. Ellis, Physics, complexity and causality. Nature 435:743 (2005)
 231

39. D.R. Hofstadter, Godel, Escher, Bach: A n Eternal Golden Braid, Basic
Books (1979)
40. F.M. Kronz, J.T. Tiehen, Emergence and quantum mechanics. Philosophy of
Science 69:324-347 (2002)
41. K. Svozil, Computational universes. In: Space Time Physics and Ractality
(Weibel P, Ord G, Rossler OE, ed) Springer Verlag, Wien, New York, p.
144-173 (2005)
42. M.J. Dodds, Top down, bottom up or inside out? Retrieving Aristotelian
causality in contemporary science. In: Science, Philosophy and Theology
(O’Callaghan J, ed) South Bend (1997)
 INFORMATIONAL DYNAMIC SYSTEMS: AUTONOMY,
INFORMATION, FUNCTION

WALTER RIOFRIO
Neuroscience and Behawiour Division
Universidad Peruana Cayetano Heredia.

The main purpose of this paper is the conceptual exploration into the pre-biotic
world, a world that we consider t o be made up of a relatively continuous sequence
of systems directed toward the origin of the first life forms. In a tentative way, we
propose the ‘Informational Dynamic System’. This type of system would constitute
the ancestral system that definitively opened the door t o the pre-biotic world.
Also, it explains, in a naturalistic way, the physical emergence of functions and
information.

1. Introduction
The so-called self-organization phenomena feature multiple types of dy-
namic systems, and we find them in different sectors of our reality. For
example, we have the BCnard cells and the Belousov-Zhabotinskii reactions
(or Brusselator when this type of chemical reaction is computer simulated).
It was Ilya Prigogine who, working on these chemical processes, coined the
term “dissipative structures”, allowing us to understand the emergence of
‘novelties’ in the systems that are far from thermodynamic equilibrium
state. From the thermodynamic point of view, these systems are open sys-
tems (they exchange energy and/or matter with their environment), and the
self-organization phenomena are produced due to the continuous changes
in the local interactions of the components that form the dynamic system
111.
The Belouzov-Zhabotinskii (B-Z) reaction is an oscillating chemical re-
action that is autocatalytic. In other words, products of the reaction in-
crease the rate of their own production. In this process, different steady-
states and periodic orbits can be observed. The Cerium oxides, Ce(II1) and
Ce(IV), are the chemical compounds in solution that produce an oscillation
dynamic in the reaction. The periodic increase in the concentration of one
of the chemical compounds and the subsequent decrease of the other is seen

232
 233

at the macroscopic level as periodic changes and rhythms of color [2].

What this type of self-organized system has in common with others of
the same type is that the continuous entropy production in the system is
actively exported (dissipated) to the outside of the system.
These types of self-organized dynamic systems reveal to us certain
generic properties that can be applied to the rest of the self-organizational
phenomena [3-5).
(1)The appearance of a global order starting from local interactions. (2)
The control of the organization is distributed throughout the whole system.
(3) They have a relative “insensitivity” to environmental changes; they are
robust. (4) Self-organized phenomena are essentially non-linear. (5) Self-
organization contains feedback processes - positive and negative. (6) The
degree of coordination or correlation between the system components is
associated with the degree of organization in the self-organized dynamic
system. (7) Self-organization is fundamentally caused by a dynamic inside
the system; because of that, the system possesses “organizational closure”.
(8) There are additional properties that begin to emerge in self-organized
systems as the internal dynamic becomes more complex.
The interest of our work will be centered on the emergence of certain
properties in determined, self-organized systems: those that we find in the
time before the appearance of the first life forms.
We will propose an initial working hypothesis with the purpose of con-
tributing to the discussion and improving the understanding or clarification
of what more than likely happened in the distant past, in the transition from
being inanimate to being animate.
Therefore, we will be occupied with exploring the nature of the local
interactions that made it possible for the appearance of the initial forms
of dynamic systems that had the global property of being an autonomous
agent [6-81.
Our strategy will be to approach the problem from a bottom-up per-
spective in naturalist terms. Consequently, we will make explicit references
in our arguments to the ways in which a dynamic organization might be
produced in systems that use matter and energy components governed by
the laws of physics, chemistry, and physical chemistry.
With this in mind, we will put forward the possibility of the emergence
of other properties that will, gradually, provide the dynamic systems we are
talking about with a character that is more similar to living phenomena.
These systems will continue to be governed by the abovementioned laws;
however, these new properties will begin to participate actively in its dy-
234

namic organization such that they would allow the emergence of the dawn
of an open-ended evolution (biological evolution).
Our conceptual proposal furthermore has two additional ends: the con-
struction of arguments that support the possibility and the necessity of
proposing one definition of information and one definition of function in
naturalist terms. In other words, we defend that both notions are not
purely thought exercises, whether in abstract, formal, or epiphenomena1
terms, or that for their definition they need some type of adscription im-
posed from the human world or from that of the human-generated devices.
In this work, we put forward a type of self-organized dynamic system
that we believe to be the system that opened the doors of the pre-biotic
world, and because of that, will begin the step from the world of the inan-
imate to the world of living systems: the Informational Dynamic System.

2. Some preliminary reflections

We consider it to be reasonable to postulate theoretic constructs that can
provide us insight into the nature of the dynamic organization that those
preparatory systems must have possessed for the future emergence of living
systems. We also believe that the origin of pre-biotic systems already carry
with them a certain level of complexity in their dynamic organization.
Put in different terms, to understand the conditions that make the ap-
pearance of living forms possible is to understand that their most basic
properties were present at the origin of the pre-biotic world. So, Informa-
tional Dynamic Systems already have in their organization those minimum
capacities that are present in whatever living system and are expressed at
levels of increasing complexity as we advance in evolution.
We can state that the self-organized type of system that is related to
living phenomena is one of a chemical nature. Hence, the normativity that
encompasses chemical reactions and the physical chemical requirements will
occupy a special place in our discussion.
In addition, it is feasible to envision the pre-biotic world made up of
a series of systems that are related in more or less successive stages. In
each step, we would find systems with certain capacities that they had
incorporated from previous systems. In turn, they would explore different
strategies for improving what they already had and would have the chance
of producing the emergence of new capacities that together with the existing
ones would prepare for the emergence of the first living systems.
Before the genesis of pre-biotic systems, we find different, self-organized
 235

systems that should gradually carry us to this new form of existence. It

must be pointed out that even though the purpose of our work is not to
discuss this prior time, it is possible for us to make certain clarifications
about it.
These very simple, inorganic, chemically composed, self-organized sys-
tems were found in extremely specific regions of the planet. The Earth’s
planetary conditions during those epochs make it possible to hypothesize
the existence of the first forms of chemical self-organization on the surface
of certain minerals [9-131 or in regions very close to volcanoes [14-181.
What must be stated clearly is the definite impossibility of trying to
find a pre-biotic system as we have visualized in our proposal. In the first
place, after so long a time since their disappearance, all traces of them
would have vanished as well. In the second, their “building blocks” must
have been metabolized by other systems in the pre-biotic world as well as
by the first living ones.
We also want to explain that when referring to the origin of the pre-
biotic universe, it is illogical to speak in terms of a Darwinian evolution
since neither DNA nor RNA had appeared, and there was nothing at all
like “genetic material” in existence. These emerged in the far distant future
of the history of the pre-biotic world. Therefore, we cannot speak of fitness
and other concepts connected with biological evolution. There is just open-
ended evolution - a type of evolution whose components are related to the
idea of evolution by natural selection.
Nevertheless, it is possible for us to refer to the capacities that permit
the gradual building of the structure and dynamic organization of those
systems in terms of their internal properties.
Therefore, at some moment, future generations of the different popula-
tions of these systems will possibly be ‘seen’ by natural selection [19-211.
If we want to suggest some reference to the succession of changes in the
dynamic systems, those that are strongly linked to the emergence of life,
we could speak of a certain proto-evolution. An example of such could be
a variant of the evolvability notion applied to the pre-biotic world [22-261.

3. The Informational Dynamic Systems

We place our theoretical reflections in the distant past, and this era tell
us that the types of chemical constituents we might find available on the
Earth would be incredibly simple chemical compounds, mainly inorganic,
yet perhaps a very small amount of organic ones. We might even consider a
236

possible abiogenic synthesis of such compounds like the thioester [27]. The
thioester compounds are those that are formed when sulfhydryl, an organic
group that contains a Sulfur-Hydrogen bond (R-SH), joins with another
compound called carboxylic acid (R’-COOH). During the chemical reaction,
a water molecule is released, forming this way the thioester compound (R-
S-CO-R’).
Taking into account the current understanding about what the condi-
tions of primitive Earth could have been like and meteor impacts during
this time period, it is possible to theorize the feasibility of the existence of
amino acids and carboxylic acids. Furthermore, the fact of massive volcanic
activity makes it possible to theorize the existence of sulfhydryl [28-321.
For the purposes of the work, we set a line of demarcation around the
origin of pre-biotic systems: it would be a type of system that has managed
a certain degree of autonomy by its own, self-maintained organization dy-
namic. Therefore, it is a type of dynamic system much more robust than
the self-organized system we mentioned at the beginning, which we will call
a “strict sense’’ self-organized system.
This opens up space for the possibility of returning to the most robust
and complex dynamic organization in the continuous succession of a type
of pre-biotic system that will end up producing the first life forms.
In that sense, we consider that our hypothetical informational dynamic
system provides an explanation for the emergence of that organizational
dynamic, and with it, allows us to understand the emergence of the global
property known as autonomy.
Generally speaking, self-organized systems are systems far from thermc-
dynamic equilibrium. BCnard cells, Belousov-Zhabotinskii reactions, living
systems, and pre-biotic systems are ones that are far from thermodynamic
equilibrium.
One difference that is crucial to our argument is that the first two exam-
ples are systems far from equilibrium because of external causes: the matter
and energy requirements that their processes needs invariably depend upon
the environmental conditions.
On the contrary, the other two cases maintain themselves in nonequi-
librium conditions for reasons that are systemically intrinsic. With this
statement, let us begin our study.
The correlation among processes is an expected phenomenon when a
system is in far from equilibrium conditions [33, 341. One important step
to the pre-biotic world is the appearance of a correlation between endergonic
and exergonic processes. The former needs energy requirements, whereas
 237

the latter produces energy.

The existence of compounds with “high energy” bonds significantly
favors this correlation between the abovementioned processes, forming a
micro-cycle that is capable of generating work (chemical work).
The inorganic pyrophosphate and/or a simple thioester could have been
the ancestral “energy currency” molecules of these types of chemical bonds.
In current cells, we find that the thioesters are bound up with the esters
formation, including those we find in complex lipids. We also find them
actively participating in the synthesis of peptides, fatty acids, sterols, ter-
penes, porphyrins, and other compounds.
With this last reference, we have just established the feasibility of main-
taining an adequate energy interchange between the exergonic-endergonic
cycle and the production of compounds (accomplishment of a chemical
work) that have much to do with the formation (and reparation) of a proto-
membrane (Amphiphilic polymers) [35, 361.
In this way, the informational dynamic systems manage to keep them-
selves in nonequilibrium conditions because there are two fundamental re-
quirements within its dynamic organization. Another way of putting it is
that in the first place, we would have the exergonic-endergonic cycle made
possible by a high energy bond carrier compound that as a group is a pro-
cess with the capability of producing work. In the second place, we would
have the protoplasmic membrane related in some way to the compound
acting like the “energy currency” molecule.
As we are visualizing the possible interaction of these types of processes,
later on, we will see how it is that they allow certain initial degrees of
autonomy with respect to the ever changing environmental conditions.
Another important topic to cover is what we refer to as the possibility
of forming oligomeric compounds by noenzymatic condensation. Much has
been said concerning abiotic chemistry in regards to this theme and until
recently the results were very limited for abiotic polymerization reactions
starting from their monomers - whether amino acids or nucleotides.
Nevertheless, a group of researchers has recently proven, experimentally,
that ‘carbonyl sulfide (COS)’ - a very simple gas formed from volcanic
activity and doubtlessly present in the pre-biotic era - intervenes in short-
chain peptide condensation.
In short, the molecular reaction involves the bond of the COS to the
amino acid by the amino group, producing afterwards a cyclical compound
called amino acid N-carboxyanhydride (NCA). The NCA will bond with
another amino acid and a dipeptide will form, freeing a COz molecule [37].
238

The results of scientific research demonstrate to us the complete possi-

bility of abiogenic short-chain peptide synthesis in very similar conditions
to those existing in the pre-biotic times. With these results, it is possible
to propose in a rational way that short-chain peptides could have man-
aged to attain a sequence of interdependent reactions: the beginnings of
protometabolism.
The three groups of processes that we have presented bring us directly
to the dynamic organization that constitutes the informational dynamic
system.

3.1. Organization of the Informational Dynamic System

In order to have the capacity of being more robust than the “strict” self-
organized systems and to be able to have the necessary and sufficient condi-
tions to begin the slow but sure road to improving their levels of complexity,
the informational dynamic system has to be made up of at least three dif-
ferent classes of process groups.
The first class of processes is fundamental for providing the system a
certain degree of independence with respect to its environment since it
generates the conditions for being in nonequilibrium state. This is the
endergonic-exergoniccycle capable of producing - in each cycle - quantities
of free energy to the system that are appropriate to generate work (of a
type fundamentally chemical).
This cycle is, in its own right, the one providing the system with the far
from equilibrium characteristic since it forms a component part of its dy-
namic organization. To be a system in the nonequilibrium state is a priority
intrinsic to the system; because of that, it is the most basic fundamental of
informational dynamic systems.
The second process group would form another important part of the
dynamic organization of these systems. This is the one that makes up the
protoplasmic membrane.
First, this component permits a separation, a physical barrier between
the system and its environment, causing from there a different distribution
of compounds, a different dynamic, and a different interaction among the
processes. It also permits certain properties of chemistry to have a focal
point of action to generate possible problematic circumstances which in turn
produce the conditioning situations for exploring different solution strate-
gies for the system in the future. Some examples we can cite are colligative
properties, cinetic chemistry, electrochemical gradients, and others.
 239

It is in the protoplasmic membrane that we will objectively be able to

observe the emergence of the autonomous capacity. It is the part of the sys-
tem that regulates the interaction with its environment as well as providing
the necessary constituents (matter and energy) so that the internal system
processes continue producing their mechanisms in conditions compatible to
their survival. In this way, system cohesiona is not disturbed [38, 391.
This part of the system organization is one that is a complete agent since
it allows for the resolution of the osmotic problems, the adequate concen-
tration of components in the internal system processes, the management
of electrochemical gradients, the reduction of chances for the entrance of
foreign elements into its organization, etc. We can state at the same time
that the system contains two different types of constriction (40,411. The
first one we call “basic constriction”, which directly maintains the f a r from
equilibrium state. It is a cycle between the endergonic processes linked
with the exergonic ones and with the added capacity of generating work (a
necessary and sufficient condition to produce the organizational dynamic
phenomena). The second permits a dynamic and agent interaction with its
external and internal environment: a semi-selective membrane (protoplas-
mic membrane) that contains the compounds which will generate processes
to give the system the capacity of managing both external and internal
conditions. Both types of constriction are so interrelated that a very basic,
direct communication exists between them.
In addition, we will find other types of processes (that make up, along
with the last two, the organizational dynamic identity of the system). The
third group is a network of reactions that would perform the organizational
dynamic’s reparation, maintenance, and reproduction processes of the in-
formational dynamic system and is in one way or another connected to the
constrictions.
The informational dynamic system is organized by the interrelation and
interdependence of these three classes of processes.
The expression of an autonomous and agent behavior implies certain
capacities that are made real in the way that the protoplasmic membrane
performs and coordinates its internal processes. For these reasons, the de-
tailed analysis of this part of the informational dynamic system is valuable.
We must remember that the informational dynamic system is the dy-

aThe idea of cohesion is applied to dynamic systems and their properties that are bonded
by an internal dynamic of relationships. This allows the establishment of a causal sub-
strate for delimiting the dynamic system identity [39]
240

namic organization that constitutes it. So, the system’s dynamic organi-
zation is in fact a collection of processes, interrelated and interdependent,
with the additional element that this interconnection among the processes
maintains a determined form of concretely connecting themselves with the
constriction that permits the system to remain in the far from thermody-
namic equilibrium state.
An agent behavior brings us to the understanding that a certain degree
of managing their internal and external conditions would exist. This man-
agement (even though very minimal at the beginning) directs our attention
to those processes that would be responsible for interacting and performing
recognition, selection, and active or passive transport processes.
And this management is done because it is “dictated by” (there is a
strong correlation with) the tendency to persist as such - to maintain itself
~ in the state that fundamentally characterizes it; the most primary aspect
of its nature is constituted by being a far from thermodynamic equilibrium
system.
It is possible to conjecture that the basic constriction causes the change
of the system’s free energy (AG,,,) to have a tendency towards negative
values. Taken as a totality of the processes and its interconnection, this
causes it to be immersed in continuous and successive dynamics of increas-
ing order.
Therefore, to stop being a far from thermodynamic equilibrium system
is the same as saying that the system no longer exists. To strongly influ-
ence the conditions for remaining in that state implies seriously placing its
survival at risk.
This central aspect of its nature is now maintained by the system itself
through the incorporation of the basic constriction. This allows the system
to maintain itself in the far from equilibrium state.
As this constitutes the most original aspect of these systems, it seems to
us that we can assume that it governs the multiplicity of transformations of
behaviors, components, capacities, and characteristics of the future system.
This way, our proposal rests upon the naturalist supposition that infor-
mational dynamic systems develop their difference from “strict sense” self-
organized systems because they have been able to develop and maintain a
capacity that is elemental for their existence: maintaining or increasing the
far from equilibrium state.
As the constriction maintaining them far from equilibrium is an intrinsic
part of their dynamic organization, there are strategies they can develop
that manage to keep this state in conditions compatible with the laws im-
 241

posed on them by the material world.

The strategies we are referring to will be two new characteristics that
emerge in the system - in the local interactions - and are directed a t main-
taining the far from equilibrium state by a direct or indirect connection
with the basic constriction: information and function.
Briefly, we think that both information and function emerge at the
same time, in unison, in the informational dynamic systems and that the
physical emergence of both notions happens in an interrelational way:
information-function. Moreover, we postulate that both information
and function are relational concepts.
For our intentions, understanding the basic difference that exists be-
tween a system in an equilibrium state and one in a far from equilibrium
state is very important. This important difference permits us to construct a
type of norm that is found in certain laws of nature without any adscription
coming from the human world: the laws of thermodynamics.
Therefore, our proposal of the notion of information-function - as a char-
acteristic emerging in the informational dynamic systems - is a relational
concept that is strongly governed (ruled) by the far from thermodynamic
equilibrium state.
The information-function characteristic has two aspects: the informa-
tional function aspect and the functional information aspect.
When we observe the transmission of some information in informational
dynamic systems, this happens because some type of function is produced.
Similarly, when we observe the presence of some function, it happens be-
cause some information has been transmitted.

3.2. I n f o r m a t i o n and Function Emergence

Both ideas of information and function are directed towards the network
of interrelations among the processes. It is in the processes and the rela-
tionships among them that both capacities emerge for the first time in the
physical world in such a way that through the function pole (“informational
function”), we observe that the contribution of the processes among each
other for their appropriate performance in the integral logic of the system’s
organization causes the system to maintain itself far from thermodynamic
equilibrium.
Through the information pole ( “functional information”), we observe
that the meaning for the system of some signal or sign (that is the physical
means of transporting some matter-energy variation) will be expressed in
242

agreement with maintaining the system in the far from equilibrium state.
However, it can also be that it was in opposition to maintaining the far
from equilibrium state.
In the first situation, the meaning of the matter-energy variation that
comes to the system will cause a positive response by the system. In the
second case, the meaning will generate a negative response by the system.
We can see that both information and function are strongly connected
by their definition with the idea of far from thermodynamic equilibrium.
And this is precisely what we propose.
Because of this, it is possible to consider, through a naturalist perspec-
tive, the far from thermodynamic equilibrium state as a basic norm that
the system imposes on itself for continuing to be the system that it is.
This way, information and function will have to be defined starting
from what we physically have right now, where there is no place for human
adscription, and hence, the possible epiphenomena1interpretation of some
of them is avoided from the start.
The differences that exist between a system in a thermodynamic equi-
librium state and a system in a far from thermodynamic equilibrium state
is something we possess firmly in our minds and have been well defined
through experimentation. The group of causes and contour conditions to
differentiate them can be used in the most explicit way possible.
Therefore, for our defense of a class of system with the characteristics
we mentioned, we would have in the realm of physical laws a naturalist
principle for building up the essential aspects in a normativity.
Effectively, a normative principle points out precisely those circum-
stances when the norm is fulfilled and the difference in those situations
when the norm is not fulfilled. For our work, those circumstances in agree-
ment with the far from equilibrium state are fulfilling the norm, and those
circumstances not in agreement with that state are not fulfilling the norm.
What is interesting about the case is that the norm we just explained
essentially permits the way these systems exist. Therefore, this normativity
is derived naturally from the physical world. Moreover, the normativity
comes from and is dictated by the system itself.
In its origins, what we would have might be a normative definition
of the information-function notion. It is from this initial appearance of
both ideas that we can postulate later on that a kind of symmetry break
would arise in the primitive idea so we would find both characteristics
acting independently (although on occasion, in some way, interacting as
they did at the beginning) and with the possibility of increasing the levels
 243

of complexity and sophistication in one of them or in both.

Everything taken together brings us to the thesis that starting the pro-
cess from inanimate to animate could have been produced by the appear-
ance of a type of dynamic system whose organization is an informational
and finctional dynamic Organization.
The normative idea of functions (we understand the functions to be
“contributing to what”) is founded precisely on this internal normativity
that is produced in the informational dynamic system by the components
and processes that sustain and constitute it. And all of this creates a
tendency to maintain the far from equilibrium thermodynamic state.
We consider the function to be located in the action that generates a
determined process and what we can understand as “what contributes” (co-
operates, favors, supplies) to the interrelation and interdependence among
the processes with a view to maintaining the far from equilibrium state [42].
On the inside of the process, each component will have the mechanism
that is proper to its chemical constitution, but that chemical action (or
group of actions) will have its raison d’gtre because of the process. Then,
each process that carries out a determined and particular action (made up
by the causal nexuses among the component mechanisms that constitute
them) will be assimilated into the dynamic organization only if this particu-
lar action is maintaining, improving, or achieving the far from equilibrium.
And this action is precisely its function in this organizational logic of the
informational dynamic systems.
Thus, a function is something that contributes or facilitates to maintain-
ing or increasing the far from equilibrium state. A dysfunction, on the other
hand, is something that does not contribute or facilitate to maintaining the
far from equilibrium state.
Concerning information, any type of signal or sign can be a carrier of
what might be information (or what we call ‘potential information’). We
consider a signal or sign to be any matter-energy variation (sound wave,
electromagnetism, concentration of a chemical compound, change in pH,
etc.). The signal or sign that is a ‘potential information’ carrier must be
in the surroundings, on the inside, or have transmitted the information
to some system component. The important thing for our idea is that the
information, properly said - The Information - “has a meaning” (very
basic semantic) that is created on the inside of the system. As Menat would
say, “It is meaningful information.” [43].
In a naturalist perspective, any signal, sign, or information must always
exist with respect to something else and not imposed by some observer
244

(we can find other approaches to Biosemantics in [44-481, among others).

Therefore, we do not accept the existence of things like signs, signals, or
information by themselves.
In this way, information will always be information for the system. How
can we understand this information for this system?
First off, that potential information that carries the signal must have
been incorporated to some process. It is in that particular process where
the information might be transmitted. Secondly, the ‘potential information’
becomes Information ( “infomation with meaning” for the system) since it
has the capacity to produce something (an effect) in the process that incor-
porated it, in some other process that is directly or indirectly connected to
the initial process that incorporated it, or with some aspect of the constric-
tions that are effectively maintaining the system far from thermodynamic
equilibrium. Third, the effect has a repercussion in the system, influencing
its own dynamic organization.
For the purposes of our work, we will use Collier’s notion of cohesion
[38, 391. Cohesion is the idea that would provide the informational dynamic
system its identity, in all its transformations in time.
Such cohesion is constituted by the group of relationship and interde-
pendencies that exist among the processes; it is a relational and dynamic
definition that encompasses the nature of the system organization.
The effect of the information that has meaning for the system can be in
the maintenance or the increase of the system cohesion. As well, the effect
could produce some level of interference in the system cohesion, possibly
interrupting one or more processes. It is clear that meaningful information
can be caused by some signal (that carries potential information) coming
from the environment like a signal that is generated in the internal dynamic
of the system. In all cases, whether an effect in favor of or in contrast to
cohesion, the system will develop some type of response that will be corre-
lated to that meaningful information and the process or processes enveloped
by the effect.
Let’s imagine an ideal situation in which all the system processes per-
form each and every one of its actions in a perfectly balanced and inter-
dependent way. In this situation, the informational dynamic system is
normally far from thermodynamic equilibrium (there is a sustained corre-
lation among the processes such that the tendency towards irreversibility
may increase).
In this ideal case, we can see what is happening in the entire system,
what happens among the processes, and inside each one of them as well.
 245

In this case, each process performs its determined function (contribution)

in which this integrated network of relationships is directed towards the
cohesion of a far from equilibrium system. At the same time, the informa-
tion generated in each process and in the interaction among this integrated
network of processes will be considered by the system as a type of system
checkup to see if the cohesion has not been disturbed.
In that sense, the state of being far from thermodynamic equilibrium
behaves like the most fundamental norm that the system imposes on itself,
for which every strategy elaborated by the system (component, process,
relationship, or matter-energy management) for maintaining or increasing
this norm will be incorporated as part of its dynamic structure.
In order to finish our argument, we present some examples that we
consider supports our working hypothesis to a certain degree.
First, it is known that the fibrodendritic projections coincide with the
isofrequency organization planes (high, medium, and low) in the Central
Nucleus of the Inferior Colliculus (CNIC). Tonotopic organization is a fun-
damental property of the auditory system (491, and it is now well estab-
lished that the morphological substrate for such tonotopy in the CNIC is
its laminar organization [50].
The inhibitory action of GABA and Gly is what shapes the temp+
ral and spectral response of Inferior Colliculus (IC) neurons. Due to the
relevance of inhibitory processes in the physiologic responses of the IC neu-
rons, it is realistic to think that we might obtain some explanations of the
acoustic information flow dynamics from the neural populations and the
histochemical and cytoarchitectural organization of this nucleus.
Working with a colleague, who is a neuroscientist, on his acoustic system
research [51], we have found a certain correlation between the distribution
of neurons in the Central Nucleus of the Inferior Colliculus and the way in
which the flow of acoustic information travels through it.
It is observed through densitometric quantitative methods that differ-
ent populations of GABAergic neurons (which are inhibitor neurons) exist
in each isofrequency plane that are oriented in transversal planes to each
isofrequency plane. Thus, a topological relationship of neurons exists with
respect to the laminar organization. This topological dynamic organization
in each isofrequency plane correlates clearly with GABA gradient concen-
tration such that it would generate a determined rhythm to the temporary
excitatory flow. If only one type of neuron population would exist in each
isofrequency plane, then this hypothetical fact could be considered by us
as a counterexample to our working hypothesis.
246

Second, erythrocytes mature in bone marrow by means of the hormone,

erythropoietin, and then are released into the blood stream in where usually
are destroyed 120 days after. Recently, levels of anemia have been discov-
ered in returning astronauts as a result of the destruction of erythrocytes.
What is surprising in this case is that in place of destroying the oldest
cells (which is the expected) the focus of the destruction is the youngest
cells. This is a phenomenon known as “neocytolysis” [52]. This leads to
the question: Could certain signals exist that would carry a determined
type of information for the organism as a whole system such that it would
make “functional” the massive destruction of the youngest erythrocytes be-
fore the oldest? It is important to point out that erythrocytes do not have
nuclei.
The topic of viruses seems to us a very revealing case as well. These
are entities that are not autonomous systems; nevertheless, in determined
conditions, they develop processes that come out to be informational and
functional. The processes for entering a cell begin with “recognition” of
the target cell receptors, whose activation by the virus will cause notable
changes in the plasmatic membrane, and end up with the virus entering
the cytoplasm [53].
Here, we can visualize certain “functional information” for the virus
that is “dysfunctional information” for the host cell. Its molecular compo-
nents are such that they can perfectly interact with the organizational logic
of the target cell and only because there is that organizational dynamic (in-
formative and functional in the target cell); it is that its components, upon
interacting, will carry determined information that makes the functional ac-
tions concrete since they contribute to its reproduction and survival. The
relational character is revealed in that these same processes are dysfunc-
tional for the cell and the information that certain components of the virus
transmit are counterproductive for the cell (to such a degree that the virus’s
use of the synthesis and replication machinery of the cell will end up de-
stroying the cell). These are interesting lines of research, attempting to find
out which signals permit “blocking” the mechanisms and processes which
monitor the internal states so that the virus ‘takes possession” of a large
part of the cellular metabolism.

References
1. Nicolis G & Prigogine I, Self-Organization in Nonequilibrium Systems:Fbom
Dissipative Structures to Order through Fluctuations. Wiley, New York
(1977).
 247

2. Strogatz SH, Nonlinear Dynamics and Chaos. Westview Press (1994).

3. Haken H, Synergetics: an introduction. Springer, Berlin (1978).
4. Bak P, How Nature Works: The Science of Self-organized Criticality.
Springer, Berlin (1996).
5. Heylighen F, Bollen J & Riegler A (Eds.), The Evolution of Complexity.
Kluwer Academic, Dordrecht (1999).
6. Maes P (Ed.), Designing Autonomous Agents: Theory and Practice From
Biology to Engineering and Back. MIT Press (1991).
7. Kauffman S, The Origins of Order: Self-organization and Selection in Euo-
lution. Oxford University Press, New York (1993).
8. Bar-Yam Y,Dynamics of Complex Systems. Addison-Wesley (1997).
9. Wachtershauser G, Before enzymes and templates: Theory of surface
metabolism. Microbiological Reviews,52: 452-484(1988).
10. Hazen RM, Life’s rocky start. Scientific American, 284 (4): 63-71 (2001).
11. Ferris JP, Montmorillonite catalysis of 30-50mer oligonucleotides: laboratory
demonstration of potential steps in the origin of the RNA world. Origins of
Life and Evolution of Biospheres,32:311-332(2002).
12. Cairns-Smith AG, Fine-tuning in living systems: early evolution and the
unity of biochemistry. International Journal of Astrobiology, 2 (2): 87-90
(2003).
13. Ricardo A, Carrigan MA, Olcott AN & Benner SA, Borate minerals stabilize
ribose. Science, 303:196 (2004).
14. Fisk MR, Giovanonni SJ & Thorseth IH, Alteration of oceanic volcanic glass:
Textural evidence of microbial activity. Science, 281: 978-980 (1998).
15. Schopf J W , Kudryavtsev AB, Agresti DG, Wdowiak T U & Czaja AD, Laser-
Raman imagery of Earth’s earliest fossils. Nature,416: 73-76 (2002).
16. Van Zuilen MA, Lepland A & Arrhenius G, Reassessing the evidence for the
earliest traces of life. Nature,418: 627-630 (2002).
17. Furnes H, Banerjee NR, Muehlenbachs K, Staudigel H & de Wit M, Early
life recorded in Archean pillow lavas. Science,304: 578-581 (2004).
18. Foustoukos DI & Seyfried J r WE, Hydrocarbons in hydrothermal vent fluids:
the role of chromium-bearing catalysts. Science, 304:1002-1005 (2004).
19. Lewontin RC, The units of selection. Annual Review of Ecology and System-
atics, 1: 1-18 (1970).
20. Sober E, The Nature of Selection: Euolutionary Theory in Philosophical Fo-
cus. MIT Press (1984).
21. Brandon RN, Adaptation and environment. Princeton University Press
( 1990).
22. Kirschner M & Gerhart J , Evolvability. Proceedings of the National Academy
of Sciences, 95: 8420-8427 (1998).
23, Radrnan M, Matic I & Taddei F, Evolution of Evolvability. Annals of the
New York Academy of Sciences, 870(1): 146 - 155 (1999).
24. Csete ME & Doyle JC, Reverse Engineering of Biological Complexity. Sci-
ence, 295(5560): 1664 - 1669 (2002).
25. Poole A, Phillips M & Penny D, Prokaryote and eukaryote evolvability.
Biosystems, 69: 163-185 (2003).
248

26. Tanay A, Regev A & Shamir R, Conservation and evolvability in regulatory

networks: The evolution of ribosomal regulation in yeast. Proceedings of the
National Academy of Sciences, 102(20): 7203-7208 (2005).
27. de Duve C, The Beginnings of Life on Earth. American Scientist, 83: 428-437
(1995).
28. Bada JL, Miller SL & Zhao M, The stability of amino acids at submarine
hydrothermal vent temperatures. Origins of Life and Evolution of Biospheres,
25: 111-118 (1995).
29. Cooper G, Kimish N, Belisle W , Sarinana J, Brabham K & Garrel L, Car-
bonaceous meteorites as a source of sugar-related organic compounds for the
early Earth. Nature, 414: 879-883 (2001).
30. Bernstein MP, Dworkin JP, Sandford SA, Cooper GW & Allamandola LJ,
Racemic amino acids from the ultraviolet photolysis of interstellar ice ana-
logues. Nature, 416: 401-403 (2002).
31. Monnard PA, Ape1 CL, Kanavarioti A & Deamer DW, Influence of ionic
inorganic solutes on self-assembly and polymerization processes related t o
early forms of life-implications for a prebiotic aqueous medium. Astrobiology,
2: 139-152 (2002).
32. Rees DC & Howard JB, The interface between biological and inorganic world
iron-sulfur metalloclusters. Science, 300: 929-931 (2003).
33. Kosztin I & Schulten K, Fluctuation-driven molecular transport through an
asymmetric membrane channel. Physical Review Letters, 93: 238102 (2004).
34. Levine RD, Molecular Reaction Dynamics. Cambridge University Press
(2005).
35. Schopf J W , Microfossils of the early Archaean apex chert: new evidence of
the antiquity of life. Science, 260: 640-646 (1993).
36. Keefe AD, Newton GL & Miller SL, A possible prebiotic synthesis of pan-
tetheine, a precursor of coenzyme A. Nature, 373: 683-685 (1995).
37. Leman L, Orgel L & Ghadiri MR, Carbonyl Sulfide-Mediated Prebiotic For-
mation of Peptides. Science, 306 (5694): 283 - 286 (2004).
38. Collier J, Entropy in evolution. Biology and Philosophy, 1: 5-24 (1986).
39. Collier J , Self-organization, Individuation and Identity. Revue Zntemationale
de Philosophie, 59: 151-172 (2004).
40. Patee HH, Laws and constraints, symbols and languages. In Towards a Theo-
retical Biology 4, Essays, Waddington CH (Ed.). Edinburgh University Press,
pp. 248-258 (1972).
41. Kauffman S, Investigations. Oxford University Press, Oxford (2000).
42. Riofrio W , La vida en sus origenes: Las propiedades b&icas. 20th World
Congress of Philosophy, Boston, Ma (1998).
43. Menant C, Information and Meaning. Entropy, 5: 193-204. (2003).
44. Millikan RG, Language, Thought and Other Biological Categories. Cambridge
MA: MIT (1984).
45. Millikan RG, Biosemantics. The Journal of Philosophy, 86 (6): 281-297
(1989).
46. Millikan RG, Biofunctions: Two Paradigms. In Functions: New Readings i n
the Philosophy of Psychology and Biology, Cummins R, Ariew A & Perlman
 249

M (Eds.). Oxford: Oxford University Press, pp. 113-143 (2002).

47. Emmeche C, A-life, Organism and Body: the semiotics of emergent levels. In
Bedeau M, Husbands P, Hutton T, Kumar S & Suzuki H (Eds.), Workshop
and Tutorial Proceedings. Ninth International Conference on the Simulation
and Synthesis of Living Systems (Alife ZX), pp. 117-124 (2004).
48. Kull K, A brief history of Biosemiotics. Journal of Biosemiotics, 1: 1-34
(2005).
49. von Bitkhsy G, Experiments in hearing. McGraw-Hill, New York (1960).
50. Oliver DL & Morest DK, The central nucleus of the inferior colliculus in the
cat. The Journal of Comparative Neurology, 222: 237-264 (1984).
51. MerchLn M, Aguilar LA, L6pez-Poveda EA & Malmierca MS, Immunocy-
tochemical and semiquantitative study on Gamma-aminobutiric acid and
glycine in the Inferior Colliculus of rat. Neuroscience, 136 (3): 907-925
(2005).
52. Rice L & Alfrey CP, Modulation of red cell mass by neocytolysis in space and
on Earth. European Journal of Physiology, 441(2-3 Suppl):R91-4 (2000).
53. Doming0 E, Webster R & Holland J (Eds.), Origin and Evolution of viruses.
Academic Press (2000).
 GRASPING THE COMPLEXITY OF LIVING SYSTEMS
THROUGH INTEGRATIVE LEVELS AND HIERARCHIES

J , M. SIQUEIROS & JON UMEREZ

Department of Logic and Philosophy of Science
University of the Basque Country (UPV/EHU)
Avda. Tolosa 70, 20018, Donostia-San Sebastian, Spain
E-mails: {jsiqueiros, jumerez} @sf. ehu. es
http://www. ehu. es/ias-research

In this paper we intend to address the conceptual challenge posed by complexity

by focusing on living systems. Biological complexity has traits that seem t o be
more amenable t o scientific scrutiny and description, without having t o turn t o
higher order capacities as sociality or human intelligence. Notwithstanding, bio-
logical complexity keeps the conceptual depth and suggestive power that appear
lacking in more formal or physical expressions of the idea. The paper presents a
conceptual overview of the different perspectives that along the 20th century were
developed with the intention to grasp biological complexity (i.e., Organicism, The-
ory of Levels of Integration, etc.). Concepts that nowadays are so common t o the
different approaches on complexity such as emergence, hierarchy, downward causa-
tion, wholeness, etc.- arose for the first time in these theoretical perspectives during
the early 20th century and in the 60's and 70's. The goal of this paper is t o point
out that these accounts still have deep insights and potentialities that are worthy
t o recast in order t o advance in the understanding and exploration of biological
complexity and complexity in a wider sense, specially the ideas of autonomous hi-
erarchies and the notion of constraints as enabling not only as restrictive material
structures.

1. Introduction. Complexity: an elusive idea

Along with many of the technological achievements during the last decades,
the re-emergence of the idea of complexity and its incorporation into many
and diverse scientific disciplines have contributed greatly to the change on
how science is done nowadays. On the one hand, complexity has proved
to be a good methodological frame for modeling and, on the other, con-
cepts like emergence or systematicity, that once were only peripheral and
marginalized, now are almost part of everyday language, since they are
tools that allow to approach problems from a new perspective.
Moreover, the impact of the idea of complexity has not been only experi-

250
 251

enced in the natural sciences but it has reached and influenced a wide span
of disciplines, from physics to the humanities and, of course, to what Simon
began to call “the sciences of the artificial.” Even so, it is remarkable how
difficult is to define the concept of complexity. Complexity is a very elu-
sive idea, due, on the one hand, to the wide range of areas of applicability
and, on the other, to the multiplicity of proposed definitions and measures.
Nevertheless, it is not our intention to offer a huge list of definitions of what
other people think what complexity is.a On the contrary, we will just very
shortly describe which general traits epitomize complex systems and then
we will characterize what kind of systems are the living ones and which are
their peculiarities (as organized complexity).
Therefore, in this paper we intend to address, at least partially, the
conceptual challenge posed by complexity (i.e., apparent contradiction of
simultaneous vagueness and usefulness) by focusing on living systems. It
is generally taken for granted that living beings are complex, even charac-
teristically so, and this attribution does not seem especially controversial.
Besides, the traits that make biological organization complex seem to be (at
least to certain degree) more amenable to scientific scrutiny and descrip
tion, without turning to higher order capacities such as sociality or human
intelligence. Notwithstanding, biological complexity keeps at the same time
some conceptual depth and suggestive power that appear lacking in more
formal or physical expressions of the notion of complexity.
Turning then to biological complexity, if we review the philosophical
panorama of the past century regarding the conceptual tools that have been
used to understand and explain living systems, we may easily find (among
other equally interesting turning points) that the fate of a cluster of ideas
revolving around aspects of complexity, such as hierarchy, levels, emergence,
organization, wholes and systems, etc., has been peculiar, ranging from
proposal and more or less easy coexistence and debate with other opposing
views in the first part of the century to total disappearance from the larger
part of that period, and to a renewed interest in its last decade. Witness
to this resurgence are, to mention two approaches relevant to this work,
the very recent mainstream appeal to build a systemic biology or systems
biology (see Ref. 1) and the not as recent constitution of the field of theories
or sciences of complexity, including Artificial Life, which we might date back

Lloyd, from MIT, has been compiling for years a “non-exhaustive list of measures of
complexity” that can be accessed at the web:
http://web.mit .edu/esd.83/www/notebook/Complexity.PDF
252

to the mid 80’s (see Refs. 2, 3, 4).b

We think, therefore, that it would be worthy to recast some of those
perspectives from the first third of the 20th century (emergent evolution,
organicism, integrative levels,. . .), as well as those fighting to get some
audience during the 60’s and the 70’s (theoretical biology, hierarchy theory,
systems science,. . . ), not only in their common generality of approach but
paying special attention to their particularities in order to highlight their
specific potentialities to advance in our understanding and explanation of
biological features and the very phenomenon of life.
We pose it that way because, even when some kind of acceptance of the
general perspective is granted, problems are continually presented. Some
are related to the presumed vagueness of the ideas, other to their contested
instrumental fertility, and yet other to their physical implementation or
their ontological nature. To answer this kind of objections, besides a general
philosophical justification of the approach, it is also convenient to provide
more specific and operative ways to apply it to precise cases or areas of
research.
Within this context we propose in this paper to recuperate two (among
many) of those specific views, each belonging to one of the two periods
mentioned before. The first is the notion of integrative levels defended by
Needham and other organicists during the 30’s (and later, for instance, by
Novikoff or Feibleman). The second is the notion of hierarchy developed
by Pattee based on the concept of constraint during the late 60’s and the
70’s (and later by others with analogous approaches, for instance, Salthe).
The idea of levels of integration has a sound grounding in the work
of organicist embryologists and theorists (Needham, BertalanEy, Woodger,
Weiss, Waddington, etc.) and it offers theoretical and empirical tools to
understand the organization of a living system as it develops. The idea of hi-
erarchy elaborated by Pattee seeks to offer operational tools as the concept
of constraint (in the sense of Ref. 5) and the distinction between holonomic
and non holonomic ones, in order to explain how levels can emerge and re-
late to one another in different ways, giving rise to specific hierarchical
arrangements, as the one showed by biological organization. The working
out of those instruments will allow us to develop an idea of biological hier-

bHowever, it is not the purpose of this short paper to deal with the logically subsequent
issue of trying to assess to what degree do these recent approaches innovate with respect
to paths previously open or just develop them further (or even if they aware of them).
Therefore, we are not going to discuss here current efforts to approach the complexity
of living systems along similar lines to those we will describe in the paper.
 253

archy as a relation among levels with different and specific dynamics which
gives way to emergent processes both upwards and downwards.
We claim that this kind of work is a very appropriate way to advance in
disclosing what complexity means. Recalling Weaver’s distinction between
problems of “simplicity,” “disorganized complexity” and “organized com-
plexity,” we intend to contribute to characterize as precisely as possible a
kind of systems, the living ones, which are paradigmatic of that organized
complexity whose challenge Weaver thought would be accessible within the
next fifty years since his paper was published in 1948.6

2. Sciences of objects and systems, some simple, some

complex
It is possible to say that a system is formed by things and the relations
among them.7 According to this, systems can be defined in terms of their
thinghood or in terms of their systemhood.8 To define a system in terms
of its thinghood is to identify it by those traits that are not part of its
systemic character. These very general criteria make possible to recognize
different systems by the things that form them. In a broad sense we could
say, for instance, that there are physical systems, biological systems, social
systems, and artificial systems that, even if they are not separate kinds,
represent different phenomenological and knowledge realms.
To define a system in terms of its systemhood is to identify it by the
relations among things. Relations are what connect and join entities giving
rise to a system. So, different things can establish similar relationships.
The other way around is also possible, the same things can establish very
different relations. From this perspective, the very subject of a systemic
approach is the analogies between the kinds of relations that things can
establish.
Complex systems are mostly understood in terms of their systemhood
instead of their thinghood. Even though, at this moment, it is almost
impossible to offer a general definition of complexity due to the variety and
diversity of phenomena that can be labeled as such (See note 1). Moreover,
the concept of complexity is related to a field that nowadays is in full
development.
Still, a general outline or a summary of complex systems according to
some of their properties is possible. Beginning at a physical and chemical
level we might find the incapability to predict global behaviors from local
component behavior -nowlinear systems (see Ref. 9). As well, an immense
254

variety of behaviors are possible due to an extreme sensibility to the ini-

tial conditions -chaotic systems.lO,ll We have already become familiar with
dissipative phenomena and the emergence of order in far from thermody-
namic equilibrium conditions -dissipative system^.^^^^^ Somehow a gener-
alization of the previous case is the emergence of robust and stable phe-
nomena that appear form unstable situations due to self-organizing prop-
erties -self-organizing systems (see Ref. 14). If we turn to biological (and
psycho-social) systems, many of the properties just mentioned for physical
and chemical complex systems get intensified through the appearance of
functional, regulative, and organizational features and, especially, through
informational and symbolic attributes. Finally, artificial systems also can
show complex behaviors, as is the case of cellular automata, Boolean net-
works, or neural networks, among others (see Ref. 15).
Apart from chaotic phenomena, most of the cases mentioned above im-
ply a high dependency on a singular relationship between the behavior of
individual components and the global or collective one, a situation that
unavoidably brings to memory the classic problem of the relationship be-
tween parts and wholes, emergent behavior, and the problem of hierarchies.
Besides, in the biological realm, this characteristic is autonomously gener-
ated, from within the very system, and not instructed from without by any
external agent.16 This is the double reason why we are proposing to focus
our attention in “grasping living systems complexity through integrative
levels and hierarchies.”
But first, turning back to Weaver’s 1948 paper, we might recall that it
explains how the complexity of a problem (or system) is related to the num-
ber of variables involved. In order to conclude this broad characterization
of complexity, we will shortly render Weaver’s classification. He recognized
three different kinds of problems:

a. Problems of simplicity are those that imply two variables (maximum

3 or 4) as in the classical mechanics before 1900.
b. Problems of disorganized complexity are faced when, after 1900, a
large amount of variables become tractable through probability theory
and statistical mechanics:
“It is a problem in which the number of variables is very large,
and one in which each of the many variables has a behavior which is
individually erratic, or perhaps totally unknown. However, in spite of
this helter-skelter, or unknown, behavior of all the individual variables,
the system as a whole possesses certain orderly and analyzable average
 255

properties.” (Ref. 6:538).

c. Problems of organized complexity appear when we deal with an inter-
mediate number of variables that cannot be averaged and it is only
nowadays that we are becoming able to address them:
“The really important characteristic of the problems of this middle
region (. . . ) lies in the fact that these problems, as contrasted with
the disorganized situations with which statistics can cope, show the
essential feature of organization. (. . . ) They are all problems which
involve dealing simultaneously with a sizable number of factors which
are interrelated into a n organic whole.” (Ref. 6:539).

2.1. Some methodological and conceptual conclusions

We might try to extract some methodological and conceptual conclusions
on the development of the idea of complexity. To begin with, simple sys-
tems are those that perfectly fit to a Newtonian and Laplacian classical
epistem01ogy.l~In order to explain them all that is required is to apply
the laws of motion, that is, no alternative descriptions of the system are
needed. A few variables are enough, linearity holds and there is a propor-
tional relation between causes and their effects. System behavior can be
deduced or predicted by knowing the initial conditions and the individual
behavior of the components. If we move on, one of the biggest problems
that Laplace already faced was the incapacity of human minds to define
all the initial conditions of all bodies in the universe, this was a situation
way beyond the two variable classical epistemology of “his” physics; this
difficulty is the first instance of complexity.
Moreover, in the case of full-fledged complex systems there is no pro-
portional relation between causes and effects, because the system behavior
cannot be reduced to the sum of the behavior of its components. Due to
this property, besides a description grounded in the laws of motion, an al-
ternative and complementary description may be required; this is to say, it
is not enough to describe the local behavior in order to predict the global
behavior. Since there is a difference between local and global behavior, two
levels of description are required, the first one is usually called “micro” and
makes reference to the behavior of local or individual components; the sec-
ond one is the “macro” because it makes reference to the global or collective
behavior of the system.
Moreover, this global behavior will be qualitatively different between
systems showing disorganized complexity and those displaying organized
256

complexity. In the first case, the behavior of the components of the system
can be described as Brownian and the global outcome can be described
as an average; examples would range from gases to more interesting ones
such as some dissipative structures or chaotic systems. The second case
would include biological and some social systems, where the behavior of
components is locally controlled by the level of the whole system (or specific
modules).
In general terms, complex systems show a global emergent behavior
that is not predictable with the classical mechanics toolkit, due to their
non-linear character. Complex systems description always imply, at least,
two levels lower and upper levels- and, in order to explain their behavior,
a full account of the relations between them is required. The levels at the
bottom give rise to global patterns (upward causation) meanwhile those
global patterns harness the behavior of the levels at the bottom (downward
causation).18 In the case of disorganized complexity, the level above may
impose what is known as boundary conditions; in organized complexity, the
level above may impose functional constraints on the level below.
It is this challenge that Simon, to cite another classical paper, seeks
to answer by claiming that the architecture of complexity is necessarily
hierarchi~a1.l~

3. Biological complexity: hierarchies

Aside from the variety of scientific objects that are recognized as complex,
the epistemology, the methodologies, and the concepts behind each of the
cases above can be historically traced back to, at least, three theoretical
frameworks: a) Cybernetics and Systems Science$ b) the work on emer-
gence at the end of the l g t h century and the first half of the 20th; and c)
the organicist theories in biology.
All these different perspectives shared a set of common interests and
intuitions, which basically were to overcome the limits of reductionism
and determinism (curiously enough they also share some participants and
spatio-temporal coordinates). In order to develop a scientific approach to
a more complex phenomenology, their effort was focused, for instance, on
a wider notion of causality, an understanding of the relationship between
levels, an appreciation of contextual aspects, etc. Of the entire conceptual

“Even if they had a different origin and development, somehow they have come t o meet
in a common corpus (see Refs. 20, 21, 22, 23).
 257

universe around complexity, we are interested in hierarchies and how they

are expressed and studied in biological cases.
Hierarchical approaches to what nowadays are considered, in its most
general sense, complex systems phenomena have been developed all along
the end of the lgth century to the p r e ~ e n t .We
~ ~may,
, ~ ~ for instance, recall
the suggestions of Mill or Lewes in the second half of the lgth century. Mill
defined what he called heteropatic laws in contrast to compositional ones
and Lewes distinguished among resultant and emergent properties. During
the 20th century the most important approaches were that of Emergent Evo-
lution developed primarily by Lloyd Morgan or Wheeler,27i28the Theory of
Integrative Levels exemplified in the works by Needham and N o v i k ~ f f , ~ ~ ~ ~
and the Materialist Emergentism well represented in the collection of papers
edited by Sellars et al.31116 Simultaneously, Woodger was claiming that the
most important feature of living beings was their hierarchical organization
(we may notice that in the 30’s BertalanfFy was closely connected to Need-
ham and W ~ o d g e r ) In
. ~ the
~ years between the forties and the sixties and
almost the seventies, hierachical approaches to natural or artificial system
seemed to disappear from mainstream science. However, resurgence began
as the work of Herbert Simon on the “Architecture of Complexity” in 1962
or Polanyi’s 1968 paper on the idea of constraint were p ~ b 1 i s h e d . By
l ~ ~the
~
1970’s the situation changed and we can highlight Howard Pattee’s work,
among others, as well as that of T u r ~ h i nin ~ 1980’s we will underline
; ~the
the contribution by Salthe,34 and today there is the work developed by
Heylighen on metasystem and evolutionary transition^.^^,^^

3.1. Organicism
Organicist theories in biology developed mainly during the interval between
the two world wars by practitioners of theoretical biology generally linked
to e m b r y ~ l o g y . Organicism
~~>~~ was a research framework that developed
an epistemology addressed to cope with the complexity of life. It might
be among the first ones in doing so in the 20th century since it dates back
to the early days of the 1910’s with the work of Ross Harrison. Among
those who were involved on the new perspective were the just mentioned
Harrison, Bertalanfi, Needham, Waddington, Weiss, and Woodger among
others. Organicism in a wider scope was a response to a set of general
claims and debates developed all along the lgth century between vitalism
and materialism (e.g., some of these were on naturalism vs. supranatural-
ism, determinism us. freedom, mechanicism vs. vitalism). A very specific
258

one was related to how biological form should be explained, a problem that
defied every mechanisic or vitalist explanation since the 18th century (and
still does to some extent), due the emergence of order apparently “for free.”
So, in a much more narrow sense organicism offered an alternative program
regarding the emergence of embryological order. The main difference be-
tween this alternative view with reference to vitalism and materialism was
the search for an emergentist account that would be consistent with physics.
All the efforts of organicism were addressed to understand the forms that
arise and proliferate in the organic world and the comprehension of the laws
that generate them. To explain form it was necessary to understand what
was responsible for its origins, maintainance and transformation. This was a
twofold situation, on one side, it was necessary to approach the evolutionary
aspects of form; on the other, its developmental features. A particular
challenge posed by organic forms to any evolutionary or developmental
account of it was that organic forms are progressively defined in the process
of growing and not piecemeal built.
In order to address these problems, organicism developed a set of ideas
and concepts that can be classified, following Bertalanfi and Morton Beck-
ner, in three groups: Organization and Wholeness, Directiveness, and His-
toricity or dynamic^.^^^^^
Organization and Wholeness. The concept of organization made refer-
ence to the hierarchical constitution of organisms. Each level was formed
by parts which, due to the relations among them, displayed at that level
rules and properties that were absent at the levels below. Within this frame
some organicists like Harrison and Needham, developed their notions on the
connection between structure and function.
Akin to the concept of organization was the notion of wholeness. Organi-
cists always emphasized the importance of the appreciation of wholeness.
This notion was about regulation, that is, about the top-down perspective
that allows to see how the rules at the level of the system exert their power
over the relations among its constitutive parts. In other words, the exis-
tence of a whole stands on the subordination of parts to the rules or laws
that characterizes the systems as such.
To make explicit the link between form, wholeness and organization,
for the organicists, form would be the expression of the whole. In this
sense, their main concern was to understand how, along development and
evolution, the relations among parts and the relations among levels give rise
to new levels of complexity; and at each step of these processes, properties
such as symmetry, polarity, and patterns are the means through which form
 259

expresses the wholeness of the system i.e. organism.

Directiveness. Directiveness is another notion shared by the organicists.
As Donna Haraway noticed, all different expressions of organicism agreed
on the Aristotelian idea of the necessity of a teleological explanation that
would be complementary to a materialist explanation (see Ref. 41:34).
Teleology was present in the organicist way of thinking as a property of
the whole; according with their view, this property is displayed at several
organismal instances, that is, at the developmental process towards specific
forms, at reproduction, and certainly in the case of behavior. Organicists
highlighted that directedness at any of the instances was not an external
imposition, but generated by the very organization of the system. It is
interesting to notice that, on this subject, Bertalanffy suggested to replace
the word teleology by that of organismic. The word organismic would
express the idea that what is characteristic of organisms is that physico-
chemical processes are organized in such a way that “they are directed to the
maintenance, production, or restoration of the wholeness of the organism.”
(Ref. 39%).
Historicity OT Dynamics. Organicism appreciated that organisms have
a developmental and evolutionary history. The historical and dynamical
character of organisms were explicitely recognized, specially in their notion
of form as a dynamical property and in the rejection of any static notion
of it. By this token there were emphasized at least three ideas: one, that
development is a process of transformations that depends to some degree on
previous states of the organic system. Second, that development is a process
that involves novelty. Finally, the historical character is also linked to the
idea that the developmental and evolutionary processes are irreversible.

3.2. Theory of Integrative Levels

In the 1930’s and 40’s a lot of theoretical work on hierarchy theory was
done by some researchers with an organicist background. Most of the
researchers that participated on the development of the hierarchy theory
proposed that nature is structured in emergent levels that are qualitatively
different amongst each other. James Feibleman, Joseph Needham, and Alex
Novikoff are some of these theorists that found inspiration in emergentist
theories as well as in organicist approaches and developed what was known
as Theory of Integrative Levels. Like organicism, this was not a reduction-
ist, mechanicist or vitalist approach, but an emergentist approach grounded
on a materialist epistemology.
260

The Theory of Integrative Levels stated that the relation among the
components of a system is what gives rise to new emergent levels and that
new properties are the outcome of the organizational and integrative prin-
ciples instead of the component properties. The purpose of taking into
account the integration of the components in the system’s organization
was to include as fundamental the properties of the level considered at the
bottom and those of the levels considered at the top, i.e., the so-called
emergent. They considered that there are some general levels that can be
identified as physical, chemical, biological and social, and that each one has
its very own laws as well as its own mechanisms through which changes or
in between levels emergence occur.
One of the most representative features of the Theory of Integrative
Levels is their stand against the reduction of biology to physics and chem-
istry; Novikoff, for example, was clear about the fact that living organisms
are not just machines made of many physico-chemical units that could be
taken apart and whose global behavior could be deduced through analysis
(see Ref. 30:210). For this theory the main biological levels were the cells,
tissues, organs, organisms, and populations.
As methodological and ontological aspects, we may consider the follow-
ing rephrased from Feibleman:42
The analysis of a system must be done considering its lowest and its
highest levels in order to achieve the most complete explanation.
An organization always belongs to its highest level.
At the end, every organization must be explained in terms of its very
own level.
It is impossible to give a full account of an organization if it is ex-
plained only in terms of its lowest or its highest level (e.g., an organism
cannot be explained just in terms of genes or ecosystems).

3.3. Pattee and recent work (Pattee on hierarchies €d the

concept of constraint)
Since the end of the 60’s, Howard Pattee developed a theory of hierarchies
that was very i n f l ~ e n t i a lBetween
.~~ 1969 and 1973 he wrote a compact se-
ries of papers that are mainly dedicated to the exposition of his hierarchical
view (see Refs. 44, 45, 46, 47, 48). This work includes the edited volume
entitled precisely Hierarchy Theory. The Challenge of Complex Systems,49
which included papers by J. Bonner, C. Grobstein, R. Levins, H. Simon
and H. Pattee himself, besides a long Postscript on “Unsolved Problems
 261

and Potential Applications of Hierarchy Theory.”48

There is nearly no treatment of the subject since the 70’s that does
not turn to, at least, his 1973 edited collection (see Ref. 34 for the best
treatment one decade later, and Ref. 50 for a recent review of the state of
the art). In a simplified way, it was his previous concern with the nature
of molecular control together with the need to find operative ways to deal
with complexity, what led Pattee to undertake the task of devising a more
comprehensive hierarchical perspective.
Following Simon’s path (see Refs. 19, 51), Pattee related hierarchical
organization to a best likelihood of evolution, but he went further in order to
explore the conditions of possibility for unlimited open-ended evolution of a
von Neumannian ancestry.43 Therefore, he complemented Simon’s, mostly
structural, kind of hierarchical order with the elucidation of a functional
kind, which he needed to give account of control attributes.
In order to give an account of this functional kind, he developed
Polanyi’s idea of the dual control of laws and constraint^.^ Polanyi’s view
of boundary conditions (or constraints) as harnessing the laws of nature
and, therefore, neither disobeying them nor being reducible to them, will
allow Pattee to develop a biological theory grounded on an illuminating
understanding of the role of constraint.
Pattee made a distinction between holonomic and non-holonomic con-
straints:

a. Holonomic constraints: These are constraints that freeze or simply

eliminate permanently some degrees of freedom making easier to de-
scribe the system and its behavior; e.g. the walls of a Petri dish;
boundary conditions, and the surface of a solid. In the cme of dy-
namical systems, holonomic constraints are rate-dependent to the
flux of matter and energy of the system so they arise when certain
conditions are present. Holonomic constraints allow the establish-
ment of a statistical and global order.
b. Non-holonomic constraints: They are material structures that are
rate-independent with respect to the inherent dynamics of the sys-
tem, though they act locally on it. By the same token, these are
non-permanent constraints, they temporally generate relations of de-
pendency among some degrees of freedom of a system. It can be said
that this kind of constraint is a material structure that embodies a
rule. An example would be a switch, which is independent of the
flow of electrons, though the switch x t s on this flow when the rule
262

conditions are satisfied. Due to its dynamical decoupling, it would

be more proper to describe it as another (higher) level.

Consequently, Pattee insisted on two somehow paradoxical but crucial

features of hierarchies: that very organized systems as hierarchies exper-
iment a simplification of their behavior, and that the limitations exerted
upon the degrees of freedom at one level result in an increase at the subse-
quent one. Those operations are performed through material devices in the
form of constraints of different kinds, mainly holonomic for structural hi-
erarchies and non-holonomic for functional ones. The very specific control
relation is introduced as selection of alternatives.
Together with the basic distinction between laws and rules and the hi-
erarchical perspective distinguishing structural and functional, the concept
of physical constraint makes room for the natural acceptance of a fully
material kind of downward causation.

4. Conclusion
The work of all these researchers on biological complexity offers a toolkit
to build an operative approach to phenomena whose epistemological impli-
cations go beyond biology and which is interesting for the study of complex
systems in general. Beginning with the necessity of a hierarchical arrange-
ment, there is further interplay among structural and functional hierarchies.
In this context, the emergence of autonomous hierarchies implies having
systems with the capacity to self-impose holonomic and non-holonomic con-
straints. Those systems and their increasingly complex behaviors appear
naturally in biological cases (and, obviously, in socio-cultural ones), but are
also quite interesting for artificial systems. Within this kind of full-fledged
complex systems, the relations among levels are described as upward and
downward causation and are made operative through the concept of con-
straint. But constraints understood as enabling mechanisms and not only
as restrictive material structures. Finally, this scenario sets the conditions
to allow the open-ended evolution of complexity.

Aknowledgements
Funding for this research was provided by grant HUM2005-02449 from the
Ministerio de Educacin y Ciencia of Spain and the FEDER program of the
E.C., and grant 9/UPV 00003.230-15840/2004 from the University of the
Basque Country. Jes6s M. Siqueiros is holding a predoctoral scholarship
 263

from t h e Consejo Nacional de Ciencia y Tecnologia (CONACyT) of Mex-

ico. Jon Umerez holds a Ram6n y Cajal senior researcher position (R y C
program is funded by t h e MEC a n d t h e UPV/EHU).

References
1. -, Systems Biology. Science, 295, 1661-1682 (2002).
2. D. Pines (Ed.), Emerging Syntheses in Science: Proceedings of the Founding
Workshops of the Santa Fe Institute. Redwood City, Addison-Wesley (1988).
3. G. A. Cowan, D. Pines, and D. Meltzer (Eds.), Complexity Metaphors, Mod-
els, and Reality. Redwood City, Addison Wesley (1994).
4. C. G. Langton (Ed.) Artificial Life Redwood City, Addison-Wesley (1989)
5 . M. Polanyi, Life's irreducible structure. Science, 160, 1308-1312 (1968).
6. W . Weaver, Science and complexity. American Scientist, 36 (36), 536-544
( 1948).
7. G.J. Klir, Facets of Systems Science. New York, Plenum Press (1991).
8. R. Rosen, Some comments on systems and system theory. International Jour-
nal of General Systems, 13, 1-3 (1986).
9. D. L. Stein (Ed.), Lectures in the Sciences of Complexity: the proceedings of
the 1988 Complex Systems Summer School. Redwood City, Addison-Wesley
(1989).
10. J. Gleick, Chaos. Making a new science. New York, Viking Press (1987).
11. E. N. Lorenz, The Essence of Chaos. Seattle, University of Washington Press
( 1993).
12. I. Prigogine, &om Being to Becoming. Time and Complexity in the Physical
Sciences. San Rancisco, Freeman (1980).
13. I. Prigogine and I. Stengers, La nouvelle alliance. Me'tamorphose de la sci-
ence. Paris, Gallimard (1979).
14. H. Atlan, Entre le cristal et la fume'e. Paris, Seuil (1979).
15. S. Forrest (Ed.), Emergent Computation. Self-organizing, collective, and co-
operative phenomena in natural and artificial computing networks. Cam-
bridge, MIT Press (1991).
16. J. Umerez, Jerarquias Autdnomas. Un estudio sobre el origen y la natumleza
de 10s procesos de control y de formacidn de niveles en sistemas naturales
complejos. Universidad del Pais Vasco/ Euskal Herriko Unibertsitatea (1994).
17. H. H. Pattee, Instabilities and Information in Biological Self-organization.
In: F. E. Yates (Ed.), Self-organizing Systems. The Emergence of Order, New
York, Plenum Press, 325-338 (1987).
18. A. Moreno and J. Umerez, Downward causation at the core of living orga-
nization. In: P. B. Anderson, C. Emmeche, N. 0. Finnemann, P.V. Chris-
tiansen (Eds.), Downward Causation, Aarhus, Aarhus University Press, 99-
117 (2000).
19. H. A. Simon, The Architecture of Complexity. Proceedings of the American
Philosophical Society, 106, 467-482 (1962).
20. W.R. Ashby, An Introduction to Cybernetics. London, Chapman & Hall
(1956).
264

21. L. von Bertalanffy, General Systems Theory. Foundations, Development, Ap-

plications. New York, George Braziller (1968).
22. L. von Bertalanffy, Perspectives o n General Systems T h e o y . Scientific-
Philosophical Studies. New York, Georges Braziller (1975).
23. N. Wiener, Cybernetics or Control and Communication in the Animal and
the Machine. Cambridge, MIT Press (1948 [1961]).
24. J. Umerez, and A. Moreno, Origin of life as the First MST. Control hierar-
chies and interlevel relation. World Futures, 45, 139-154 (1995).
25. J. S. Mill, A System of Logic, Ratiocinative, and Inductive. London, Longan
and Green (1843).
26. G. H. Lewes, Problems of Life and Mind. London, Trench and Turber & Co.
Kegan Paul (1875).
27. C. L. Morgan, Emergent Evolution. New York, Henry Holt and Cn. (1923).
28. W. Wheeler, Emergent Evolution and the Development of Societies. New
York, Norton (1928).
29. J. Needham, Integrative Levels: A Revaluation of the Idea of Progress. Ox-
ford, Clarendon Press (1937).
30. A. Novikoff, The concept of integrative levels and biology. Science, 101, 209-
215 (1945).
31. R. W. Sellars, V. J. McGill, and M. Farber (Eds.), Philosophy for the Future.
The quest of modern materialism. New York, Macmillan (1949).
32. J. H.Woodger, Biological Principles: A Critical Study. London, Routledge
and Keegan Paul (1929).
33. V. Turchin, The Phenomenon of Science. New York, Columbia University
Press (1977).
34. S. Salthe, Evolving Hierarchical Systems. Their Structure and Representa-
tion. New York, Columbia University Press (1985).
35. F. Heylighen, The growth of structural and functional complexity during
evolution. In: F. Heylighen, J. Bollen, A. Riegler (Eds.), The Evolution of
Complezity. Dordrecht, Kluwer Academic (1999).
36. F. Heylighen, Evolutionary transitions: how do levels of complexity emerge?
Complezity, 6 (l),53-57 (2000).
37. K. Ruiz-Mirazo, A. Etxeberria, A. Moreno, J. Ibbnez, Organisms and their
Place in Biology. Theory in Biosciences, 119 (3-4), 209-233 (2000).
38. A. Etxeberria and J. Umerez, Organism0 y Organizaci6n en la Biologia
Te6rica Vuelta a1 organicismo 50 aiios despuhs? In: A. Ambrogi (Ed.) Medio
siglo de doble hdice. Edicions Universitat de les Illes Balears (in press) (2006).
39. L. von Bertalanffy, Modern Theories of Development. A n Introduction t o
Theoretical Biology. London, Oxford University Press (1933).
40. M. Beckner, The Biological Way of Thought. Berkeley, University of Califor-
nia Press (1968).
41. D. J. Haraway, Crystals, Fabrics, and Fields. New Haven, University of Yale
Press (1976).
42. J. K. Feibleman, Theory of Integrative Levels. The British Journal f o r the
Philosophy of Science, V (17), 59-66 (1954).
43. J. Umerez, Howard Pattee’s theoretical biology. A radical epistemological
 265

stance to approach life, evolution and complexity. Biosystems, 60 (1-3), 159-

177 (2001).
44. H. H. Pattee, Physical conditions for primitive functional hierarchies. In: L.
L. Whyte, A. G. Wilson and D. Wilson, Hierarchical Structures, New York,
American Elsevier, 161-177 (1969).
45. H. H. Pattee, The problem of biological hierarchy. Towards a Theoretical Bi-
ology 3, Drafts, Edinburgh, Edinburgh University Press (3), 117-136 (1970).
46. H. H. Pattee, Laws and constraints, symbols and languages. Towards a The-
oretical Biology 4, Essays, Edinburgh, Edinburgh University Press (4),248-
258 (1972).
47. H. H. Pattee, The physical basis and origin of hierarchical control. In: H.
H. Pattee, Hierarchy Theory. The challenge of complex systems, New York,
George Braziller, 73-108 (1973).
48. H. H. Pattee, Postscript. Unsolved problems and potential applications of hi-
erarchy theory. In: H. H. Pattee, Hierarchy Theory. The challenge of complex
systems, New York, George Braziller, 131-156 (1973).
49. H. H. Pattee (Ed.), Hierarchy Theory. The challenge of complex systems.
New York, George Braziller (1973).
50. V. Ah1 and T. F. H. Allen, Hierarchy Theory: A Vision, Vocabulary and
Epistemology. New York, Columbia University Press (1996).
51. H. A. Simon, The Organization of Complex Systems. In: H. H. Pattee (Ed.),
Hierarchy Theory. The challenge of complex systems, New York, George
Braziller, 1-27 (1973).
 SIMULATION AS FORMAL AND GENERATIVE SOCIAL
SCIENCE: THE VERY IDEA*

N U N 0 DAVID1, JAIME SIMAO SICHMAN2 & HELDER COELH03

‘Department of Information Science and Technology, I S C T E
Av. das ForGas Armadas, 1649-026 Lisboa, Portugal
E-mail: Nuno. DavidOiscte.pt
Computer Engineering Department, University of Sao Paulo
Av. Prof.Lucian0 Gualberto 158 tv.3, 05508-900 Sao Paulo SP Brazil
E-mail: Jaime.Sichman@poli.usp.br
3Faculty of Sciences, University of Lisbon
Bloco C5, Piso 1, Camp0 Grande, 1749-016 Lisboa, Portugal
E-mail: hcoelhoOdi. fc.ul.pt

The formal and empirical-generative perspectives of computation are demonstrated

to be inadequate to secure the goals of simulation in the social sciences. Simulation
does not resemble formal demonstrations or generative mechanisms that deduc-
tively explain how certain models are sufficient t o generate emergent macrostruc-
tures of interest. The description of scientific practice implies additional epistemic
conceptions of scientific knowledge. Three kinds of knowledge that account for a
comprehensive description of the discipline were identified: formal, empirical and
intentional knowledge. The use of formal conceptions of computation for describing
simulation is refuted; the roles of programming languages according to intentional
accounts of computation are identified; and the roles of iconographic programming
languages and aesthetic machines in simulation axe characterized. The roles that
simulation and intentional decision making may be able to play in a participative
information society are also discussed.

1. Introduction
The analysis of complex systems is being pursued with increasingly more
sophisticated information technologies. In particular, the area of computer
simulation has acquired a decisive role for analysing societies as complex
systems, leaving behind the history of simulation as a secondary method-
ology in the social sciences. The sources of analogy between agent-based

*The title of this paper is inspired by James Fetzer’s article “Formal Verification: The
Very Idea”. See Ref. 1.

266
 267

technologies and social scientific models fomented an unprecedented inter-

disciplinary effort, which has been creating countless interfaces of research,
across the computer and social sciences.
Several reasons exist for conducting a philosophical analysis of this sci-
entific domain. In general, social science simulation has contributed to an
inter- and multi-disciplinary scientific praxis,a thereby establishing new al-
ternatives to traditional scientific methodologies. This should lead to the
elaboration of new philosophical perspectives about the rules of the game
as they are played in ~ i m u l a t i o n . ~
Within the scientific community, the
existence of methodological aspects that deserve better analysis is recog-
nized. For some, the use of formal models, resulting from the computational
nature of simulation, has been considered not only an addition to the estab-
lished methods but the basis for the emergence of “proper social sciences”.‘
For others, the classical theory of computation does not support an ade-
quate model of reality for simulation in the social sciences, and therefore
the formal perspective of computation is not enough.d At any rate, the
difficulties in constructing methodological perspectives on simulation raises
interesting questions about the kind of scientific knowledge that simulation
is providing.
Once the philosophy of simulation is analysed, it becomes clear that
most essays do not take into account methodological and philosophical as-
pects of computer science, but are grounded mostly on aspects of social
science. Among the diversity of perspectives that may be adopted, most
should lead to recognizing additional ways to understand the concepts of
computation and programming languages. However, few if any philosophi-
cal analyses in this field considered theoretical and practical limits of com-
putation, as well as its new approaching challenges, well understood. Specif-
ically, among the questions confronted by social science simulation is the
extent to which formal and empirical methodologies are sufficient to de-
scribe the goals and methods of the discipline. In this paper we claim that
they are not.
As we demonstrated in David et al. (2005)’, simulation reveals new
conceptions about the kind of scientific knowledge that computers provide.

aOn the interdisciplinary structure of the scientific community in agent-based social

simulation see Ref. 2.
bSee the introduction to the issue of JASSS on epistemological perspectives on simula-
tion, Ref. 3. See also Ref. 4.
‘See specifically Ref. 5. See also the debate in Ref. 6.
dSee Ref. 8.
268

Meanwhile, the meaning of social simulation is strongly connected with

two traditional epistemic conceptions of scientific knowledge. The first
conception is the formal view of mathematical knowledge, reminiscent of the
computer science formal tradition and congruent with the idea of Simulation
as numerical calculation. The second refers to the experimental character
of simulation, insofar as scientists run their programs in computers like an
experimental set-up. Programs, in both senses, are viewed as descriptions
of social theories or phenomena that, unlike most theories in the social
sciences, are viewed as formal mathematical models.
The conflation of the formal and the experimental perspectives of simu-
lation lead many scientists to pose simulation as a way of testing represen-
tations of social theories or phenomena that could, in principle, be deduced
from general principles. This is based on three tacit methodological as-
sumptions that draw on the formal tradition of classical computational
theory, which we will challenge throughout this article:

Assumption 1 . The process of executing a program in a computer can

be formally modelled, and thus understood as an automatic process of for-
mal inference. This position was explicitly advocated by Epstein ( 1999)l0,
drawing directly on the Turing-Church thesis.e
Assumption 2. The sensitivity of complex models to initial condi-
tions does not permit deducing its simulated results from general princi-
ples, although it would be possible to do it in principle. Casti, for instance,
described this conception in a very explicit way:
“In principle, one could trace all the statements in the program repre-
senting the simulation and discover exactly why things unfolded the way
they did (. . . ) it just cant be done in practice.” (Casti, 2001, p.14, our
emphasis)”
Assumption 3. The experimental character of simulation in the so-
cial sciences can draw on the experimental character of simulation in the
natural sciences.f Insofar as the sensitivity of general complex models ex-
plains the inability to deduce its simulated results from general principles,
the sensitivity of social-scientific models explains similarly the inability to
deduce its simulated results from general principles. Casti, for instance,
tacitly conflated the present and the previous assumptions as follows:

CForan introduction to classic computer theory, also known as the Church-Turing thesis,
see e.g. Ref. 9.
‘For a comprehensive critic see Ref. 8; see also Ref. 12.
 269

"For the first time in history we are in a position to do bona fide

laboratory experiments on these kind of complex systems. (. . . )
We can use these surrogates as laboratories for carrying out the
experiments needed to be able to construct viable theories of com-
plex physical, social, biological and behavioural processes." (Casti,
1997, p.35, our emphasis)13
Recently, a number of essays debated epistemological perspectives on
social science simu1ation.g Either by viewing simulation as a process of
imitation12, stylized facts 14, or intentional adequacy between programs
and theory', the tendency is to emphasize the interpretative character of
social science in simulation, notwithstanding its application as a useful
methodology for approaching complexity. The essay presented by David
et al.', in particular, advocates the perspective of intentional computa-
tion as the approach able to comprehensibly reflect the multiparadigmatic
character of social science in terms of agent-based computational social sci-
ence. Conversely, essays presented by Moss and Edmonds15 or Boero and
Squazonni" argue that simulation provides the social sciences with a pow-
erful instrument to generate empirical evidence, and thereby, contribute to
better social sciences. Whereas these two trends may not be contradictory,
the formal character of computer science pervades the scientific culture of
simulation, as well as most of its methodological arguments. However, the
formal perspective does not seem to be compatible with any such views.
In this chapter the classical account of computation is demonstrated to
be inadequate to secure the goals of simulation in the social sciences. Sim-
ulation does not resemble formal deductive demonstrations or generative
mechanisms that explain how certain agent-based models are sufficient to
generate emergent macrostructures of interest. The justification of results
implies additional epistemic conceptions of scientific knowledge, which are
tacitly being used in the discipline. Three kinds of knowledge that account
for a comprehensive description of the discipline from an epistemological
point of view are identified: formal, empirical and intentional knowledge.
Intentional knowledge should be considered an outcome of an experimental
exercise, albeit not empirical, acquired within a context of limited consen-
sus.
The structure of our argument in this chapter is composed of two parts.
In the first part the formal and generative perspectives of computation in

gSee the introduction to the issue of JASSS on epistemological perspectives on simula-

tion, Ref. 3.
270

literature are recalled, proceeding afterwards to refuting the use of such con-
ceptions for describing the scientific practice of simulation. In the second
part the role of programming languages in simulation, according to inten-
tional accounts of computation, is discussed. Two types of programs and
programming languages in simulation are identified: Programs as text and
programs as icons; and languages as abstract machines and languages as
aesthetic machines. The use of abstract languages confirms that the method
of simulation incorporates formal and empirical methodologies. The use of
aesthetic languages demonstrated that it depends fundamentally on inten-
tional methodologies. The roles that intentional decision making may play
in a participative information society are also discussed.

2. An Ontological Confusion
Conventional methodologies of computer science model the mechanism of
executing a program in a computer like a process of formal inference. In
complexity sciences, the process of running a program was described not
only as an automatic inference procedure but as a formal deductive proce-
dure itself. The computer is seen as a mechanism of formal calculus, where
programs represent mathematical functions that map inputs into outputs.
The calculus can be modelled in several ways, one of which ascribes the
computer the capacity to prove first-order theorems according to a fixed
vocabulary. Considerations of brevity and simplicity lead us to call this tra-
dition the FDE argument, as per ‘Formal Deduction through Execution’.
Notwithstanding, our goal is quite the opposite, namely to demonstrate
that simulation shall not be legitimized under the presumption of resulting
from a calculus of formal inference. Additional conceptions of knowledge
are needed.

2.1. Against Genemtive Suficiency of Growing Artificial

Societies From the Bottom Up
The relationship of simulation with the formal perspective of computation
was rigorously advocated by Epstein (1999, p.44)lo in his account of “gen-
erative social science”, where he writes:h
“ . . . if one accepts the Church-Turing thesis then every computa-
tion - including every agent-based computation - can be exe-
cuted by a suitable register machine. It is then a theorem of logic

hThe concept of generative social science was also adopted in Ref. 17.
 271

and computability that every program can be simulated by a first-

order language.”
The point is the following:
“ . . . for every computation there is a corresponding logical deduc-
tion, and this holds even when the computation involves ‘stochastic’
features since, on a computer, these are produced by deterministic
pseudo-random number generation. Even if one conducts a statis-
tical analysis over some distribution of runs, each run is itself a
deduction.” (Epstein, 1999, p.44)lo
The methodological point of social science simulation would be to gener-
ate observed social phenomena in computers and thus deductively explain
the social phenomena, insofar as there is an intellectual tradition upon
which “we deduce propositions expressing observations from other more
general propositions” . i Accordingly, scientists “seek to explain macroscopic
social phenomena by generating it in an agent-based computational model”.
Moreover, “in that event, we can claim that they [the explanations] are
strictly deductive.” (p.43, our emphasis and brackets)”
We may state Epsteins argument of generative sufficiency as follows:

The Generative Sufcciency Argument. Agent-based models pro-

vide formal demonstrations that a given microspecification is a suficient
condition, albeit not a necessary condition, to generate a macrostructure
of interest.

And this leads Epstein to conclude:

‘‘J?rom an epistemological stand point, generative social science, while
empirical, is not inductive, at least as that term is typically used in the
social sciences.” (1999, p.43)lo

2.2. ‘Generative’ from a Philosophy of Computer Science

Perspective
The arguments of Epstein can be analyzed from the perspective of the phi-
losophy of computer science. The term ‘formal’ is ubiquitous in computer
science. Smithlg, for instance, has acknowledged the ambiguity of the term:

‘Epstein‘s account of scientific explanation seems to be inspired by the work of classical

empiricists, such as in Ref. 18.
272

“People may believe that developing an idea means formalizing it,

and that programming languages are formal languages, and that
theorem provers operate on formal axioms - but few write ‘formal’
in formal axioms or daily equations. Moreover, a raft of different
meanings and connotations lie just below the surface. Far from
hurting, this apparent ambiguity has helped to cement popular
consensus. Freed of the need to be strictly defined, formality has
been able to serve as a lightning rod for a cluster of ontological
assumptions, methodological commitments, and social and histor-
ical biases. Because it is tacit, goes deep, has historical roots, and
permeates practice, formality has been an ideal foil, over the years,
with which to investigate computation. Almost a dozen different
readings of ‘formal’ can be gleaned from informal usage: precise,
abstract, mathematical, a-contextual, digital, explicit, syntactic,
non-semantic, etc.”

In a Church-Turing theoretic account, the meaning of ‘formal’ appears

connected to the antisemantical reading mentioned above. That is, the idea
that a symbolic structure is formal just in case it is manipulated indepen-
dently of its semantics, in which it is assumed that a theorem is derived by
an automatic inference regimen.
To some extent, the pervasive use of the term ‘formal’ arises from con-
flating the terms ‘program computation’ and ‘program execution’ into one
single meaning, conveying the same ontological status to two fundamentally
distinct p r o c e ~ s e s The
. ~ concept of ‘abstract machine’ in classical compu-
tation is conflated with the concept of ‘physical machine’ in software en-
gineering. The observed behaviour of a program executing in a computer,
which should be the subject of research in the first place, gives way to
a computation that is, in essence, a formal model or a theory itself. As
Epstein himself states (1999, p.44)lo, “each run is itself a deduction” -
the actual execution of a program in a physical machine is a theory or a
computation in itself.
It would be reasonable to question whether the idea of Epstein is not
to illustrate anything more than an ideal perspective of computation in
real computers. Yet, his efforts are not limited to illustrating that sim-
ulation implies deductive and generative conceptions of scientific research
but that simulation implies a new kind of scientific empirical research. The
point of simulation, in such a sense, is to provide agent-based specifications
for which the corresponding program execution should generate patterns
 273

that match empirical data - for instance, to implement in a computer an

agent-based archaeological model that succeeds in generating behaviours
that match empirical data and thus explain why the Kayenta Anasazi pop-
ulation of Long House ValleyJ vanished at some point from the valley (1999,
p.44)’O. The role of computer science, therefore] is one of formal inference,
whereas the role of social science simulation is to use computers to generate
behaviours that should match social scientific data.
Apart from the arguments of Epstein, there is wide evidence confirm-
ing a tacit association of simulation with formal deduction, for instance, by
suggesting the high levels of control and objectivity that scientists attribute
to simulation results. The claim of Castill that it is possible to trace all
statements in the simulation is obvious evidence. Prietula et a1.” advocate
that: “computational models are generally less noisy, easier to control, more
flexible, more objective”. Sawyer explains: “the contributions of artificial
societies to sociological theory will be primarily to theories that are char-
acterized by logical rigor (. . . ) ] which allow precise and logical deductions
from abstract principles to empirical hypotheses (. . . ) Artificial societies
resemble axiomatic theory in the sense that their propositions are explic-
itly stated in the form of algorithms or program code and valid derivations
may be drawn systematically by running the program” (2003, p.332)’l. For
Axelrod, one of the advantages of simulation is that “there are no messy
problems of missing data or uncontrolled variables as there are in experi-
mental or observational studies” (1997, ~ . 2 7 ) ’ ~ .
The most cited sentence of Axelrod, describing simulation as a bridge
between induction and deduction, is actually not far from the idea. Whereas
it is claimed that simulated data must be analysed inductively, it is sug-
gested that the data are a necessarily valid consequence of the rules speci-
fied:

“Simulation is a third way of doing science. Like deduction, it

starts with a set of explicit assumptions. But unlike deduction, it
does not prove theorems. Instead, a simulation generates data that
can be analyzed inductively. Unlike typical induction, however, the
simulated data comes from a rigorously specified set of rules rather
than direct measurement of the real world. While induction can
be used to find patterns in data, and deduction can be used to find
consequences of assumptions, simulation modelling can be used as

J A small region in northeastern Arizona.

2 74

an aid to intuition.” (Axelrod, 1997, ~ . 2 5 ) ~ ~

However, insofar as the data can be analysed inductively, it seems dif-

ficult to conceive how it was generated in the first place without “direct
measurement of the real world”. Computers in this sense do not belong to
the “real” world. That is to say, the tacit methodological assumption of
social science simulation would propose the same ontological status to both
a program execution in a physical machine and a formal computation in an
abstract machine. We shall call this assumption the FDE argument, as per
Formal Deduction through Execution:

The Argument of Formal Deduction through Execution. It is

based on three methodological beliefs, which we will show to be contradic-
tory with one another:
A) The mapping of classical computer theory into the logic of the
method of social science simulation. B) The conflation of the terms ‘com-
putation’ and ‘execution’. C ) The experimental character of simulation,
upon which an unexpected result can be a reflection of a mistake in the im-
plementation (bug) or a surprising consequence of the program itself (see
e.g. Axelrod, 1997, ~ . 2 7 ) ~ ’ .

2.3. Refutation of Formal Deduction through Execution

In this research, it is claimed that the role of computation in social sim-
ulation is not one of formal deduction. The goal is to refute the FDE
argument. From a scientific-philosophical perspective our goal may seem
somehow trivial. However, we demonstrated that the formal perspective
of computer science is recalcitrant within social science simulation. More-
over, the objection to the FDE argument suggests yet another objection
to current philosophical thinking in the literature: The objection to char-
acterizing simulation as a basic and alternative epistemic conception to
deduction or induction.
The objection to the FDE argument parallels an old, but still om-
nipresent, debate in computer science, centered around the merits of the
so-called “formal verification of programs”. The merits of formal methods
for verifying programs became particularly controversial by the end of the
Eighties and in the beginning of the Nineties after James Fetzer published
an eloquent article in the ‘Communications of the ACM’’>23.On one side
of the debate were gathered those who considered that computer program-
 275

ming was reducible to mathematics - inspired by Hoare or Dijkstrak -

and the other side was composed of those who saw it as applied mathemat-
ics or empirical science. The claim of Fetzer was that the formal verification
project carried misleading ideas. The seminal work of Hoare is an obvious
example:

“Computer programming is an exact science in that all the p r o p

erties of a program and all of the consequences of executing it in
any given environment can, in principle, be found out from the
text of the program itself by means of purely deductive reasoning.”
(Hoare, 1969, p.576, our emphasis)25

These ideas often turned out to be misleading. For instance, it was often
claimed that the intended behaviours of computers could be completely
specified and verified before the corresponding programs were executed on
specific computers, by means of purely formal methods. Computer science
would be viewed as pure mathematics rather than as applied mathematics
or empirical science.
Fetzer‘s philosophical refutation of the formal verification project con-
sisted of distinguishing two kinds of programs: those that are and those that
are not in a suitable form to be compiled and executed in a machine. The
difference between the programs is that the former is verified by reference to
abstract machines, whereas the latter require the existence of compilers, in-
terpreters and target machines. Compilers, interpreters and processors are
properly characterized according to specific target physical machines. Inso-
far as a program in an abstract machine does not possess any significance
for the performance of a target machine, the performance of that program
can be verified formally and conclusively. Conversely, to the extent that the
performance of a program possesses significance for the performance of a
target machine, that program cannot be conclusively verified a priori. The
program must be verified empirically by means of program testing. Hence,
the formal verification project is not viable.’ Program testing should be
the crucial technique to ascertain the proper behaviour of programs and
computers, notwithstanding the use of formal methods during the stages of
analysis and design.
Our goal is to informally reduce the FDE argument in social science sim-
ulation to the formal verification project in computer science. Consider the

kSee e.g. Ref. 24.

‘For more details see Ref. 1 or 23.
276

formal verification project according to its most radical terms. The intent
would be to create formal methodologies that could guarantee that a given
specification would correspond to the behaviour of a program executing in
a computer. That is, to find deductive procedures to verify conclusively
the correctness of a program P in relation to a specification F : I 4 0,
in order to guarantee that the execution of P with inputs I would result
exactly into the specified outputs 0. The following argument reduces FDE
to the formal verification project.

The Argument of FDE Refutation. Consider a specification F1 :

I1 -+ 01 and a program P1 as text that can be read, edited, printed.
The computation of P1 with inputs I1 is denoted by Pl(I1) = 01 and
the execution of P1 after implementation is denoted by Pl'(I1) M 01.
Suppose that Pl(I1) = 01, according to a proof of partial correctness.m
Suppose, however, that Pl'(I1) M 0 2 , that is the computation and the
execution of P1 leads to different results, compatible with assumptions B
and C in the FDE argument. However, according to assumptions A and B in
the FDE argument, there is a specification F2 : I1 + 0 2 and some program
P2 such that the computation with input I1 leads to 0 2 , i.e. P2(11)= 0 2 .
So, a specification F2 : I1 -+ 0 2 exists such that the execution of P1
and the computation of P2 satisfy F2. The formal verification project is
thus possible: the behaviour of P1 execution (as well as P2 computation)
necessarily corresponds to the specification F2 : I1 -+ 0 2 .

In short, from a methodological point of view, the FDE argument can-

not be sustained. At the very least it is misleading. Even though it does
not suggest that both the computation and the execution of P1 necessar-
ily give the same outputs, which would prima-facie instantiate the formal
verification project, it may suggest that the execution of P1 necessarily cor-
responds to a formal computation of some program P2, which is a method-
ological absurdity. The formal perspective of computation is not enough to
support an adequate model of reality for simulation in the social sciences.
The logic of simulation implies distinct types of program verifications that
reflect epistemological distinctions in the kind of knowledge one can have
about programs. One obvious type of knowledge is empirical knowledge
obtained through program testing. Another one is intentional knowledge,8
which is discussed in subsequent sections.

mOn proofs of partial correctness see e.g. Ref. 25.

 277

A final comment emerges regarding the perspective of Axelrod22 that

poses simulation as a contrast to both induction and deduction, which is
after all not far from the formal perspective of Espstein. Whereas Axelrod
defines induction as the “discovery of patterns in empirical data”, deduction
is understood as the specification of “a set of axioms and proving conse-
quences that can be derived from those assumptions” (1997, ~ . 2 4 ) Yet,~~.
there is no reason for not viewing deduction as a kind of empirical enquiry.
Popper26,as many other ‘deductivists’ in the philosophy of science, would
say that there is no such thing as induction.

Inference through Em piric a1 E n q u i r y

I I de duc tive or
deductive@ ar
inductively or
abductive ly ...
i
!
1

I
I
hductive@ar
abductively ...

Natural Process

Figure 1. A theory square.

Inference through Empirical Enquiry

I I de duc tivc ly or
deductive@ or I
inductively or I I hductivelyor
abduct& ly ...
I I hbductively ...
I
! I
Program Execution

Figure 2. A second theory square.

Consider Figure 1. The theory square represents a traditional epistemic

question in the philosophy of science with respect to empirical science:
Are correlations between natural properties enquired inductively or deduc-
tively? Likewise, it is equally fair to use a theory square in relation to the
behaviour of a program executing in a computer, just like in Figure 2. In-
deed, whether the enquiry may be inductive, deductive, or even abductive
depends on the relevant methodological conception of scientific enquiry,
which is by no means a specific dilemma of simulation. To define the epis-
278

temic specificities of simulation based on such contrast does not seem to be

significantly informative to the point in question.

3. The Role of Programming Languages

An informative way to analyse the epistemic status of simulation is to ex-
amine how simulations are programmed. Program implementation plays
a fundamental part in constructing a simulation. The role of intentional
methodologies becomes crucial once it is realised that, rather than one kind
of program and programming language, two kinds of program and program-
ming language are used in the implementation of a simulation. Simulations
are constructed by means of successive language translators, from program
to program, by using simulation platforms, compilers and interpreters. In
computer science we usually view a programming language as a n abstract
machine and a program as a textual and static entity, which may be read,
edited, printed. However, in social simulation we have identified yet another
type of programming language and kind of program, which involve the use
of intentional knowledge. At least two kinds of programming languages are
used in the process of implementing simulations, iconographic and textual
programming languages.

Iconographic Programming Languages. Iconographic program-

ming languages consist of a set of subjective rules that model the behaviour
of an aesthetic machine. A subjective model, eventually associated with a
specific domain composed of organisational or aesthetic abstractions, such
as groups of agents, grids, movement, constraints, roles, levels, messages,
societies, or specific behaviours, ‘segregation rules’, ‘sexual rules’ and ‘ge-
netic crossing’, ‘culture’, ‘race’, ‘influence’, ‘friendship’, ‘innovation’, ‘state
nations’ or ‘political actors’. For instance, the CORMAS platform,” as well
as the Swarm simulation system,O support aesthetic human-machine inter-
faces to specify interactions between individuals and groups sharing renew-
able natural resources. The commands in the programming languages are
icons represented on the screen rather than structures in textual languages.
A program is a set of selected icons with no definitive explicit order. The
user selects icons with the mouse, after which an automatic code generator
transforms icons into sections of code in high-level programming languages,
such as SmallTalk or Objective C. However, the first order logics of classic

nCornmon-pool Resources and Multi-Agent Systems, see http://cormas.cirad.fr.

Osee http://www.swarm.org.
 279

computer theory can hardly describe these transformations. Moreover, the

sections of code are later linked in some arbitrary way by the user. The
specific mapping from icon-level language to high-level language is subjec-
tive, domain specific and validated according to a limited level of consensus.
In the case of participative-based simulation, stakeholders may be involved
in the specification of iconographic programs, but hardly in the semantic
process of mapping iconographic programs to high-level programs.

Textual Programming Languages. Textual programming languages

include the usual high-level and low-level languages. A high-level language,
as defined by Fetzer2?, is a set of abstract rules that model the behaviour of
an abstract machine. High-level languages have clear formal semantics, and
contrary to icon-level languages, the meaning of their commands must not
be subjective. The advantage of programming with high-level languages,
such as Java or Objective C, is that there is a one-to-many relationship be-
tween the commands that can be written in a high-level language and the
counterpart operations that are performed by a machine executing them,
on the basis of their translation into machine language. The function of
interpreters and compilers is to create a causal mechanism so that programs
written in high-level languages may be mapped to low-level languages and
later executed by target machines whose operations are causally affected
by machine code, which usually consists of sequences of zeros and ones.27
A low-level language is a set of abstract rules that model the behaviour of
abstract or target machines. Typically, the lowest-level language program-
mers use is Assembly language, where there is more or less a one-to-one cor-
respondence between commands and operations. Low-level programming
languages therefore play two roles: First, that of an abstract machine, in a
way analogous to high-level languages but where, second, unlike high-level
languages, there is a one-to-one causal relationship between the commands
that occur within a programming language and the operations performed
by a target machine. The programming language stands for a virtual ma-
chine that may be understood as an abstract entity, which may or may not
be causally connected with a target machine.

The distinction between programs as text and programs as icons, as well

as between abstract and aesthetic machines, reveals the intentional charac-
ter of social simulation methodologies. Whereas the consistency between
two different abstract machines can be specified formally and verified em-
pirically, the consistency between abstract and aesthetic machines must be
280

verified intentionally. Indeed, the modelling of target machines by means

of abstract machines can be defined according to formal logic, drawing on
classic computer theory. Hence, the relative consistency between the high-
level and low-level abstract machines can be demonstrated formally, as well
as tested empirically against the observed behaviour of the program. More-
over, insofar as most high-level programming languages stand for abstract
machines rather than physical machines, we can say that the process of
implementing a high-level program involves the construction of a sequence
of embedded models that are causally connected to a target machine.
In contrast, whereas abstract machines are specified with well-defined
formal semantics, the meaning of iconographic languages and aesthetic ma-
chines is negotiated intentionally by the members of the team implementing
the simulation, as well as by the stakeholders involved. Since iconographic
languages must be mapped into high-level, and ultimately into low-level
languages, it becomes clear that the process of implementing icon-level
programs involves the construction of a sequence of embedded models con-
nected intentionally to a target machine. The target machine is modelled
intentionally according to a limited level of consensus and tested experimen-
tally, albeit not empirically, against the observed behaviour of the icon-level
program.
From this point of view, the implementation of a program can be viewed
as the action of embedding models within models, where the notion of em-
bedding may be envisioned as an intentional, causal or logical relation. In
Figure 3, Fetzer’s diagram of language embedding27 is expanded with icon-
level programs and aesthetic machines. The thin arrows represent a possible
relation between programs and machines represented by programming lan-
guages. The thick arrow represents an actual relation between a low-level
program and a target machine. The series of three black dots stands for
the possible existence of automatic code generators, compilers and inter-
preters that effect some causal connection between programs/machines at
different levels of embedding. In addition, the series of unfilled dots stands
for the existence of implementation teams and stakeholders that exercise
intentional connections between abstract and aesthetic programs/machines
at different levels, according to a limited level of consensus.
The use of iconographic programming languages demonstrates that the
logic of simulation incorporates formal and empirical methods, but largely
surpasses the use of formal and empirical methodologies. The results of
a simulation are outcomes of experimental set-ups, but the results of the
experiments can hardly be represented by material conditions of necessity
 281

PROGRAMS MACHINES

HIGH LEVEL

LOW LEVEL

Figure 3. Programs and languages 89 models - an extension t o the diagram of Fetzer

with icon-level programs and aesthetic machines.

relating facts about the objective behaviours of the program. The results
are appropriately characterized by conditions of intentionality that relate
aesthetic components in the program, negotiated according to a limited
level of consensus.

4. A Different Idea of Simulation

We shall recall the traditional assumptions of simulation in the social
sciences, presented in the introduction, which we have been challenging
throughout this article. Firstly, whereas it is true that the process of ex-
ecuting a program can be modelled according to a mechanism of formal
inference, it does not seem to be significantly relevant to the methodology
of social science simulation. Secondly, whereas sensitivity to initial con-
ditions is an important issue in the social sciences, it is not the primary
reason explaining the inability to deduce simulation results from general
principles, at least from a technical point of view.
A tremendous semantic gap exists between the formal-empiricaland the
intentional perspective of computation, both of which are used to interpret
the behaviour of simulations. The misleading assumption of generative
social science is that both formal and intentional representations can be
legitimised deductively, insofar as the process of program execution can
be understood as a formal deductive mechanism. This reasoning is unsus-
tainable. Firstly, the vocabularies of the low-level abstract machine (e.g.
memory registers, bit logical operations), as well as the vocabularies of the
282

high-level machine (e.g. complex data structures, objects, graphics), are

not identical to the vocabularies of the icon-level machine (e.g. agents,
grid, movement, culture, segregation rules). The parts that those vocabu-
laries designate in the world are not the same; from a strict formal point of
view the consistency between machines is incommensurable. Secondly, al-
though the consistency between low-level and high-level abstract machines
can be verified empirically, the consistency between abstract and aesthetic
machines must be verified intentionally. But unlike the process of empirical
adequacy, no formal computational theory is available to justify the process
of intentional adequacy.P
The characterization of simulation as a research practice implies ad-
ditional epistemic conceptions of scientific knowledge that are being used
tacitly in the discipline. There are at least three different kinds of knowl-
edge that can be acquired from computer simulation: formal, empirical
and intentional knowledge. Intentional knowledge should be considered an
outcome of an experimental exercise, albeit not empirical, acquired within
a context of limited consensus. Social science simulation, like the social
sciences, is interpretative and diverse theoretically and methodologically.
To imagine that simulation could integrate the archipelago of the social
sciences, at least as far as that may depend on the establishing of wide
consensuses, like those found in the natural sciences, would be a mistake.
The conditions for the acceptance of a simulation depend on the particular
theoretical-methodological context of the social scientist, can be interpreta-
tive and subjective, and may depend on the socioeconomic and sociocultural
context. The perspective of intentional computation seems to be the one
able to reflect the multiparadigmatic character of social science into social
science simulation.
Finally, the observation that social science simulation is multiparadig-
matic demonstrates the need for participative contexts. As a growing in-
formation technology, simulation is being used to assess concrete socioe-
conomic and environmental/ecological problems. Simulation is a useful
methodology to approximate scientists and stakeholders. However, only
within a specific interpretative context can a specification or a program be
considered as a set of sufficient conditions to explain the observed behaviour

PMoreover, as we mentioned before, even if we ought to stay on the empirical plan

(which is not the case), whether the epistemic conception of inquiring the behaviour of
computers could be an inductivist or a deductivist one that would be by no means a
specific dilemma of simulation.
 283

of a simulation. Indeed, from a sociological perspective, simulation should

help us pose the process of science as critical thinking in a democratic
context.

Acknowledgements
Jaime S i m k Sichman is partially supported by CNPq, Brazil, grants
482019/2004-2 and 304605/2004-2.

References
1. James Fetzer, Program Verification: The Very Idea, Communications of the
ACM, 31, 1048-1063(1988).
2. Nuno David, Maria Marietto, Jaime S. Sichman, Helder Coelho, The
Structure and Logic of Interdisciplinary Research in Agent-Based Social
Simulation, Journal of Artificial Societies and Social Simulation, 7(3),
ihttp://www.soc.surrey.ac.uk/JASSS/7/3/4.htmli,(2004).
3. Ulrich Frank and Klaus Troitzsch, Epistemological Perspectives on Sim-
ulation, Journal of Artificial Societies and Social Simulation, 8(4),
ihttp://jasss.soc.surrey.a~.uk/8/4/7.html~ (2005).
4. Carlos Gershenson, Philosophical Ideas on the Simulation of Social
Behaviour, Journal of Artificial Societies and Social Simulation, 5(3),
ihttp://jasss.soc.surrey.ac.uk/5/3/8.htmli,(2002).
5. J. Kluver, C. Stoica and J. Schmidt, Formal Models, Social Theory and
Computer Simulations: Some Methodologi-
cal Reflections, Journal of Artificial Societies and Social Simulation, 6(2),
ihttp://jasss.soc.surrey.ac.uk/6/2/8.html~(2003).
6. Rosaria Conte, Bruce Edmonds, Scott Moss, Keith Sawyer, Sociology and
Social Theory in Agent Based Social Simulation: A Symposium, Computa-
tional and Mathematical Organization Theory, 7(3), 183-205(2001).
7. Nuno David, Empirical and Intentional Verification of Computer Programs in
Agent-based Social Simulation, Ph.D., University of Lisbon (in Portuguese)
(2005).
8. Nuno David, Jaime S. Sichman, Helder Coelho, The Logic of the Method
of Agent-Based Simulation in the Social Sciences: Empirical and Intentional
Adequacy of Computer Programs, Journal of Artificial Societies and Social
Simulation, 8(4), ihttp://jasss.soc.surrey.ac.uk/8/4/2.htmli,(2005).
9. C. Papadimitriou, Computational Complexity, Addison-Wesley (1994).
10. Joshua Epstein, Agent-Based Computational Models And Generative Social
Science, Complezity, 4(5), John Wiley & Sons, 41-59 (1999).
11. John Casti, Would-Be Business Worlds. Complexity, 6(2), John Wiley &
Sons, 13-15 (2001).
12. Gunter Kuppers and Johannes Lenhard, Validation of Simulation: Patterns
in the Social and Natural Sciences, Journal of Artificial Societies and Social
Simulation, 8(4), ihttp://jasss.soc.surrey.ac.uk/8/4/3.htmli,(2005).
284

13. John Casti, Would-be Worlds: how simulation is changing the frontiers of
science, John Wiley & Sons (1997).
14. Bernd-0. Heine, Matthias Meyer and Oliver Strangfeld, Stylised Facts
and the Contribution of Simulation to the Economic Analysis of Bud-
geting, Journal of Artificial Societies and Social Simulation, 8(4),
ihttp://jasss.soc.surrey.ac.uk/8/4/4.htmli (2005).
15. Scott Moss and Bruce Edmonds, Towards
Good Social Science, Journal of Artificial Societies and Social Simulation,
8(4), ihttp://jasss.soc.surrey.ac.uk/8/4/13.htmli(2005).
16. Riccardo Boero and Flaminio Squazzoni, Does Empirical Embeddedness
Matter? Methodological Issues on Agent Based Models for Analytical So-
cial Science, Journal of Artificial Societies and Social Simulation, 8(4),
ihttp://jasss.soc.surrey.ac.uk/8/4/6.htrnlj,(2005).
17. Joshua Epstein, Robert Axtell, Growing Artificial Societies: Social Science
from the Bottom Up, MIT press (1996).
18. C. G.Hempel and P. Oppenheim, Studies in the Logic of Explanation, Phi-
losophy of Science, 15, 567-579 (1948).
19. Brian Cantwell Smith, The Foundations of Computing, Smiths in-
troduction to a series of books that report his study of computing
in the books The Age of Significance: Volumes IVI. Available at
ihttp://www.ageosig.org/people/bcsmith/papersi(1996).
20. M. J. Prietula, K. M. Carley and L. Gasser, Simulating Organizations, MIT
Press (1998).
21. Keith Sawyer, Multiagent Systems and the Micro-Macro Link in Sociological
Theory, Sociological Methods and Research, 21(3), 37-75 (2003).
22. Robert Axelrod, Advancing the Art of Simulation in the Social Sciences,
Simulating Social Phenomena, Springer Verlag, 21-40 (1997).
23. James Fetzer, Computers And Cognition: Why Minds are not Machines,
Studies in Cognitive Systems, 25, Kluwer Academic Publishers (2001).
24. Edsger Dijkstra, A Discipline of Programming, Prentice-Hall (1976).
25. C. A. R. Hoare, An Axiomatic Basis for Computer Programming, Commu-
nications of the ACM, 12,576-580 (1969).
26. Karl Popper, The Logic of Scientific Discovery, Routledge Classics
(1935/1992).
27. James Fetzer, The Role of Models in Computer Science, The Monist, 82(1),
La Salle, Illinois 61301,20-36 (1999).
 A COMPROMISE BETWEEN REDUCTIONISM AND
NON-REDUCTIONISM

ERAYOZKURAL
Department of Computer Engineering
Bilkent University
Bilkent 06800 Ankara, Turkey
E-mail: erayo@cs.bilkent.edu.tr

This paper investigates the seeming incompatibility of reductionism and non-

reductionism in the context of complexity sciences. I review algorithmic infor-
mation theory for this purpose. I offer two physical metaphors t o form a bet-
ter understanding of algorithmic complexity, and I briefly discuss its advantages,
shortcomings and applications. Then, I revisit the non-reductionist approaches in
philosophy of mind which are often arguments from ignorance t o counter physical-
ism. A new approach called mild non-reductionism is proposed which reconciliates
the necessities of acknowledging irreducibility found in complex systems, and main-
taining physicalism.

1. Introduction
New mathematical techniques have enabled scientists to gain a better un-
derstanding of non-linear dynamical systems, while similar advances have
occurred in computer science for constructing models of natural phenom-
ena as information processing systems. A particularly interesting research
direction in complexity sciences is the study of conditions under which self-
organization occurs. '
Complexity theorists pay special attention to the
concept of emergence. Gershenson and Heylighen give a typical definition
of emergence:
Emergent properties characterize a system that cannot be reduced
to the sum of its parts.
The authors then give the most striking example that an animate cell is
composed of parts that are inanimate. Thus, to be animate must be an
emergent property since the system cannot be explained in terms of its
parts. In the same paper, the authors show analysis/reductionism as a
shortcoming of classical science, which is indeed the case.

285
286

However, the identification of non-reductionism with this particular no-

tion of “emergent property” leaves something to be desired. In this paper,
I will argue that the mechanical view of nature is compatible with Hey-
lighen’s definition of complexity:

complexity of a system increases with the number of distinct com-

ponents, the number of connections between them, the complexities
of the components, and the complexities of the connections.

And I will suggest that complex phenomena like emergence may be under-
stood through the mechanical view.
The rest of the paper is organized as follows. In the next section, I will
review the relevance of the information theory approach to quantification
of complexity. In Section 3, I will highlight the relation of dualist and non-
reductionist theories in philosophy of mind, suggesting that extreme non-
reductionism leads to extra-physical notions. In Section 4, I will outline
mild non-reductionism as a physicalist approach for resolving the tension
between reductionism and non-reductionism.

2. Information theory and complexity

2.1. Basic definitions
Algorithmic information theory (AIT) gives a precise mathematical defini-
tion of complexity. Algorithmic complexity (or entropy) H ( z ) of a message
z is the length of the shortest program that generates it. a This definition
captures the essential difference between the complexities of
0101010101010101010101010101010101010101010101010101010101010101
1011100001110100110110110000010011011010100010111000101111000110
While the first sequence can be generated by a simple program consisting of
a loop, it takes a longer program to create the second random sequence. In
contrast to classical information theory, this definition has also the advan-
tage that it is an absolute quantity per message, rather than characterizing
a random source. The algorithmic definition is deemed more fundamental
than classical information theory for classical information can be derived
from algorithmic information.
This definition has not been well received in the philosophy community
due to some basic misunderstandings of the theory. Obviously, for there

aFormally, H ( z ) = min(lp1 : U ( p ) = z} where U is a prefix-free universal computer,

U ( p ) is the output of program p and Ipl is its length. 3,4
 287

to be a program, a programming language has to be chosen. Due to the

invariance theorem, this choice makes only a constant difference in com-
plexity. However, people have been confused over whether this means that
the definition is not absolute as it is claimed.
An achievement of AIT is the proof that other mathematical definitions
of randomness and incompressibility are identical. That is to say, one
cannot find a (much) shorter program for a random sequence, with the
consequence that minimal programs are random. This discovery has been
confusing as well, causing people to claim that they can denote with a
single symbol a message of any length (in which case nothing would be
random). While this can be done, such a decoder would be useful for
decoding only one message, while a universal computer can handle any
message. If one “hacks” a universal computer to achieve this, one will
have to encode the additional information inside the computer, and thus
algorithmic information is not lost.

2.2. Physical complexity

We offer the following purely physical metaphors to make it easier for the
reader to form an intuition about the physical relevance of the information
theory point of view.

0 physical message complexity: volume of the smallest machine that

can output a message in any fashion. The condition is that the
output message is encoded as is, it should be written symbol for
symbol. Like the AIT definition, with increasing complexity, we
would need a larger machine to achieve this.
0 constructive complexity: volume of the smallest machine that can
precisely build a desired object (up to physical limits, or required
accuracy).

These metaphors are given to furnish the understanding of algorithmic

complexity as a physical concept. Every bit in a sequence of bits extends
in space, and must exist as a distinguishable physical property. That is to
say, length may be conceived as the physical length.
Considering physical message complexity, the volume of this machine
is similar to the length of a minimal program in a programming language.
However, the physical definition is literally universal. The smallest such

bPlease note that a formal argument is beyond the scope of this paper.
288

volume does not depend on the choice of a computer; since it effectively

uses our physical universe as the universal computer in the definition of
algorithmic complexity. Likewise, we would not think that we could design
a storage medium that is more compact beyond a certain physical limit.
More naturally, one may consider algorithmic complexity in relation to
thermodynamic entropy. A discussion of various approaches to physical
complexity may be found in Bennett’s work.

2.3. Adequacy of algorithmic complexity

Algorithmic complexity beautifully captures our basic intuitions about com-
plex systems. Recalling Heylighen’s definition, we can view a complex sys-
tem as a graph (or network) where vertices are components and edges are in-
teractions. According to subadditivity of algorithmic information, the com-
plexity of a message is the sum of the complexities of its components, with
+
the addition of a constant term. That is, H ( z ,y) = H ( z ) H(y/z) 0(1)+
where H(y/z) is conditional (or relative) algorithmic complexity, H ( z ,y)
denotes the complexity of concatenation of messages z and y, and 0(1)is
a constant. Conditional algorithmic complexity H(y/z) is the length of the
shortest program that generates y given the shortest program for z. The
complexity of each component therefore does add to overall complexity.
However, it is not a simple sum of the complexities of parts. If the com-
ponents are identical, then there will be little increase. If the components
are all very different, each will contribute a large amount of algorithmic
complexity. The algorithmic complexity of the graph of the system will
also increase with the number of vertices, the number of edges, and the
randomness of the topology of the edges.

2.4. Shortcomings of algorithmic complexity

The definition of algorithmic complexity neglects the temporal dimension
which is indispensable in the study of dynamical systems. Without time, we
could not formulate such concepts as trajectory and stability. Algorithmic
complexity does not distinguish between passive stores of information and
active users of information, thus it has to be supplanted by descriptions of a
dynamical system in the study of many natural complex systems. However,
there are theorems connecting algorithmic information theory and dynam-
ical systems. Also of interest is the relation of algorithmic complexity

=A regular topology takes fewer bits to specify than an irregular one.

 289

and cellular automata dynamics. lo On the other hand, time aspect can be
put into new definitions of complexity. The “logical depth” definition of
Bennett incorporates time. l1
There are significant features of biological systems that algorithmic
complexity does not model. For instance, algorithmic complexity disre-
gards redundancy in the data, while redundancy and fault-tolerance are
fine features of biological systems. Biological systems also seem to achieve
optimizations other than reducing program-size, e.g., time and energy op-
timizations. Critics have also argued that algorithmic complexity does not
model the interestingness or usefulness of a message. In classical informa-
tion theory, the notion of “relevant information” has addressed the problem
of interestingness. l2 Hutter has pursued in detail an approach based on
algorithmic probability including agent models that maximize an arbitrary
utility function. l3
Another shortcoming is the fact that algorithmic complexity is uncom-
putable. Therefore we always have to stick to much simpler approximations
if we need to explicitly calculate the complexity of a message. This is not
a shortcoming in theory, only a shortcoming in practice.

2.5. Applications and related work

Mutual algorithmic information H ( s : y) between two messages quantifies
how much algorithmic complexity is shared. In the visual descriptions of
a car and a plane, “wheel” could be part of shared information. The joint
information H ( a , y ) and mutual information are related via H ( s : y) =
+ +
H ( s ) H(y) - H ( s ,y) O(1). Mutual information measures how much
easier it is to compress two strings together, than separately.
Mutual information can be used to quantify the amount of knowledge a
biological agent has about the world by the mutual information between the
agent and the world. While this a good characterization of the amount of
knowledge, it says nothing about the dynamical aspects. The operational
facets of intelligence could be incorporated by modelling which events it
could predict, an idea that may be important since some theoretical neu-
roscientists propose that it is one of the main functions of the brain. l4
It is also worth mentioning that a minimal program can be used to make
predictions; Solomonoff presents a particularly accessible introduction to
applications of algorithmic complexity in A1 using inductive methods. l5
Some philosophical applications of AIT may be found in a paper by Chaitin
including a thesis about perception. Another interesting work relating
290

the information theoretic point of view to complexity sciences is due to

Bialek et al. 17, suggesting that predictive information has a close relation
to the measures of complexity in dynamical systems theory.

3. Non-reductionism and physicalism

Proponents of non-reductionism often bring the human mind as an exam-
ple of an irreducible system. In philosophy of mind, the tension between
reductionism and non-reductionism has been a major driving theme of con-
temporary research. A motivating question is to find a theory in which
physicalism becomes compatible with non-reductionism. Physicalism is the
view that mental events are ultimately physical, i.e., material. I will use
the term extra-physical for those concepts which do not befit a physical
description.
Complexity sciences may be seen as embracing a physicalist methodol-
ogy, for the tools of a complexity scientist are usually physical: trajectories,
dynamics, interactions, entropy, etc. This attitude does not carry over to
philosophy of complexity. In philosophical discussions we often see a drift
towards some old concepts in philosophy such as dualism.

3.1. Non-reductionism in philosophy of mind

A fundamental problem in philosophy of mind is that it does not seem there
is a place for qualia such as color, or pain in a human organ. I can see green,
but when I open up a brain I do not find any green, only neural tissue. On
the other hand, there seems to be a strong relation between experience and
the physical state of brain. For instance, we can invoke certain sensations
by stimulating the brain with electrodes. In philosophy of mind, we describe
this relation as: mental events supervene on physical events. We can speak
of a general principle of supervenience in philosophy of mind: there are no
mental differences when there are no physical differences.
However, it is difficult, if not impossible, to fathom a relation between
this complex, impenetrable organ and our basic perceptions. Thus, the
philosophers have been creating new ideas for centuries, imagining that the
mind must be something else than the brain. This has traditionally had
other reasons such as the belief that we are special. Descartes thought that
while the animals were automata, we were more than mere machinery. One
reason for this was the apparent inadequacy of a mechanism. A machine
could do only one kind of a thing, but the human mind seemed to be capable
of anything. Thus, the human mind could not be a machine. Of course,
 291

Descartes would not have known the concept of a universal machine which
was discovered by Alan Turing, a machine so flexible that it can mimic any
other machine. l8
Descartes proposed that the immortal soul that he imagined interacted
with the human body through the pineal gland. This was not a good idea
as examinations of the pineal gland showed. In general, the view that
the stuff of mind is distinct from matter is known as substance dualism.
Descartes subscribed to this view; he thought that mind did not extend in
space like matter. There are two other kinds of ontological dualism. P r o p
erty dualism suggests that mental properties are fundamentally different
than material properties. The weakest statement of dualism is predicate
dualism 19, which states that mental (or psychological) predicates cannot
be reduced to material predicates, although supervenience holds.
A primary non-reductionist argument against physicalism is the knowl-
edge argument. 2o Imagine a neuroscientist who has been born and raised
in a black and white room. The neuroscientist knows every physical fact
about the brain. However, she has never seen red; she does not know what
red is like. The argument then quickly concludes that, the knowledge of
red cannot be reduced to physical facts about the brain, so it must be
extra-physical.

3.2. Arguments from ignorance

An extreme non-reductionist attitude may result in dualism. Both the non-
reductionist predicate dualism and knowledge argument views purport to
show that ordinary physicalism is insufficient to explain the mind. The
proponents of these arguments seem to arrive at their conclusions through
reinforcing our ignorance in cognitive science. 21
Predicate dualism relies on the fact that there are still many things
unknown about the physical processes in the brain. The claim is made
that there are no strict bridge laws connecting mental events and physical
events. If there were such laws, it would be possible to fully reduce psy-
chological states to physical descriptions. Assume that this is not possible.
Then, clearly, there is something more to psychological states than can be
explained by a physicist. It is also quite difficult to see how the part of
the psychological state that cannot be reduced to a physical description
could legitimately exist as part of our causal world. 22 While it is possible
to defend that a sufficient form of supervenience would hold in this case, l9
this view has the flavor of an idea that depends on our lack of progress
292

in neuroscience. We do not have a comprehensive neuroscientific theory

of how consciousness comes about in the brain. This lack of knowledge
is termed the “explanatory gap” and much has been said of it. 23 Thus,
the predicate dualist has a lot of leeway to persuade the readers about the
non-existence of bridge laws, since neuroscientists have not shown them yet.
However, this skepticism does not seem too different from invoking super-
natural forces in any phenomenon that science cannot explain currently. It
also seems to contradict the expected uniformity of natural law. In partic-
ular, why should we think that such bridge laws seem present everywhere
but the brain?
In the case of the knowledge argument, we are forced to believe that
the ultimate neuroscientist can not know what “red” is. However, even
today’s neuroscientists know a great deal about the human vision system.
Furthermore, if experience of red is a particular physical event as is usually
thought by neuroscientists, then it is acceptable that reading a book might
not result in the same brain activity. Given appropriate apparatus, how-
ever, the neuroscientist might be able to invoke any color sensation. Today,
artificial vision is a reality. As science advances, arguments like the knowl-
edge argument lose their ontological appeal, although they may be useful
in illustrating the difference between subjective experience and objective
knowledge. 24

4. Mild versus extreme non-reductionism

The complexity view encourages us to leave behind certain tendencies of
classical science. One such tendency may be called extreme reductionism.
We often see observations of complex systems such as human societies being
fitted to linear or quadratic equations, disregarding complex phenomena
like phase transitions. An extreme inductive bias towards simplicity is not
uncommon among scientists. Classically, everything must have a clear-cut,
simple and perfect explanation, such as the law of motion F = ma. Today,
we see that even in physics, the facts are not so simple. The world is
imperfect, complex, dynamic and not easily predictable.
However, the alternative to extreme reductionism must not be extreme
non-reductionism, which may bring about dualism via either of the non-
reductionist arguments in Section 3.1. In fact, dualism is one of the classical
tendencies that we must abandon. There are still scientists who believe in
an inexplicable soul. On the other hand, many other scientists have broken
this psychological barrier and see themselves as part of an evolving, complex
 293

nature. Nevertheless, we still need to clarify in what sense our minds are
irreducible.

4.1. Degrees of irreducibility

In Section 2, I have reviewed the information theoretic definition of com-
plexity. Algorithmic complexity of a system is a matter of degree. However,
the system is fully reducible to small physical interactions. This is the view
of mild non-reductionism. Indeed, viewing a complex system as consisting
of local interactions was a notable achievement of complexity sciences. In
the sense of supervenience physicalism, the properties of such a complex
system supervene on the local interactions, and of course the structure of
these interactions and the components as discussed in Section 2.3, all of
which are physical.
An advantage of mild non-reductionism is that it is fully compatible
with the mechanical view of nature. Algorithmic complexity clarifies our
intuitions about reducibility. Ordinarily, we view reducibility as an all or
none issue. However, reducibility does not make sense without certain im-
portant variables. Reduction to what, by what amount, and in many cases,
by which method? A physicalist would think that everything, including
mental states, are fully reducible to physical states. However, in the light
of Section 2.2 the physicalist would also agree that we can reduce only so
much. If we would like to store all the information about complex entities
such as one's psychology, or a living cell, then we cannot hope to compress
all this information beyond an optimally small volume of physical space.

4.2. Consequences of irreducibility

If we were to take away only a single piece of information from this optimal
description, the reproduced system may degrade in an important way or
stop functioning altogether, because there is no fault-tolerance in an optimal
description. The minimal program for anything is irreducible. When we
look at it, it is a random arrangement of symbols. Yet, it holds crucial
information. Note that, if something is random, there can be no more
analysis of it. Thus, all regularities in the data have been found in a
minimal description. If of course we knew all the regularities in the data,
this also means, intuitively, that we can realize the best prediction. The
analysis has occurred to the maximum limit in the process of compression
if we could have achieved maximal compression.
Unfortunately, in the real-world, we can never guarantee that we have
294

found a minimal program, or we can never know how close we are to a min-
imal program. Both of these have turned out to be uncomputable, and we
finite beings have to be satisfied with approximations. Beyond the uncom-
putability of algorithmic complexity there are incompleteness results that
place serious limits to analysis which have been rigorously investigated by
Chaitin. However, the incompleteness results in mathematics do not show
that it is impossible for scientific inquiry to be successful in general. Rather,
they deal in what a machine can/cannot infer from given axioms. On the
other hand, it has been possible for scientists to discover new postulates
such as that the speed of light is constant. As a consequence, we can never
be sure whether we have the most elegant theory in any domain. Intuitively,
we can imagine that the more the complexity, the more effort it will take
to compresslanalyze the data. Then, we need to diminish complexity.

4.3. The need for abstract talk

We can not demand that all sciences are reduced to physics before we
begin research. The possibility of reduction to physics is not an issue, the
feasibility is. We have to talk at a higher level of abstraction that makes
the problem smaller, because we often cannot face it in all its glory.
Let us turn our attention to the problem of piecewise analysis and anal-
ysis at different levels of description. Often, all the details about a system
are not known, or are irrelevant to our purpose. Then, we must have a way
of dealing with the relevant aspects of the system. For instance, one might
investigate the data transfer rate of the brain, if we are so certain that it is
a computer. We then need a reliable way to choose a level of abstraction
in which it becomes possible to talk meaningfully about the proposition.
That is to say, it should be neither a lower nor higher level of abstraction
than necessary. In the case of the data transfer question, we might like to
retain only the information about data transmission capabilities of individ-
ual links and the network topology. We do not need to concern ourselves
with philosophical puzzles to work at this level. In general, we would only
like to deal with information that allows us to predict a variable of our
choice.
There is another reason for abstract talk. In a mechanical universe, we
may expect the probability of a rule to be inversely related to its complexity.
That is to say, we do not expect arbitrarily complex rules to pop up. In
theory some explanations have been given as to why Occam’s razor holds; l5
our natural way of beginning with simple theories and progressing towards
 295

more complex theories seems to have a sound mathematical basis. This

would also explain why random events are not common. If randomness were
the common case, then our brains could make little use of all the information
about the past they are storing. Pure randomness means precisely that one
cannot use previous observations to predict the future. However, this is
not the case, we routinely predict and control events around us. This can
happen only because their algorithmic complexities are small. In other
words, randomness is bounded. Mountains do not appear at once, and
weather does not reverse every second. While in dynamical systems, a
view is that butterfly effects do occur, in the cosmic scale everything is
pretty much uniform. It is a fortunate fact that this predictability also
occurs to a healthy extent at the level of human beings. That is, mild
non-reductionism also leads to mild indeterminacy. It is not productive to
imagine that the macro-scale world is as unpredictable as the most random
quantum events. For sufficient prediction, a simple abstract theory might
be found by induction.
Another reason for abstract talk comes from the question: reduction
to what? When we say reduction, we do not mean reduction to an alien
language that we do not understand. Instead, we mean a satisfactory ex-
planation in familiar terms. That is to say, if we know x about the world,
we would like the new explanation y to be relative to this knowledge, i.e.,
that it is a short explanation in terms of x. We demand this, because oth-
erwise we cannot integrate y into our body of knowledge. This constraint
is captured well by the notion of conditional algorithmic complexity. It is
possible that an alien creature may have developed a quite different body
of knowledge, and we will have to spend a lot of effort to decypher its mus-
ings, i.e., the meaning would have large conditional complexity relative to
our culture.
Another relevant idea in algorithmic complexity is the notion of algorith-
mic independence (lack of mutual information). If through observations,
a brain can detect “islands” of information, then it can exploit this p r o p
erty. l6 Such a capacity for abstraction from sensory information seems
to be present in our brains, however we can make few speculations on its
properties at the present.

4.4. Algorithmic complexity and emergence

We are now in a better position to review the idea that animate-ness of
a cell is more than the sum of its inanimate parts. Or similarly, that a
296

brain is more than the sum of its neurons, I have already argued that
reduction to a physical description is necessary for the scientifically minded
complexity researcher who rejects ontological dualism. First, let us have a
look at the meaning of such predicates like animate or conscious that denote
complex entities. These have been most likely invented for the purpose of
avoiding excessive expcplanations, i.e., to be able to talk at a correct level of
abstraction. When we use these terms, we do not feel the need to explain
them further, however we can all recognize their presence through their
outward appearances. Furthermore, in many common sense problems, we
need no more information about such words except for the basic function of
distinguishing a dead animal from a living animal, etc. As countless works
in philosophy show, it takes a great effort to do otherwise and try to give
a complete description.
On the other hand, the animate-ness of a cell supervenes on continuing
success of a number of complex subsystems that are tightly interconnected.
Should one of these crucial systems fail, the cell is no more animate. While
like any other machine, the description of the cell is compositional, the
topology and nature of interconnections are non-trivial.
The outward appearances of complex phenomena are usually not satis-
factory for a scientist. Complexity community was not satisfied with the
definition of “animate” that it meant being able to reproduce or having
genetic material. They rightly sought generic properties of the systems
that we call animate. For instance, that they are stable, yet have dynamics
at the edge of chaos, that they are made up of parts that are being re-
placed over time. Attempts have been made to characterize not only earth
biology, but any animate matter, and further efforts have been made to
characterize evolution. For instance, Chaitin has given a rough mathemat-
ical definition of life 25 where he suggests that the interacting parts of such
a system (similar to components of a system in Section 2.3) are self-similar,
i.e., they have high mutual algorithmic complexity. He also suggests the
famous halting probability R as an abstract model of evolution; the approx-
imation procedure to calculate it is presented as a mathematical model of
entire evolution. On the other hand, researchers have already successfully
reconstructed evolutionary trees from gene sequences using an information
theoretic method. 26
The very existence of these generic explanations in complexity sciences
suggests that the concepts of animate-ness and evolution may be given ab-
stract descriptions that are independent of particular instances. It may
be possible to extract the essential commonality in all living systems into
 297

a clear description by choosing an affordable level of explanation. These

descriptions may turn out to be much simpler than the dynamics of a cell
which is a complex machine. The architecture of the cell may be impos-
sible to infer from the history of evolution prior to the cell, that is to say
it contains novel algorithmic information relative to that history. Given
the “inanimate” components of a cell, we can construct a large number of
systems, however, only a few of them will result in proper function. The
algorithmic complexity of the fit organization relative to the set of compo-
nents may be interpreted as corresponding to new “valuable” or “emergent”
bits invented by evolution.
The information theoretic view of emergence is similar to Bedau’s notion
of “weak emergence” as he distinguishes between nominal, weak, and strong
emergence. 27 However, in the ongoing discussion, emergence is independent
from the distinction between micro and macro levels.

5. Conclusion
The physical versions of algorithmic complexity help us consider it in a
purely physical way, and free of problems in the abstract version. I have
shown that algorithmic complexity corresponds to our intuition about com-
plexity, and further clarifies it with theorems. In Section 3 some dangers
of extreme non-reductionism have been shown by examining the knowledge
argument and predicate dualism which rest on epistemological and abso-
lute irreducibility. These arguments have been criticized as arguments from
ignorance. With the same attitude, we would have hardly advanced science
to this point. In Section 4,I have proposed using precise and quantitative
conceptions of complexity as they clarify our intuitions. Irreducibility of a
system comes in degrees, and it may be understood physically. Random-
ness means the absence of further reducibility, and we cannot reason about
random events. The reason for abstract talk is examined as a matter of pro-
cessing feasibility, purpose, inductive reasoning and opportunity. Finally,
instead of stating emergency of a property X as “ X is more than sum of its
parts”, I suggest that we can use the precise language of information theory.
In particular, “emergent property” can be given a satisfactory meaning in
AIT, corresponding to algorithmic information in a system that does not
reside in its components or prior history, thus is impossible to predict using
the information therein.
Mild non-reductionism cannot be the final word on this controversial
subject at least for the reason that algorithmic complexity has obvious
298

shortcomings. While irreducibility is so important, our intuitions may not

suffice. Algorithmic complexity helps us sharpen our methodology with
mathematical reasoning, and sheds new light on concepts such as irre-
ducibility, limits to analysis, predictability and emergence.

Acknowledgments
I a m grateful to Cristian S. Calude for technical corrections and suggesting
references. Thanks t o anonymous reviewers, Peter Jedlicka, Michael Olea,
Gregory Chaitin, and William Modlin for their comments which helped
improve this paper.

References
1. Carlos Gershenson and Francis Heylighen. When can we call a system self-
organizing? In W Banzhaf, T. Christaller, P. Dittrich, J. T. Kim, and
J. Ziegler, editors, Advances i n Artaficial Life, 7th European Conference,
ECAL 2003 LNAI 2801, pages 606414. Springer-Verlag, 2003.
2. Carlos Gershenson and Francis Heylighen. How can we think the complex? In
Kurt Richardson, editor, Managing Organizational Compledty: Philosophy,
Theory and Application, chapter 3. Information Age Publishing, 2005.
3. Information and Randomness - A n Algorithmic Perspective. EATCS Mono-
graphs in Theoretical Computer Science 1. Springer-Verlag, 1994.
4. Gregory J. Chaitin. Algorithmic Information Theory. Cambridge University
Press, 1987.
5. Thomas M. Cover and Joy A. Thomas. Elements of Information Theory,
chapter 1. Wiley Series in Telecommunications. Wiley-Interscience, 1991.
6. W. H. Zurek. Algorithmic randomness, physical entropy, measurements, and
the demon of choice. In A. J. G. Hey, editor, Feynman and computation:
exploring the limits of computers. Perseus Books, 1998.
7. Charles H. Bennett. How to define complexity in physics, and why. In W. H.
Zurek, editor, Complexity, Entropy, and the Physics of Information, volume
VIII, pages 137-148, 1980.
8. A.A. Brudno. Entropy and the complexity of the trajectories of a dynamical
system. IPrans. Moscow Math. SOC.,44:127, 1983.
9. Fabio Cecconi, Massimo Falcioni, and Angelo Vulpiani. Complexity char-
acterization of dynamical systems through predictability. Acta Physica
Polonica, 34, 2003.
10. Jean-Christophe Dubacq, Bruno Durand, and Enrico Formenti. Kolmogorov
complexity and cellular automata classification. Theor. Comput. Sci., 259( 1-
2):271-285, 2001.
11. Charles H. Bennett. Logical depth and physical complexity. In Rolf Herkin,
editor, The Universal Turing Machine: A Half-Century Survey, pages 227-
257. Oxford University Press, Oxford, 1988.
 299

12. N. Tishby FC Pereira and W Bialek. The information bottleneck method.

In Proceedings of the 37th Annual Allerton Conference on Communication,
Control and Computing.
13. Marcus Hutter. Universal Artificial Intelligence: Sequential Decisions Based
on Algorithmic Probability. Springer-Verlag, 2005.
14. William Bialek. Thinking about the brain. In H Flyvbjerg, F. Jlicher, P. Or-
mos, and F. David, editors, Physics of Biomolecules and Cells: Les Houches
Session L X X V , pages 485-577. EDP Sciences, Les Ulis; Springer-Verlag,
Berlin, 2002.
15. Ray Solomonoff. The application of algorithmic probability t o problems in
artificial intelligence. In M. Kochen and H. M. Hastings, editors, Advances
in Cognitive Science, A A A S Selected Symposia Series, pages 210-227. 1988.
16. Gregory J . Chaitin. Two philosophical applications of algorithmic informa-
tion theory. In Proceedings DMTCS’03 Springer Lecture Notes in Computer
Science, volume 2731, pages 1-10, 2003.
17. William Bialek, Ilya Nemenman, and Naftali Tishby. Predictability, complex-
ity and learning. Neural Comp, 13:2409-2463, 2001.
18. Marvin Minsky. The Emotion Machine, chapter 8. Draft.
http://web.media.mit.edu/-minsky/EB/ebB.html.
19. Donald Davidson. Mental events. In L. Foster and J. W. Swanson, editors,
Experience and Theory. 1971.
20. Frank Jackson. Epiphenomena1 qualia. Philosophical Quarterly, 1982.
21. Paul S. Churchland. The hornswoggle problem. Journal of Consciousness
Studies, 1996.
22. Dan Hutto. Davidson’s identity crisis. Dialectica, 52(1), 1998.
23. Ned Block and Robert Stalnaker. Conceptual analysis, dualism, and the ex-
planatory gap. Philosophical Review, 108:l-46, 1999.
24. Tim Crane. Subjective facts. In H. Lillehammer and G. Rodriguez-Pereyra,
editors, Real Metaphysics, pages 68-83. Routledge, London, 2003.
25. Gregory J. Chaitin. Toward a mathematical definition of life. In R.D.Levine
and M. Tribus, editors, The Maximum Entropy Formalism. MIT Press, 1979.
26. Rudi Cilibrasi and Paul M. B. Vitnyi. Clustering by compression. ZEEE
Transactions on Information Theory, 51(4):1523-1545, 2005.
27. M. A. Bedau. Downward causation and autonomy in weak emergence. Prin-
cipia, 6:5-50, 2003.
 THE COMPLEXITY OF INFORMATION-PROCESSING
TASKS IN VISION

JOHN SYMONS
University of Texas, El Paso

The basic presupposition of cognitive science is that mental life is more

complicated (or complex) than it appears; there is more to memory, atten-
tion, perception and the like, than meets the inner eye. In most mainstream
cognitive science, the hidden workings of the mind are treated as informa-
tion processing systems. In defense of their approach, cognitive scientists
often point to the importance of articulating the complexity of the informa-
tion processing processes underlying cognition and perception. The use of
complexity in defense of mainstream cognitive science deserves detailed ex-
amination, because of its importance and because it is more philosophically
problematic than has been recognized to date.
This paper focuses primarily on the use of complexity in the science of
vision. The information-processing model gained supremacy, in large part,
because of its success in the science of vision. Visual perception was one of
the preeminent areas of investigation and served as the testing ground for
the most important conceptual presuppositions of the discipline.
By the mid 1980s, it was commonly assumed that a satisfying explana-
tion of vision should permit us to program a computer to see. From the
computationalist perspective, a satisfying explanation of some aspect of
mental life is one that allows us to understand the phenomenon in question
as a series of steps articulated in ways that can easily be presented to our
colleagues without the appearance of question-begging.
However, the attempt to articulate the mind’s informational complexity
raises a series of fundamental questions. In the cognitive science of vision,
we can ask, for instance:

(1) What is the relationship between the information-processing com-

plexity that we hypothesize and the actual embodied act of seeing?
(2) What predictive power do we gain via an articulation of the
information-processing complexity of the visual system?

300
 30 1

(3) How can we correctly capture the information processing complexity

of a particular system?
(4) Is it always necessary to provide an information-processing account
of a system in order to provide a proper explanation?

One can easily imagine explanatory contexts where the information-

processing complexity that cognitive scientists hope to understand is sim-
ply not relevant. After all, most scientific explanations do not take an
information-processing form and for the most part, it is not obvious why
one should opt for an explanation of this kind. There might be some justifi-
cation for an information-processing style analysis in engineering contexts,
but even here, it is not clear that engineers would really need to know the
information-processing properties of the solutions they develop. In addi-
tion to their emphasis on the importance of non-enthymatic, proof-like pro-
cesses in explanations, the computationalist commitment to an information-
processing accounts of psychological and perceptual processes results from
a representationalist conception of mind. As we shall see, the notion of
information-processing complexity serves as the intersection for a tangled
set of basic questions in the philosophy of cognitive science.
This paper examines some of the arguments in support of the infor-
mation processing approach to the complexity of visual perception. David
M a d s defense of the approach serves as the focus of the analysis. Marr’s
posthumously published book Vision: A Computational Investigation into
the Human Representation and Processing of Visual Information (1982)
presents the most prominent computational theory of visual perception.
Vision provides an overview of work conducted by Marr and his colleagues
at the Artificial Intelligence Laboratory at the Massachusetts Institute of
Technology from about 1973 to the late 1970s. Written under the pressure
of serious illness and completed after his death by friends and colleagues, it
is widely regarded as the most important book in cognitive science. Vision
outlines a set of methodological and theoretical principles that have helped
to consolidate the self-image of cognitive science as a discipline. Marr’s
computational theory of the visual system is seen as providing a model
for theoretically sophisticated investigation into the biological mechanisms
responsible for mental life. So, for example, Steven Kosslyn claims that
Marr’s view provided “. . . an enormous advance over much of the previ-
ous work in neuroscience, which tended to minimize theory, and over most
of the previous work in artificial intelligence, which tended not to rest on
careful analyses of a problem.” (Kosslyn 1994,35) For the purposes of this
302

paper, M a d s text provides is an influential example of the reasoning that

shaped the prevailing view of complexity in the visual system.

1. Recognizing Complexity
Accounts of the visual system from the computational perspective almost
invariably begin by contrasting their approach with earlier attempts to
provide “direct” or “na’ive realist” theories of perception. Central to this
contrast is the notion of complexity. Computationalists see their work as
uncovering the complexity of the information-processing task involved in
visual perception that previous theorists had neglected. When scientists
are introducing and defending their approaches to vision, they regularly
begin by taking a stand on the notion of complexity. So, for example,
Alan Yuille and Shimon Ullman begin their introduction to computational
theories of low-level vision as follows:
Vision appears to be an immediate, effortless event. To see the
surrounding environment, it seems we need only open our eyes and
look around. However this subjective feeling disguises the immense
sophistication of the human (or animal) visual system and the great
complexity of the information processing tasks it is able to perform
(Yuille and Ullman 1990, 5)
Computationalists understand their predecessors as having ignored the
complex process underlying visual experience. Nicholas Wade and Michael
Swanston also echo this emphasis on the underlying process when they write
that: “The world around us seems concrete and immediate, and our abil-
ity to perceive it is easily taken for granted. . . Clearly, there must be some
process that gives rise to visual experience.” (Wade and Swanston 1991, 1)
By contrast with their predecessors’ neglect, they contend that the compu-
tational approach to perception as offers the best route to recognizing and
untangling the information processing complexity of the visual system. In
his introduction to cognitive science, Paul Thagard explicitly connects our
recognition of complexity to the problem of crafting a computer program
that sees:
For people with normal vision, seeing things seems automatic and
easy. You look at a room and immediately pick out the furni-
ture and people in it. The complexity of vision becomes apparent,
however, when you try to get a computer to do the same thing.”
(Thagard, 1996, 96)
 303

Introductory discussions of the science of vision show a striking level of

uniformity in their theoretical presuppositions with respect to complexity.
They echo a long tradition in cognitive science and cognitive psychology
which takes a two pronged approach to defending its approach to vision
and to mental life more generally. First, cognitive scientists criticize what
they claims is the naivety of their predecessors, who fail to see the com-
plexity of the processes under consideration. Second, they argue that the
correct approach to depicting and understanding the complexity of visual
perception involves the use of an information-processing model.
As early as Ulric Neisser’s 1967 introduction to cognitive psychology we
read the following:

Even psychologists who ought to know better have acted as if they

believed [in naive realism] [. . . ] The information reaching the eye
is subjected to complex processes of analysis, extended in time.
Visual experience results from some of these processes. [. . .] Per-
ception is not a passive taking-in of stimuli, but an active process
of synthesizing or constructing a visual figure. Such a complex
constructive act must take a certain amount of time. (1967, 16)

Neisser and others argued that the time taken to complete a process is
an indication of the level of complexity involved. It is important to note
that the complexity which interests cognitive scientists is said to exist at the
level of information-processing, rather than at the biological or behavioral
levels. Given the emphasis on the informational level analysis, sources of
evidence at the biological or even at the behavioral level tend to be marginal
to their choice of explanation. Given the nature of our access to the puta-
tive underlying information processing, the evidential constraints governing
articulations of informational complexity become somewhat problematic.
The time that experimental subjects take to perform visually-based tasks
serves as the primary empirical basis for the construction of theories of
information-processing complexity. However, the problem of understand-
ing what the informational complexity of a process amounts to and how
it should be studied is underdetermined by timing evidence alone. Timing
measurements alone leave room for a variety of explanations that are com-
patible with the evidence. Consequently, other, non-empirical constraints
must play a role in one’s decision to favor one articulation of information
processing complexity over another. Philosophical presuppositions, play an
especially prominent role in this kind of decision, as we shall see.
304

2. How Marr saw vision

Marr describes vision as a computational process wherein the brain con-
structs a three dimensional representation of the distal scene based solely
on the illumination of the surface of the retina. According to Marr, the vi-
sual system consists of three (or four, if one counts the initial effect of light
on the retina) hierarchically organized stages. At each stage a new aspect
of the representation is constructed from information originally contained
in the retinal image. For Marr, the visual system functions somewhat like
an assembly line whose final product is a useful, finished model of the visual
scene suitable for the organism’s manipulation. Along the way, each stage
in the visual system is explained in terms of the transformation of one rep-
resentational structure into another according to relatively fixed rules. No
new material or content is added to the representations as they undergo the
sequence of computational transformations. Instead, increasingly specific
features are extracted from information derived from the original retinal
image, until the visual system generates a fully elaborated model of the
visual scene. Models of this kind represent a data-driven approach to per-
ception. Data-driven models of perception are generally contrasted with
theories that view belief and other higher-level phenomena as playing a
role in the construction of what it is that the organism perceives. A gestalt
theory of perception would be an example of a top-down as opposed to a
purely data-driven theory.
The specific claims involved in his description of how the visual system
actually accomplishes these stages have proven less important to cogni-
tive science than the methodological framework articulated in the opening
and closing chapters of Vision. Following Marr’s methodological example,
traditional cognitive science has assumed that the first step in the investiga-
tion of mental life is the construction of an information-processing model.
This construction involves determining a set of well-defined information-
processing problems that the researcher believes the visual system must
solve. Marr recommended that these problems should, at least initially, be
studied in-themselves, apart from consideration of the underlying neuro-
science.
Marr’s suggested methodological tripartition in Vision; his distinction
between computational, algorithmic and implementational levels of anal-
ysis in cognitive science is consonant with the philosophical functionalist
intuition that, when it comes to matters of the mind, neuroscience can be
safely ignored, since mental states can be approached at the computational
 305

(or functional) level.

The notion of information-processing complexity plays a central role in
the approach and as we shall see, the claim that these systems are complex
is motivated by the assumption that the visual system traffics in infor-
mation bearing representations. By insulating the information-processing
approach from the constraints of biological or behavioral evidence, cog-
nitive scientists were free to craft a system of rules and representations
that solved the problems of perception that they believed the visual system
evolved to solve. However, because they are crafted in response to com-
putational characterizations of the problems in question, such solutions do
not necessarily help us to understand the mechanisms in actual biological
vision systems.

3. Information processors vs. resonators

As mentioned above, the computational or information-processing view is
almost always introduced by contrast with secalled direct theories of per-
ception. Marr follows this practice when, early in Vision, he distinguishes
his approach from J. J. Gibson’s. Gibson, who is widely regarded as the
most important modern proponent of the direct theory of perception, denied
that perception is mediated by sense data or representations and sought
instead to provide an ecological theory of vision that accounted for the
dynamical and embodied characteristics of the senses.
Gibson recognized that visual perception is mediated by highly complex
biological processes. His denial of what Susan Hurley called the classical
sandwich - the distinction between perception, cognition and action- is gen-
erally overlooked in computationalist characterizations of so-called direct
theories of perception. (Hurley 1998) Rather than tackling his criticism
of the functional-level amount, critics assume that the ecological theorists
share their analysis of the complexity of vision at the functional level and
then that they simply fail to explain that complexity properly. This is what
motivates the computationalist emphasis on Gibson’s sketchy and highly
simplified accounts of the internal processes that subserve the senses. So,
for example, Marr takes Gibson to task for ‘seriously underestimat[ing] the
complexity of the information-processing problems involved in vision and
the consequent subtlety that is necessary in approaching them.’ (1982, 29)
M a d s criticism of Gibson focused specifically on the idea of the senses
as organizing behavior in relation to ecological invariants. According to
Marr, Gibson’s approach was the product of his failure to understand in-
306

formation processing (1982, 29) and he wrote:

Although one can criticize certain shortcomings in the quality of

Gibson’s analysis, its major and in my, fatal shortcoming lies at a
deeper level and results from a failure to realize two things. First
the detection of physical invariants, like surface images, is exactly
and precisely an information processing problem, in modern ter-
minology. And second, he vastly underrated the sheer difficulty of
such detection. (1982, 30)

Gibson would object to the idea that the brain does anything like com-
puting the solutions to problems. For example in a passage cited by Marr,
he writes “the function of the brain, when looped with its perceptual organs,
is not to decode signals, nor to interpret messages, nor to accept images, nor
to organize the sensory input or to process the data, in modern terminol-
ogy.” (as cited in Marr 1982, 29) Consequently Marr and Gibson differed
fundamentally on the nature of brain function. For Gibson, the problem
with an account like Marr’s is that it relies on what he called “the mud-
dle of representation” (1988, 279) and that it takes what Gibson called a
“molecular” rather than a “molar” approach to perceptual systems. (1966,
52). The purpose of Gibsonian explanations, by contrast, is to show how
“the receptive units combine and their inputs covary in systems and sub-
systems” (1966,52) Both Gibson and Marr are interested in understanding
the function of vision. However, their views of what constitutes an explana-
tion of this functional level are different. This functional level difference is
not debated, and the computationalists, who assume that cognitive science
is in the business of providing complex information-processing accounts,
see Gibson as simply failing to understand the problem. In Marr’s words:
“Although some aspects of his thinking were on the right lines, he did not
understand properly what information processing was, which led him to
seriously underestimate the complexity of the information-processing prob-
lems involved in vision and the consequent subtlety that is necessary in
approaching them.” (1982,29)
From Marr’s perspective, the important lacuna in Gibson’s approach,
was his failure to articulate the process involved in extracting informa-
tion about the environment from the flowing array of ambient energy. For
Marr, it is not enough to say, as Gibson does, that the organism simply
‘Lresonates”with certain invariant features of its visual environment. Marr
believed that he could explain, in information processing terms, how the
organism extracted these invariant features in the first place. In Gibson’s
 307

terms, Marr can be understood as attempting to explain how an organism

comes to LLresonate” with some aspects of the environment rather than oth-
ers. At its core, Marr’s project is an attempt to explain how we recognize
things. For Gibson, by contrast, recognition happens via a set of biolog-
ical and ecological mechanisms. These mechanisms can be described, but
beyond that no further explanation is really necessary. The purpose of the
functional level analysis according to Gibson is simply to show the interac-
tion and covariation of what he called the “receptive units”. Marr viewed
the appeal to mechanisms that resonate with certain features of their envi-
ronment as missing precisely the information-processingstrategies by which
we recognize, to use his example, certain cat-like shapes as cats.
Denying representationalism, as Gibson did, does not entail a denial of
the possibility of an explanation of recognition per se. Gibson’s criticism of
representationalism is unequivocal, and yet it leaves space for explanations
of how we might come to judge, for instance that some object that we
directly perceive is a cat. He writes for instance:

Direct perception is what one gets from seeing Niagara Falls, say,
as distinguished from seeing a picture of it. The latter kind of
perception is mediated So when I assert that perception of the
environment is direct, I mean that it is not mediated by retinal
pictures, neural pictures, or mental pictures. Direct perception
is the activity of getting information from the ambient array of
light. I call this a process of information pickup that involves the
exploratory activity of looking around, getting around, and looking
at things. This is quite different from the supposed activity of
getting information from the inputs of the optic nerves, whatever
they may prove to be. (1979, 147)

Gibson was also interested in providing an explanation for the capacity

to recognize distinct objects as instances of a particular kind. However,
Gibsonian accounts of how such recognition takes place, do not employ
the representationalist strategies used by the computationalists. Rather
than processing information, for Gibson, the organism is simply reacting to
patterns of information in its environment. A more appropriate target for
Marr’s frustrations and criticism would have been the handful of philosc-
phers who argue that it is simply impossible to explain how we recognize
cats as cats, tables as tables, etc. Marr’s desire to explain how we can, for
example, see a cat as a cat or a table as a table, provides the basic impetus
for his idea that we should treat vision as a set of information-processing
308

problems. Marr actually cites John Austin’s discussion of what it means to

say that we recognize or fail to recognize the real shape of an object. (1982,
29) Marr criticizes Austin’s claim that there is no such thing as the real
shape of a cat. Austin writes :

What is the real shape of . . . a cat? Does its real shape change
whenever it moves? If not, in what posture is its real shape on
display? Furthermore, is its real shape such as to be fairly smooth
outlines or must it be finely enough serrated to take account of each
hair? It is pretty obvious that there is no answer to these questions-
no rule according to which, no procedure by which, answers are to
be determined. (Austin 1966, 67 as cited in Marr 1982, 31)

Contrary to the skepticism that Austin promotes, Marr shows how each
of Austin’s questions can be given an answer, perhaps not the kind of answer
that would satisfy the philosophical skeptic, but certainly the kind that
would satisfy ordinary scientific curiosity. Marr sees himself as providing
precisely these answers in the fifth chapter of Vision where he provides an
account of how the visual system generates representations of shape. The
purpose of M a d s account in that chapter is to show how the visual system
might provide some content for a judgment that some represented shape
can be identified as an instance of some concept- - -how, for example,
some represented shape is the shape of a cat.
To conclude this section, we have seen that M a d s criticism of direct
theories of perception is misdirected. Gibson was not suggesting that Marr’s
explanatory goals could not be achieved. Philosophers like Austin and
perhaps occasionally Wittgenstein may have held such positions, however,
as we have seen, where Gibson and Marr really differed was with respect
to the representational approach to perception. The significant difference
centers on the notion that the complexity of the information-processing task
must be articulated in representational terms. Gibson certainly believed in
some sort of mediating process that supported perception. What Gibson
denied was the need for a system of representations to serve as mediators
between perception, cognition and action. By contrast, the notion that the
brain generates a series of representations is central to M a d s view.

4. Representation and Information Processing Complexity

The idea that brains traffic in representations was encouraged by re-

 309

newed interest in the processes underlying behavior in psychology. While

cognitive psychology as a discrete field of study only emerged in the early
1960’s, the idea that recourse to representations and psychological processes
is necessary to explain behavior was already in the air thanks to Tolman’s
cognitive mapping experiments in the 1930’s. Theories of brain function
that posited internal manipulation of symbolic structures seemed to solve
a number of important problems for experimental psychology. They pro-
vided a model of the kind of internal processing that Tolman’s cognitive
mapping experiments were thought to require and, through the physical
symbol system hypothesis, they offered a neat solution to the problem of
the relationship between psychological and physical properties. Represen-
tations stored in a physical form in a computer could be understood as
having both semantic and causal properties.
Relying on what had, by then, become a standard line of thought in
cognitive science; Marr claims that we must have representations in the
brain, because in order for an organism to know something about an ob-
ject, it must be able to represent the object to itself. (1982, 3) Marr offers
a clear definition of what he means by ‘representation’. He writes: ” A
representation is a formal system for making explicit certain entities or
types of information, together with a specification of how the system does
this.” (1982, 20). Marr’s line of thought with respect to representation
runs roughly as follows: in order to know we must have representations in
the brain, and these representations are (or are part of) a formal system
that makes certain kinds of information explicit along with a set of instruc-
tions or rules for interpreting these representations. In this paper, I will
not provide an analysis of Marr’s view of representation per se, instead I
will focus on his unique and extremely influential computational account
of the relationship between representations. This is where the notion of
information-processing complexity plays an important role.
According to Marr’s view, cognitive scientists should articulate
information-processing problems into their component sub-problems and
present an explanation that consists of a string of computable solutions
to each of these problems. However, one unresolved issue faced by these
explanations involves the criteria for judging whether they fit the mecha-
nism under consideration. For instance, in biological systems a relatively
simple neural mechanism such as the edge detecting cells in the visual
system may solve what computationalists treat as a complex information-
processing task. In such cases, the neural mechanism can be interpreted as
computing a range of alternative algorithms for the information-processing
310

task. The challenge for the computationalist is to decide which of the set
of possible solutions to chose.
In his defense of the dynamical systems approach to perception and
cognition, Tim Van Gelder made the valuable point that simple systems
can accomplish feats that can also be construed as incredibly complex
information-processing tasks. (Van Gelder 1995;1991, Van Gelder, T. J.,
& Port, R. 1995) His example of such a system is Watt’s governor. In the
late 1700’s James Watt invented a device for controlling the relationship be-
tween the speed of a steam engine and the pressure of the steam driving the
machine. This device itself is simple, however, given computationalist incli-
nations, we could interpret it as solving a complex information-processing
task. Careful analysis of the problem could, for example, lead us to write a
computer program attached to a system of sensors (input) and valves (out-
put) which could calculate the adjustments that are necessary to maintain
a constant relationship between the pressure, speed, load etc. While noth-
ing stops us from producing such a system, (let’s call it Turing’s governor)
Watt’s governor was a far simpler and more elegant solution.a
How much would an articulation of this task in terms of a computer
program - the Turing Governor - add to our understanding of what it is that
the Watt Governor accomplishes? Practically speaking, a project to develop
Turing’s Governor would almost certainly have been an impediment to the
creation of a technological solution. It is possible that such a solution may
have been of some theoretical interest, but the chances of such theoretical
advancement would hardly be sufficient to abandon Watt’s solution
Theorists of direct perception, like Gibson and others, saw the senses
along the lines that Watt took to the governor. The senses were to be
understood in terms of the dynamic interaction between the organism and
the environment. Sensory systems were collections of feedback loops that
tended towards a kind of equilibrium between organism and environment.
In the case of the visual system, these visuomotor loops obviously involved

aThe Watt governor is a spindle with two hinged arms. Connected t o the engine’s
flywheel, the spindle turns at a rate that varies according t o the demands on the engine
(speed and load). As the spindle turns, the hinged arms are driven towards a horizontal
position by centrifugal force. The arms are connected t o valves that control the release
of steam such that as the flywheel spins faster, the arms of the spindle move upwards,
slowing the release of steam and thereby slowing the engine. If the flywheel slows, one
gets the opposite effect, the arms descend and the valve opens again, thereby increasing
engine speed. This simple device thereby establishes equilibrium between a variety of
forces without any recourse t o the complex information-processing techniques that would
have been involved in our imaginary Turing governor.
 311

some kind of mediating biological process, but these processes were thought
to be more like Watt’s governor than my imaginary Turing Governor. For
direct theorists of perception, a detailed presentation of the interaction of
the organism and the environment was a sufficient explanation of the senses.
Computationalist criticisms of direct theories of perception attack this
basic idea that the senses can be understood solely in terms of the interac-
tion of their component parts. The study of cells, for example, was seen as
failing to explain how an information-processingtask, like recognizing a cat
as a cat, could be accomplished. To compensate for this apparent deficiency,
an additional level of explanation was introduced wherein the dynamic re-
lationship between organisms and their environment could be analyzed into
components that had no direct relationship to physical objects or to observ-
able behavior. This level, the level of information-processing complexity,
permits the development of theories of perception that incorporate the all-
important notion of representation.
This complexity bears no direct relationship to the philosophical prob-
lem of computational complexity and has relatively little to do with Shan-
non and Weaver’s formal definition of the notion of information. Instead,
the standard strategy in cognitive science is to recommend that perception
be broken down into a series of stages consisting of transformations over in-
formation bearing states or representations. As we have seen, articulations
of this complexity may often have no correlate in physical reality. Conse-
quently, it is often very difficult to know what it is that we are being told
to recognize as complex.

5. Conclusion
The problem with Marr’s argument for the importance of information-
processing complexity is its implicit assumption that we can understand
what the visual system does or what its function is, without knowing its
physiological or anatomical form. Marr’s basic assumption, understood in
its strongest form, detaches accounts of this complexity from biological or
other sources of evidence and precludes the possibility that any new dis-
coveries regarding the structure or physiology of the brain or mechanism in
question could change the way we understand its computational or func-
tional properties. This weakness is due to the lack of evidential constraints
on the characterization of information processing complexity.
Marr qua scientist was a not dogmatic adherent to the autonomy of
the functional level and would probably have been sufficiently skeptical in
312

scientific practice not to stick to the strongest version of his methodological

principle come what may. Nevertheless, Marr’s methodology does recom-
mend that scientific practices be governed by a principle that is equivalent
to the belief that a suitably attuned philosopher or psychologist can deter-
mine psychological functions without concern for messy details of biology
or ecology.
Historically, the bias in favor of the functional or computational level of
investigation over the evidence of neuroscience did not originate with Marr.
It dates back at least to the classic early works in cognitive psychology and
is clearly articulated in Neisser’s Cognitive Psychology. (1967) Again, the
dismissal of biological evidence rested, almost from the start on the analogy
between minds and computer programs, as Neisser writes:
The task of the psychologist trying to understand human cognition
is analogous to that of a man trying to understand how a computer
has been programmed. In particular, if the program seems to store
and reuse information, he would like to know by what “routines”
or LLprocedures”this is done. Given this purpose, he will not care
much whether his computer stores information in magnetic cores or
in thin films; he wants to understand the program not the “hard-
ware.” By the same token, it would not help the psychologist to
know that memory is carried by RNA as opposed to some other
medium. He wants to understand its utilization, not its incarna-
tion. (1967, 6)

This radical denial of biology was not tempered by developments in

neuroscience. For example, it is common to read more recent works in
cognitive science arguing that:

Information processes can and must be studied without reference

to the physics or biology of the system carrying them out.. . the
algorithm for multiplication can be studied and understood in in-
teresting ways without any knowledge of how it is implemented by
neurons in the human brain.. . In general an important clarity of
understanding is attained by separately considering an information
process as a formal symbol manipulation process and as a physical
or biological process. (Stilling et al. 1987 4-5)

Even in the congressionally declared decade of the brain one could still
read cognitive scientists arguing along similar lines. In his (1996) Paul
Thagard lists ‘Neurobiological Plausibility’ as the fourth of five criteria for
 313

evaluating theories in cognitive science. (1996, 13) In his judgment, only the
‘Practical Applicability’ of a theory to educational techniques and design
ranks as a less import3nt criterion.
The only real theoretical constraints on computational explanations are
the limits of computability. This is a virtue insofar as it seems to allow the
investigator to abstract from the mental life of particular cognizers to ”cog-
nition as such,” but it is a vice when it comes to offering the kinds of expla-
nations that most of us seek from the brain and behavioral sciences. Given
an appropriate interpretation, we can imagine an information-processing
account capturing the essence of the behavior that interests us, but we
can only make this claim once we imagine putting the kind of biological or
robotic mechanisms of affection and transduction in place that will allow
us to imagine our interpretation of the algorithm being enacted. The char-
acterization of those mechanisms was precisely the goal of direct theorists
of perception like Gibson.
While Marr’s work and influence continue to be celebrated in both philc-
sophical and scientific circles, the science of the visual system has begun
to follow a very different path from the one Marr’s theoretical framework
prescribed. For instance, it is now clear to most workers in the field that de-
taching work at the functional or computational level from considerations of
neural anatomy and physiology is a less fertile strategy than taking neural
architectures seriously as a source of insight and evidence. Even computa-
tionally inclined investigators of vision have parted ways with Marr. For
example, the work of computationalists like Steve Grossberg begins by pay-
ing attention to the patterns of connectivity and the laminar structure of
visual cortex. Rather than believing that computational and biological lev-
els can be held separate in any scientifically useful sense, Grossberg claims
to identify the neural architecture with the algorithm.
Marrls view was, in Patricia Churchland’s phrase ‘brain-shy’ whereas
much of what has happened since the early-80’s has been brain-friendly,
and highly successful. Revisiting Marr’s account of complexity is impor-
tant if we are to understand how the norms governing explanation in the
computational functionalist framework have fared in the practical business
of scientific investigation.

Acknowledgements
I am grateful to Carlos Gersherson for his patience and assistance and to
two anonymous referees for some very helpful comments.
314

References
1. Austin, J. 1967. How to do things with words. Cambridge: Harvard University
Press.
2. Gibson 1966 The Senses Considered as Perceptual Systems Boston:
Houghton Mifflin.
3. Gibson, J.J. 1979. The Ecological Approach to Visual Perception Systems.
Boston: Houghton Mifflin.
4. Hurley, S.-1998. Consciousness in Action. Cambridge: Harvard University
Press.
5. Kosslyn, S. 1994. Image and bmin: The resolution of the imagery debate.
Cambridge: MIT Press.
6. Marr, D. 1982. Vision: A Computational Investigation into the Human Rep-
resentation and Processing of Visual Information. San Francisco: W.H. F’ree-
man.
7. Neisser, U. (1967) Cognitive Psychology. Englewood Cliffs, N.J.:Prentice-
Hall.
8. Stillings, N.A. et al. (1987) Cognitive Science: A n Introduction. Cambridge,
Mass.: MIT Press.
9. Thagard, P. 1997. Mind: A n Introduction to Cognitive Science. Cambridge:
MIT Press.
10. Van Gelder, T . J. 1995 What might cognition be, if not computation? Journal
of Philosophy, 91, 345-381.
11. Van Gelder, T. J., & Port, R. 1995. It’s About Time: An Overview of the
Dynamical Approach t o Cognition. In R. Port & T. van Gelder ed., Mind as
Motion: Eqlomtions in the Dynamics of Cognition. Cambridge MA: MIT
Press.
12. Van Gelder, T . 1998. The Dynamical Hypothesis in Cognitive Science. Be-
havioral and Brain Sciences. 21, 1-14
13. Wade, N. and M. Swanston, Visual perception: an introduction. London:
Routledge
14. Yulle, A.L.& S. Ullman, (1990). ‘Computational theories of low-level vision’.
In: D. N. Osherson, Kosslyn, S. M. &. J.M Hollerbach (Eds), Visual Cogni-
tion and Action, Volume 2, pages 5-39, MIT Press.
 ON THE POSSIBLE COMPUTATIONAL POWER
OF THE HUMAN MIND

HECTOR ZENIL* & FRANCISCO HERNANDEZ-QUIROZ+

Math Department, Faculty of Science,
National University of Mexico (UNAM)

-Forthcoming in: Carlos Gershenson, Diederik Aerts, and Bruce Edmonds (eds.)
Philosophy a n d Complexity: Essays o n Epistemology, Evolution, a n d
Emergence, World Scientific, 2006-

The aim of this paper is t o address the question: Can an artificial neural network
(ANN) model be used as a possible characterization of the power of the human
mind? We will discuss what might be the relationship between such a model and its
natural counterpart. A possible characterization of the different power capabilities
of the mind is suggested in terms of the information contained (in its computational
complexity) or achievable by it. Such characterization takes advantage of recent
results based on natural neural networks (NNN) and the computational power
of arbitrary artificial neural networks (ANN). The possible acceptance of neural
networks as the model of the human mind’s operation makes the aforementioned
quite relevant.

1. Introduction
Much interest has been focused on the comparison between the brain and
computers. A variety of obvious analogies exist. Based on several thoughts,
some authors from very diverse research areas (Philosophy, Physics, Com-
puter Science, etc.) claim that the human mind could be more powerful
than Turing Nevertheless, they do not agree on what
these “super-Turing” capabilities mean. Consequently, there is no univer-
sally accepted characterization of this extra power and how it could be
related to the human mind, even though there is strong defense of these
authors’ theories based on whether or not humans are “super-minds” ca-
pable of processing information not Turing-computable.

*zenil@ciencias.unam.mx
t fhq@hp.fciencias.unam.mx

315
316

Nowadays, it is accepted that the nervous system, including the spinal

cord and the neocortex, control behavior. In a simplified picture, the ner-
vous system could be viewed as a device which receives input from various
senses and produces output in the form of some action. From a computa-
tional point of view, the word “produces” could be harmlessly substituted
by “computes”, in the sense that the mind computes in the way a Turing
machine does. One of the most interesting and challenging tasks in sci-
ence is to assume this as a possibility in order to understand how these
computations are performed. Thus, it is not surprising that the field of
Neuroscience attracts many researchers not only from the fields of Biology
and Computer Science, but also from the area of Physics.
Contributions to the mind’s neural computation require a strong effort
towards interdisciplinary collaboration. As Teuscher and S i ~ p e pointed
r~~
out, there are very few problems stemming from neural computation on
which a theoretical computer scientist can commence work without further
interaction or collaboration with neuroscientists.
Since the artificial model was inspired in the natural, in this paper we
will explore the bridge between them on the basis of recent work. Since arti-
ficial neural networks have been inspired from their biological counterparts,
it is natural to ask about the feedback from the artificial to the biological.
We attempt build such a bridge by taking the standard model of ANNs and
the way in which ARNN admits arbitrary weights, in order to character-
ize its computational power by using the complexity of the weights or the
timing of the pulses.
Most of what happens in our brain does so without our being aware of
it, so that many things that seem easy to us involve highly complex neu-
ral circuits and operations. Most practical applications of artificial neural
networks are based on a computational model involving the propagation of
continuous variables from one processing unit to the next.
Inspired by a simplistic vision of how messages are transferred between
neurons, computer scientists invented the artificial computational approach
to target a wide range of problems in many application areas. Biological
neurons are connected by synapses, which are the links that carry messages
between them. Using synapses neurons can carry pulses to activate each
other with different threshold values. Neurons are the computational part
of the network while links connect different neurons and enable messages to
flow among them. Each link is a connection with a weight that affects the
carried message in a certain way. In fact each link receives a value from an
input neuron, multiplies it by a given weight, and then passes it to another
 317

neuron. Between this process several sources of computational power can

be identified. This means that the source of the computational power of
an NN might have at least three possible sources: (a) one carried by the
operations in the neuron itself; (b) the message between neurons involving
pulses, action potentials (APs) and timing and (c) the complexity of the
neural network weights in terms of their capability to encode information
which implies (a) and (b).
As biologically-inspired devices Artificial Neural Networks (ANNs) have
been suggested as a computational theory of mental activity. An ANN is
mathematically defined as a weighted directed graph in which each vertex
represent a neuron and each arrow a connection.
Definition 1 . A n Analog Recurrent Neural Network or ARNN" is
a finite network of nodes (called neurons) and connections, wherein the
synaptic weight associated with each connection (determining the coupling
strength of the two end-nodes) is a real (analog) value.
An ARNN is basically an artificial neural network allowing loops (re-
current) and irrational (even non-Turing computable) numbers as weights
(analog).
We consider the question about the possible computational power of
the human mind in this paper as a problem of computability in general and
tractability or practical computability out of its scope then making only
emphasis on the computability power defined by the Turing degrees and
beyond.
The key arguments which form the bases of our position are:
(1) ARNNs are a plausible model of the human brain because they
define a wide range of different levels of computational power.
(2) At least some characteristics of the human mind arise from the
human brain.
(3) ARNNs are therefore a plausible model of the human mind.
(4) It has been argued that the human mind could be depicted or

a The A R N N dynamics i s given by a mapping:

F : W ~ X ( 0 , l ) M+ R N
an which each neuron xi in t i m e ( t + l ) behaves as described by the following non-linear
equation:
M
aijzj(t) + b i j u j ( t ) + ci (2)
j=1 j=1
318

fully simulated by Turing machines, and others, as it has a “super-

Turing” computational power.
(5) Therefore, prima facie, it seems reasonable to explore claims on
the human mind’s supposed Turing or “super-Turing” capacities
through an analysis of ARNNs in relation to recent work on the
brain’s own neural network.
( 6 ) That an analysis of ARNNs supposed Turing and “super-Turing”
capabilities (and other more well established findings) gives rise to
a taxonomy of computational capabilities.
(7) The taxonomy of computational capabilities gives rise to prima facie
reasonable hypotheses concerning the human mind. The contribu-
tion of this paper to our knowledge is to build up this argument and
generate hypotheses binding artificial models to the natural.

With reference to argument number 1, the ARNN model even allows to

consider a situation in which the weights and the ARNN become equiva-
lent in power to automaton of lower power (including the Turing machine).
However, the ARNN (and a wider generalization proposed in this paper,
to be fully explored in another, yet to be published) allows us to consider
simultaneously, all possible scenarios. Claim number 2 is widely accepted
in different degrees of engagement. It is the case of the mind/body problem
and our claim is based on the consideration of the mind as the brain op-
eration. Hence, no distinction is made between brain and mind. However,
it is our assumption that the mind does not subtract any computational
power and fully inherits the whole computational power of the brain. If
mind adds some computational power to the overall system our arguments
still apply as lower bound. By contrast, if the mind subtract computational
power, our claims should be reformulated in terms of it and our arguments
will lay as upper bound.
Our intent is to discuss the mathematical features which a model should
possess, if it is to aspire to certain computable power. Our choice of an
ARNN model rests on two considerations: (a) it has been proposed as a
model with (potentially) non computable capabilities (provided some re-
quirements are met; see later); (b) as a neural network model, neurosci-
entists might find it easier to relate it to their own empirical work -most
of their research is based on (pseudo)analog values- and therefore they can
put forward and test concrete hypotheses to confirm or refute the model.
Additionally, the ARNN represents a refinement on what had been typically
regarded as the mathematical definition of an ANN, which did not make
 319

any distinction between the complexity of the weights in terms of the com-
putational power beyond the seminal work of Minsky. Nevertheless, neural
networks which run on digital computers operate on a Turing-computable
subset of irrational numbers, a strong restriction that determines a pri-
ori its computational power. Hence, some enthusiasm generated by im-
portant experimental and theoretical results cannot be extended easily to
applications because there is no straightforward way to make real numbers
available even assuming its possible existence. Digital hardware implemen-
tation uses a finite number of bits for weight storage and rational restraint
values for firings rates, weights and operations and remain limited to a
computational power. Even analog implementations, which are often cited
for their ability to implement real numbers easily (such as analog quanti-
ties), are limited in their precision by issues such as dynamic range, noise,
VLSIb area and power dissipation problems. Thus, in theory and in prac-
tice, most implementations use rational numbers, or at most, a subset of
irrational numbers, the Turing-computable ones.
The classical approach of computability theory is to consider models op-
erating on finite strings of symbols from a finite alphabet. Such strings may
represent various discrete objects such as integers or algebraic expressions,
but cannot represent general real or complex numbers, even though most
mathematical models are based on real numbers. The Thing machineG7and
its generalization in the form of the Universal Turing machine (UTM) is the
accepted model for computation, while under the Church-Turing thesis, it
is considered the authoritative model for effective computation. However,
some researchers have proposed other models in which real numbers play
the main role in effective computations.
Machines with “super-Turing” capabilities were first introduced by Alan
TuringGg,which investigated mathematical systems in which an oracle
was available to compute a characteristic function of a (possibly) non-
computable set. The idea of oracles was to set up a scheme for investigating
relative computation. Oracles are Turing machines with an additional tape,
called the oracle tape, which contains the answer to some non-computable
characteristic functions. An oracle machine is an abstract machine used
to study decision problems. It can be visualized as a Turing machine with
a black box, called an oracle, which is able to determine certain decision

bVery Large Scale Integration are systems of transistor-based circuits into integrated
circuits on a single chip. For example, the microprocessor of a computer is a VLSI
device.
320

problems in a single step. However, Turing gave no indication on how such

an oracle might be implemented. The Turing machine can write an input
on its own tape, then tell the oracle to execute it. In a single step, the
oracle computes its function, erases its input, and writes its output to the
tape. It is possible to posit the existence of an oracle, which computes
a non-computable function, such as the answer to the halting problem or
some equivalent. Interestingly, the halting problem still applies to such
machines; that is, although they can determine whether particular Turing
machines will halt on particular inputs, they cannot determine whether ma-
chines with equivalent halting oracles will themselves halt. This fact creates
a hierarchy of machines according to their Turing degree, each one with a
more powerful halting oracle and an even more difficult halting problem.
With such a method, an infinite hierarchy of computational power can
easily be constructed by positing the existence of oracles that perform pro-
gressively more complex computations which cannot be performed by ma-
chines that incorporate oracles of lower power. Since a conventional Turing
machine cannot solve the halting problem, a Turing machine with a Halt-
ing Problem Oracle is evidently more powerful than a conventional Turing
machine because the oracle can answer the halting question. It is straight-
forward to define an unsolvable halting problem for the augmented machine
with the same method applied to simpler halting problems that lead to the
definition of a more capable oracle to solve that problem. This construction
can be continued indefinitely, yielding an infinite set of conceptual machines
that are progressively more powerful than a Turing machine. This build
the hierarchy of Turing degrees:
Definition 2. The Turing degree of a subset A of natural numbers is
the equivalence class consisting of all subsets of B equivalent to A under
Turing reducibility. The degree of a set A is denoted by deg(A). The least
element in the partial order is denoted b y 0 and is the degree of all recursive
sets (computable sets).
In other words, two sets of natural numbers have the same Turing degree
when the question of whether a natural number belongs to one can be
decided by a Turing machine having an oracle that can answer the question
of whether a number belongs to the other, and vice versa. So the Turing
degree measures precisely the computability or incomputability degree of a
subset X .
Definition 3. A Turing reduction from a A to B is a reduction which
easily computes A assuming B , i.e. A is computable by a Turing machine
with an oracle for B .
 321

Because a language can be always codified by a subset of natural num-

bers, in terms of computability theory, sets and languages are equivalent.
Definition 4 . Formally, a language L is recognized b y a Turing ma-
chine with an oracle A if f o r every string s the machine using A as a n oracle
halts o n input s if S E L . A language B is Turing-reducible to a language A
if a Turing machine equipped with an oracle f o r A can decide B .
Models of hyper-computation tend to be of two general types: One uses
oracles or oracles in disguise, and the other uses infinite computation in fi-
nite time. Whether or not such machines are physically constructible -most
experts believe they are not- studying them improves our understanding of
the recursion theory.
On the other hand neural networks have been studied as computational
devices. In 1956, Kleene showedz7 how to simulate finite automata using
McCulloch and Pitts neurons41 and proved that when the weights of the
networks are constrained to be integers, the languages accepted by them are
exactly the regular languages. On the other hand, M i n ~ k yshowed ~ ~ that
neural networks with boolean neurons can simulate any finite automaton.
More recently, Hava Siegelmann5’ presented a computation model based on
analog recurrent neural networks (ARNN). All this work establish a very
strong connection between artificial neural networks and automata theory.
Both automaton (including the Turing machine) and neural networks are
characterized by the languages they accept and hold the same hierarchy.
Siegelmann offered a proof5g in which the set of languages accepted by
networks with rational weights are exactly the recursively enumerable lan-
guages=. Siegelmann showed that ARNNs are strictly more powerful than
the Turing machine model in that they can perform computations provably
uncomputable by a universal Turing machine. Disregarding the fact that
it seems unrealistic for most experts that those ARNN could be someday
constructible it is not easy to discard if such devices are already present in
nature taking advantage on the physical properties of the universe or, at
least, the way in which they can perform computations over its physical en-
vironmentd. According to G. Kreisel himself it is an open question whether

‘Davislg rightly pointed out that even if a subset of non rational numbers is used,
namely the set of Turing-computable irrationals, the class of languages recognized by
neural networks remains the same, as Siegelmann’s proof on the power of networks with
rational weights readily extends to nets with computable irrational weights (as Turing
already did with his machines).
dAnother interesting question raises by its own right: if there exists a natural device
with such capabilities how might we be restricted to take advantage of the same physical
322

there is any ”natural phenomenon” leading to an uncomputable numberg.

These neural networks satisfy some classical constraints of computa-
tional theory: the input, output, and transitions, are discrete and finite.
But the overall system is not really finite since it uses arbitrary real num-
bers, and it is known that it is powerful enough to encode all non-recursively
enumerable languages. Consequently, a connection between the complexity
of the networks in terms of their information complexity and their compu-
tational power had been developed, spanning a hierarchy of computation
from the Turing model to a “super-Turing” mathematical model.

2. Modeling the Brain

A clear difference between the brain and the computer is that a digital
computer operates by performing sequential instructions from an input
program, whereas there is no clear analogy of such a performance in the
human brain. However, since any Turing machine working on several tapes
is finally equivalent to a single tape Turing machine, the difference may
be functional rather than fundamental. The brain as a fully parallel de-
vice, could be reduced to a sequential input under some very significant
restrictions (for example being discrete or finite). Nowadays, such ques-
tions are the subject of scientific and philosophical debate since we have
the computational resources to develop live experimentse. However, Digital

properties in order t o build an artificial equivalent device? Much of the defense of the
work mentioned above have precisely centered on questions such as if we are taking
advantage of the resources we have in nature.
eThere are at least three important projects currently running: A Linux cluster running
the MPI NeoCortical Simulator (NCS), capable of simulating networks of thousands
of spiking neurons and many millions of synapses, was launched by Phil Goodman at
the University of Nevada. Blue Brain, a 4,000 processor IBM BlueGene cluster, was
used t o simulate a brain in a project started in May 2005 in the Laboratory of Neural
Microcircuitry of the Brain Mind Institute at the EPFL in Lausanne, Switzerland, in
collaboration with lab director Henry Markram. It has as its initial goal, the simulation
of groups of neocortical columns which can be considered the smallest functional units of
the neocortex. Also running is the NCS, t o be combined with Michael Hines’ NEURON
software. The simulation will not consist of a mere artificial neural network, but will
involve much more biologically realistic models of neurons. Additionally, CCortex, a
project developed by a private company Artificial Development, planned t o be a complete
20-billion neuron simulation of the Human Cortex and peripheral systems, on a cluster
of 500 computers: the largest neural network created t o date. Different versions of the
simulation have been running since June 2003. CCortex aims t o mimic the structure of
the human brain with a layered distribution of neural nets and detailed interconnections,
and is planned t o closely emulate specialized regions of the Human Cortex, Corpus
Callosum, Anterior Commissure, Amygdala and Hippocampus.
 323

computers were not designed to be models of the brain even when they are
running neural networks to simulate its behavior within their own compu-
tational restrictions. Most fundamental questions are however related to
its computational power, in both senses: time/space complexity and degree
of solvability. Most computational brain models programmed to date are
in fact strictly speaking, less powerful than a UTM. Researchers such as
Stannett63 have speculated that “if biological systems really do implement
analog or quantum computation, or perhaps some mixture of the two, it is
highly likely that they are provably more powerful computationally, than
Turing machines. This statement implies that true human intelligence can-
not be implemented or supported by Turing machines, an opinion shared by
Roger P e n r ~ s e who
~ ~ , believes mechanical intelligence is impossible since
purely physical processes are non-computable. A position strongly criti-
cized by many researchers in the field since its authors first propagated the
idea.
However, assuming some kind of relation between the mind and the
brain’s physical actions, neural networks may be accepted as a model of the
human mind operation. Since such a mindlbrain relation is widely accepted
in different ways and levels, we concern ourselves with the computational
power of these devices and the features that such networks must possess.
We will address which from our point of view, goes to the crux of the
matter when the question of the computational power of the brain is raised,
that is, its solvability degree. This means that we do not concern ourselves
with what could be the recipe in which a simulation could run, since if we
restrict ourselves to the discussion of artificial neural networks running on
actual digital computers, we will be restricted to the lowest computational
degree of solvability. From this, it can be easily deduced that if there is
a fundamental difference between the architecture of the brain and digital
computers, then the efforts of the artificial neural networks to fully simulate
the brain either for the purpose of study or reproduction, are destined to
have fundamentally different degrees of power.
Based on certain references5 as well as our own research, we have iden-
tified at least five mathematical descriptions in which “super-Turing” ca-
pabilities have been formally captured: super- task^^^^* and accelerated
Turing machines, Weyl machines or Zeus machines7, Trial and Error
machines50, Non-Standard Quantum Computation, and the Analog Re-
current Neural Networks59. We have also identified other proposals con-
cerning Turing machines with some kind of interaction between them or the
e n ~ i r o n m e n t ’ ~those
~ ~ ~models
~ ~ ~ ,provide a basis for the following claims:
324

(1) Minds are not computers, because (most) thought processes are not.
(a) It can “violate” Godel’s theorem, therefore is not a com-
puting machine, a claim made most famously by Lucas2’ in 1961.
(b) Mind can “solve” the “Entscheidungsproblem” , there-
fore is not a computing machine.
(2) Minds are computing devices but not of the same power of Turing
machines (maybe Godel himselff).
(a) There are special operations that occur in the brain
which are not Turing computable, a claim made most famously by
Penr~se~~.
(b) The mind could be a machine but with access to a cer-
tain oracle (from an external source or from a previously coded
internal source).

All hyper-computational models presented to date are purely theoret-

ical, and we may well ask whether they can actually be implemented in
the sense that the universal Turing machine is implemented or pseudo-
implemented in a common digital computer. A personal hyper-computer
would be no more an implementation of a hyper-computer model, than a
personal computer is of a UTM and nobody has physically implemented
any kind of hyper-computer. Most models seems to take advantage of (a)
a type of oracle device or (b) an infinite number of steps in a finite amount
of time.
Jack Copeland has pointed out13 an interesting fact concerning the way
Turing machines work. He stated that Turing machines are closed systems
that do not accept input while operating, whereas the brain continually
receives input from the environment. Based on this observation, Copeland
has proposed the coupled Turing machine which is connected to the envi-
ronment via one or more input channels13. However, as Christof Teuscher
and Moshe Sipper have pointed any coupled machine with a finite

‘In his 1951 Gibbs lecture” Godel attempts to use incompleteness t o reason about
human intelligence. Godel uses the incompleteness theorem to arrive at the following
disjunction: (a) Either mathematics is incompleteable in this sense, that its evident ax-
i o m can never be comprised in a finite rule, that is t o say, the human mind (even within
the realm of pure mathematics) infinitely surpasses the powers of any finite machine, or
(b) or else there exist absolutely unsolvable diophantine problems (or absolute undecid-
able propositions) for which it cannot decide whether solutions exist. Godel finds (b)
not plausible, and thus he seems have believed that the human mind was not equivalent
t o a finite machine, i.e., its power exceeded that of any finite machine, the term used
originally for Turing machines
 325

input stream can be simulated by a UTM since the data can be written on
the machine’s tape before it begins operation. From dynamic systems we
often decide almost in an arbitrary way, when a system will be closed in
order t o handle it. However, the chain of such external systems, potentially
infinite (even just by loops) can create a non-linear system which could
truly be more complex and maybe more powerful. Some other claims and
critics have been made in this regard.
Some other authors claim that the “super-mentalistic” perspective is
not a scientific one, as it implies the possibility of assigning non-reducible
phenomena t o some sort of information processing. However, we believe
that this fact does not preclude a study on what formal properties can be
required from non-Turing models of the human mind. A clarification of
these properties would help us understand to what extent “super-Turing”
models of the mind can or cannot be considered in a scientific way.

3. A computational model for the human mind

3.1. Determining the power of a neural network by the
complexity of its weights
Neural Networks are able t o encode information by several ways. If brain’s
neural network computes equal or less than the Turing degree of Turing
machines without oracles, their weights should code only whole or ratio-
nal numbers. However, if brain’s NN is more powerful it is clear that it is
possible to use an oracle to build a neural network that simulates a Turing
machine with that oracle in order to recognize any arbitrary language5’.
This can be done by encoding the oracle into a real number r which the
neural network “consults” in order to know if a certain input (or a trans-
formation of it) does belong to the language encoded by r.
Because we are interested in the set of Turing degrees of the weights of
a neural network, and since not always Turing degrees are comparable, it
is necessarily t o use the notion of a maximal element.
Definition 4 . Let be a partially ordered set (A,<). Then an element
r c A is said to be maximalg if, for all aEA,r$a.
We are going to take as “information” to the repository of languages.
In Computer Science and Linguistics a language is a set of symbols from

gNote that the definition for a maximal element is true for any two elements of a partially
ordered set that are comparable. However, it may be the case that two elements of a
given partial ordering are not comparable as in the case of Turing degrees as Post proved
326

an alphabet. As it is well known, every string on an arbitrary alphabet

can be encoded into a string on the binary alphabet {O,l}. In the same
way, a real number with infinite expansion can also be represented by an
infinite binary string. Putting together these facts, a language can be easily
encoded into a real number r in the interval [0, 1) taking the n digit of the
expansion of r as the value of the characteristic function of the language
applied to the n string in {0,1}* ordered lexicographically. Additionally,
this encoding is unique. Then given a language L , rL is the real number
that encodes it.
Definition 5. A real number r L is computable i f and only i f it is the
limit of a n effectively converging computable sequence of rational numbers.
T h e Turing degree of a real number r L i s defined as the Turing degree of its
binary expansionh.
Example 1. This definition is quite natural and robust. For instance,
let P be the set of natural numbers defined by P={nEN(the digit n of the
binary expansion of (A-3)is 1). P is evidently computable. Let be R N a
neural net with weights A and A-n. Since A is Thing-computable and A-n
too then R N will compute only those languages coded by A and 8-71,
Example 2. However if h is used as a weight, where h is the Chaitin
constant 0 defined as the probability of halting of a Turing machine Mi for
an input xi R N is going to compute definable languages but non-Turing
computable. It is evident that R N computes h and those in its same
complexity degree. If the set of maximals of R N has only h then the
Turing degree of R N is going to be the Turing degree of h, that is 0'. Then
if the human mind is capable to hold x,h or any other value as a connection
between neurons it will determine its computational power by taking the
set of maximals of the weight's Turing degrees.

3.2. Extracting information from the weights

A more interesting question is how to verify what type of languages a
weight has coded in order to determine the computational power of a neu-
ral network. That means what language a real number encodes. Since the
function of the encoding is bijective it is just needed to analyze one by
one the digits of the binary expansion of the real number in order to re-
construct the encoded language. Of course, languages encoded in this way

hit can be seen as a subset of natural numbers (or in fact the whole set of natural numbers
encoded in a single real number concatening all). It is evident that not all real numbers
are computable (they are also identified as random in Chaitin theory).
 327

go well beyond recursively enumerable languages and most of them will go

beyond any Turing computable procedure. It is clear that if a weight really
encodes a non rational number it is going to be impossible to extract all
digits in a finite time and space for most of them. However, if all weights
in the brain neural network are finite it will represent an argument for a
computationalism position on the human mind (and depending upon the
accepted relation between the brain and mind). If a computable irrational
number is taken, for example, T , the language it encodes is clearly recur-
sively enumerable but infinite. However this procedure obscures the rules
(of a suitable Turing machine) that generate the language and it is very
difficult to see how such rules could be derived from the number itself and
then impossible to determine if the substring n ( ~for ) the first n digits of
K is really going to be K at the end or any other number sharing the first
n digits. Even further, it seems to be impossible to distinguish between a
computable or non-computable number by this way.
Relativization of arbitrary neural networks by using the computational
power of their weights by any procedure could allow us to classify them
into the well known traditional hierarchies. For this purpose, we will use
an oracle to build a neural network that simulates a Turing machine with
that oracle to recognize a language. Let M be the neural network we will
build. First, we encode the input w into a real number r , as we saw in the
last section, then we code the oracle given in the same way. Let o be the
real number that encodes the oracle. For each entry i coded in r , the oracle
coded in o is consulted, when the answer is yes, the digit i is 1, when it is
not, the answer is 0.
Then M has two parts, one a sub-network N will simulate a Turing
machine, and the other, named 0 will simule the oracle machine5g. Let
w ( i ) be the oracle answer for the input i, we denote by c the concatena-
tion c = w(O)v(l) ...w ( i ) ...w(n),and by d(c) a Cantor encoding5g of T . The
Cantor encoding can be avoided since we are not concerned in this paper
in timelspace complexity. According to Davis’’ it is enough a straight
encoding based on length and lexicographical order’
Then, the network M can be described59 as a composition of two sub-
networks: the first, N , for which T, is a weight, receives the input w, and

‘~-1, a-2, b-3, aa-4, ab-6, ba+6, bb-+7,aaa-8, aab-+9,aba-10, . . . For example
if L is defined by all the strings that begins with an “a” followed by an arbitrary number
of “b”sL=(a, ab, abb, abbb, ...}, then the set S will be (2, 5, 11, 23, ...} and the
real number encoding L will be L,=0.0100100000100000000000100... (in fact it is a non
rational number in this case).
328

after a fixed computation N submit yw t o 0, where y comes from w after

the computation of the Turing machine simulation, the sub-network which
performs the oracle machine and has d ( c ) = o as weight. The output of the
sub-network 0 will depend on a binary activation functionJ .
The input arrives on two binary input lines. The first is a data line,
which is used to carry a binary input signal, when no signal is present, the
output is zero. The second is a validation line, which indicates when the
data line is active. It takes the value one while the input is present and
zero when not. There are also two output lines, that take the same roles.
These conventions allow using all external signals to be binary. This gives
rise t o a taxonomy of computational capacities summarized in the table of
the following section.
If any arbitrary neural network N is equivalent to a network N' with
two sub-networks M and 0 that simulate the Turing machine and the
oracle component respectively, then the computational power of the whole
network N will be determined by the sub-network 0. Therefore, it makes
sense to classify a neural network by means of the Turing degree of its oracle
component. The Turing degree of an oracle in a neural network is the set of
maximals of the Turing degrees of the weights of the network. The Turing
degree of each weight depends on its encoding capacity.

3.3. Other sources of computational power: spikes and

operations
In recent years, data from neurobiological experiments has made it increas-
ingly clear that biological neural networks, which communicate through
pulses, use the timing of these pulses to transmit information and to per-
form computation. This realization has stimulated a significant growth of
research activity in the area of Spiking (or Pulsed) neural networks (SNNs
or PNNs), theoretical analysis, as well as the development of new ANN
paradigm^^^,^^. From the perspective of our main concern in this paper,
an important question is: What type of information in terms of complex-
ity, might be carried by an action potential? Each action potential could
represent a single bit of information, similar to a serial digital communica-
tion channel without error checking. However, the information per action
potential may differ according to the function of the neural network. In

jA sigmoidal type function called the signal function defined as: signal(z) = 0 if xS0
and 1 in other case.
 329

traditional neural network models the timing of computation steps is usu-

ally “trivialized” . Clearly, precisely timing of spikes would allow neurons t o
communicate much more information than with essentially random spikes.
Otherwise, such temporally-coded information is lost as a source of possi-
ble computational power in those traditional models. It has been shown
that the use of pulse correlations in addition to pulse rates can significantly
increase the computational power of a neural network. These timing en-
codings finally contribute to the overall power of such networks.
Biologically-inspired computation paradigms take different levels of ab-
straction when modeling neural dynamics. The production of action po-
tentials or spikes has been ab- stracted away in many rate-based neuro-
dynamic models, but recently this feature has gained renewed interest. In
biological neural networks, information is transmitted by the conduction
of action potentials along axons, and information processing takes place at
the synapses, dendrites, and soma of neurons28. It has been shown that
firing correlations play a significant computational role in many biological
neural systems.
Models of biological neural networks in terms of dynamic systems have
been studied and formalized by H ~ d g k i n - H u x l e yand
~ ~ FitzHugh-Nagumo
equation^^^^^^ and others. Recently experimental evidence has accumu-
lated during the last few years, which indicates that many biological neural
systems use the timing of single action potentials (or “spikes”) to encode
information1~3~6~51~24~21~70~53,55~56@. Experiments have shown that in vitro
biological neurons fire with slightly varying delays in response to repetitions
of the same current injection’. Even when this behavior may be explained
by noisy versions of neural networks, the source for some kind of possible
codification in such processes remains. However, noise certainly affects the
computational power of networks of spiking neurons for analog i n p ~ t ~ l ? ~ ~ .
Nevertheless, empirical experiments in which complex fluctuations exist, as
seen for example in EEGk signals that are generally taken as noise, may be
indicative of complex dynamics for processing information similar t o that
of pulse propagation networks or spiking neurons, maybe continuously over
time.
A central theme in this paper is the coding capabilities of neural net-

kElectroencephalographyis the neurophysiologic measurement of the electrical activity

of the brain by recording from electrodes placed on the scalp or, in special cases, on the
cortex. The resulting traces are known as an electroencephalogram (EEG) and represent
so-called brain-waves.
330

works in which its computational power lies. Most accepted descriptions of

the human neural network involve finite and discrete quantities, but some
of them remain as continuous variables from one processing. Computer
scientists have been studying ANNs for many years. Although ANNs were
inspired by real biological networks, typical ANNs do not model a number
of aspects of biology that may prove to be important. Real neurons, as we
have seen for example, communicate by sending out little spikes of voltage
called action potentials (APs). ANNs, however, do not model the timing of
these individual APs. Instead, ANNs typically assume that APs are repet-
itive, and they model only the rate of that repetition. However some of the
computational power of a biological neural network could be derived from
the precise timing of the individual APs or other properties inherent to the
biological neural network that it is not fully right-modeled. Regular ANNs
could never model such a possibility defined as their currently are.
Additionally, all ANNs running over digital machines are incapable of
simulating any possible analogical signal or full irrational encodings if they
were present in human brain’s architecture. Methods for estimating the
computational power of neural circuits and their relation to artificial neu-
ral networks models have been established. Maass30,32733134 and Markram3’
have recently argued that “in contrast to Turing machines, generic compu-
tations by neural circuits are not digital, and are not carried out on static
inputs, but rather on functions of time”. These kinds of experiments could
ultimately provide a definitive answer to the critical issues which are the
concern of this paper.
These models are also interesting because even if a neural network is
constrained to simple weights, namely whole or rational numbers, it is pos-
sible to achieve an extra power from the timing process, encoding what
is not possible to encode in the weights. Since they are similar in power
to analog signal processing, they can be compared to the traditional hier-
archies we have already explored. Even when a mathematically rigorous
analysis of the computational power of networks of spiking neurons has so
far been missing, they are equivalent to the levels of such hierarchies, sim-
ply by replacing weights encodings with pulse timing encodings. In other
words, the computational power is transferred from weights to spikes, and
in the presence of a mixed model with both weights and spikes encod-
ings, its final computational power will be the power of the most powerful.
Both pulse frequencies and correlations which are computationally relevant
can be seen as operations involving potentially any possible real numbers
in principle equivalent to a weight encoding. Hence, the overall power of
 331

these PNNs or SNNs is determined by the maximal Turing degree of the

union of the Turing degrees of both weights and time encodings. It has also
been shown that SNNs are, with regard to the number of neurons that are
needed, computationally more powerful than other neural network models.
These models are obviously Turing reducible when variables (the input and
all internal operations) are restricted to Turing-computable values if such
model is close under its operations. However when arbitrary values are
allowed then they can compute possible non-Turing functions. M a a . 9 ~has ~~
shown that, with regard to temporal coding of real-valued variables any
continuous function can be approximated arbitrarily closely. In this case,
the model depends on the continuity of time.
On the other hand, since it is possible to compute fifrom simple opera-
tions between whole numbers (as the hypotenuse of a Pythagorean triangle
with unitary sides) and even to make proofs of irrationality without calcu-
lating their decimal expansion, it might be possible to prove if brain’s neu-
ral network can hold non-computable numbers if it is able to perform some
special operators where the discrete operations, such as primitive recursion
or bounded sums and products, are replaced by operations on continuous
functions, such as integration or linear i n t e g r a t i ~ n ~The
~ . sets of those
real functionss4 can be definable by a general-purpose analog computer”
or GPAC, which is a general computing model evolving in continuos time’.
Then even if weights and spikes are restricted to Turing-computable func-
tions, the brain neural network might be not closed if at least an operation
is not recursive possibly (see and 16) taking the involved values to non-
Turing degree at some point of the computation. Traditional functional
operations like Sequential Composition, Parallel Composition or definition
by Cases, Primitive Recursion or simply Recursion, Bounded Recursion,
Partial Projection, Cut-off Subtraction and the order functions like the
Minimization or Projection and Bounded minimization take computable
functions into computable functions. By contrast, It is well known that
limits take computable functions beyond the class defined by the power of
Turing machines. For a more detailed description of such operations see
a. The power of human brain operations remains unknown, but even if

162,16 The class of R-recursive functions is very large. It contains many traditionally
non-computable functions, such as the characteristic functions of sets of the arithmeti-
cal h i e r a r ~ h y ~ *Experimental
%~~. proofs beyond this level could be more difficult if not
impossible since the construction of a sequence of real numbers which can not be com-
putably diagonalized is used to prove that there are continuous functions without a
Turing degree.
332

non-traditional operations were possible both theoretical and empirical ev-

idence seem to be hard or impossible since the verification of non-Turing
operators require non-Turing inputs and outputs which seems to be undis-
tinguishable from those Turing-computable. Nevertheless neurological tests
could be designed in order to achieve some advance in this direction.
One could suppose that given the evolution of neural networks models,
several better approaches to the biological neural network model could be
expected. Other possible sources could be unrepresented in current artificial
models. Experimental and theoretical research in the field should continue
to relate it more to the biological model. ma as^^^ suggests SNNs as the
third generation of the artificial neural network model. Each one seems
to offer better approaches to the experimental evidence of the brain’s own
neural network.
Now, we can build a simplified hierarchy of computational power re-
garding the two possible sources explored in this paper (weights and spikes,
which are equivalent in terms that they can be replaced one for the other
preserving the whole complexity of the NN):

Neural Network Architec- Computational Power

ture
R Non-computable num-
bers
Q Rational numbers
Z Whole numbers
Turing machine with an oracle depend-
ing on the maximal Turing degree of of
the weights: w1,w2, . . .’ n E R and pulses
pl,p2, . . .’ rn E R, not always in the Arith-
metical Hierarchy
at most Turing machine
at most bounded automata

At the bottom level, we have neural networks computing Turing-

computable languages if the oracle is empty. On the other hand, weight’s
and spikes’ complexity could be determined -assuming that the neural net-
work is closed under isomorphism- then they can be decoded into an oracle
which will determine its computational power. If the neural network per-
form not closed operations under isomorphism, the neural network will
compute in a range determined by the bounds of such operations, possibly
determining an interval within this simplified hierarchy.
 333

4. Discussion and Conclusion

Even when the answer concerning the computational power of the human
mind should come from neurophysiological and interdisciplinary research,
including the correspondence between the natural and the artificial neural
network models, the question remains of mathematical and philosophical
interest as it is possible to explore all possible scenarios inside the com-
putability theory. We add that if we were able to determine some key
properties of the physical universe and the mind we could determine the
computational power of the mind, among them:

0 if our physical universe is discrete or continuous, and in which case

0 if it can hold non-computable numbers;
0 if it can hold non-computable operations;
0 if the brain inherits those properties;
0 if the mind can take advantage of them, which depends on the
precise relationship between the mind and the physical operation of
the brain.

As it can be seen, these are not small requirements. However once we

can ascertain some fundamental properties of the mind it could be possible
to know to which computational level it belongs by its maximal Turing
degree. The value of this paper lies in its effort to restore some of the
claims which have been made and proven for artificial neural networks to the
natural model and the problem of determining the mind’s computational
power. Then, it is not just about the way in which a neural network can
achieve one or more computational powers, but also which features a brain
should have to potentially take its power to one level or another: either
less, equal or more powerful than Turing machines.
Then, if every language with a given alphabet can be encoded and po-
tentially extracted into a real number (from weights or spikes), this real
number can be used as the main component of a neural network that simu-
lates an oracle Turing machine whose oracle is the characteristic function of
the language. Therefore by determining the neural network encoding capa-
bility and knowing the type of operations permitted in it, its computational
power could be determined.
We would like to conclude, by remarking that a full model of the mind
has a very difficult task ahead. We hope that this paper has succeeded in
shedding light on the current research and ways to build certain bridges
from the artificial to the natural model, as a possible way to eventually
334

determine the computational power of the human mind taking in consider-

ation all possible sources of power.

Acknowledgments
The authors would like to thank the editors and referees for very helpful
comments during the preparation of this paper.

References
1. M. Abeles, Corticonics. Neural circuits in the cerebral cortex, Cambridge, Eng-
land: Cambridge University Press, 1991.
2. A. Aertsen, M. Erb, and G. Palm, Dynamics of functional coupling in the
cerebral cortex: an attempt at a model-based interpretation, Physica D, 75,
103-128, 1994.
3. M. Arbib (Ed.) The Handbook of Brain Theory and Neural Networks, 2nd
Edition. Cambridge and MIT Press. 1072-1076, 2002.
4. P. Benacerraf, Tasks, super-tasks, and the modern elastics, J. Philos. 59 765-
784, 1962.
5. S. Bringsjord and M. Zenzen, Superminds, People harness hypercomputation
and more, Kluwer, 2003.
6. W. Bialek, , F. Rieke, R. Steveninck, de Ruyter van, and Warland, D. Reading
a neural code. Science, 252, 1854-1857.
7. G. Boolos, J. P. Burgess and R. C. Jeffrey, Computability and Logic, Cam-
bridge Universky Press, 2002.
8. Mark Burgin, Super-Recursive Algorithms (Monographs in Computer Science),
Springer, 2004. ’

9. Cristian S. Calude, Information and Randomness : A n Algorithmic Perspec-

tive (Texts in Theoretical Computer Science. An E A T C S Series), Springer,
2002.
10. Manuel Campagnolo, Cristopher Moore, and Jose Felix Costa. Iteration, in-
equalities, and differentiability in analog computers, Journal of Complexity,
16(4):642-660, 2000.
11. B. B. Jack Copeland, Hypercomputation, Minds and Machines 12: 461-502,
Kluwer Academic Publishers, 2002.
12. B. B. Jack Copeland, and D. Proudfoot. Alan Turing’s forgotten ideas in
computer science. Scientific American 280, 4, 77-81, 1999.
13. B. B. Jack Copeland and R. Sylvan, R. Beyond the universal Turing machine.
Australasian J. Philosophy 77, 1, 46-66, 1999.
14. J. FBlix Costa, O n Hava (Eve) T. Siegelmann’s book Neural Networks and
Analog Computation”, Universidad TBcnica de Lisboa-Birkhauser, 1999. 68-
83 University of Nevada, Reno.
15. J. Mycka and FBlix Costa, JosB. emphReal recursive functions and their hi-
erarchy, Journal of Complexity, 20(6):835-857, 2004.
16. J. Mycka and FBlix Costa, J o s B . What lies beyond the mountains, computa-
 335

tional systems beyond the Turing limit, European Association for Theoretical
Computer Science Bulletin, 85:181-189, 2005.
17. M. Davis, Computability and Unsolvability, Dover Publications, 1958.
18. M. Davis, The Undecidable, Dover Publications, 2004.
19. M. Davis, The Myth of Hypercomputation, Alan Turing: Life and Legacy of
a Great Thinker, Springer, 2004.
20. R. FitzHugh, Impulses and physiological states in models of nerve membrane,
Biophysical Journal, 1:445466, 1961.
21. W. Gerstner, R. Kempter, J. van Hemmen and H. Wagner, A neuronal learn-
i n g rule for sub-millisecond temporal coding, Nature, 383, 76-78, 1996.
22. K. Godel, Some Basic Theorems on the Foundations of Mathematics and
Their Implications, in Feferman, et al., 1995, 304-323, 1951.
23. A.L. Hodgkin, A.F. Huxley, A quantitative description of membrane current
and its application to conduction and excitation in nerve, J. Physiol. 117 500-
544, 1952.
24. J.J. Hopfield, Pattern recognition computation using action potential timing
f o r stimulus representation, Nature, 376, 33-36, 1995.
25. J.J. Hopfield, Neural networks and physical systems with emergent collective
computational abilities, Proc. Natl. Acad. Sci. USA 79 25542558, 1982.
26. K. Judd, K. Aihara, Pulse Propagation Networks: A Neural Network Model
That Uses Temporal Coding by Action Potentials, Neural Networks, Vol. 6,
pp. 203-215, 1993
27. S.C. Kleene, Representation of events in nerve nets and finite automata.
In C.E. Shannon and J. McCarthy, editors, Automata Studies, pages 3-42.
Princeton University Press, Princeton, NJ, 1956.
28. C. Koch, Biophysics of Computation: Information Processing in Single Neu-
rons, Oxford University Press, 1999.
29. JR. Lucas, Minds, Machines and Gdel. Philosophy, 36:112-127, 1961.
30. W. Maass, R. A. Legenstein, and N. Bertschinger. Methods for estimating the
computational power and generalization capability of neural microcircuits. In
L. K. Saul, Y. Weiss, and L. Bottou, editors, Advances in Neural Information
Processing Systems, volume 17, pages 865-872. MIT Press, 2005.
31. W. Maass, Lower bounds for the computational power of spiking neurons.
Neural Computation, 8:140, 1996.
32. W. Maass. O n the computational power of neural microcircuit models: Point-
ers to the literature. In Jos R. Dorronsoro, editor, Proc. of the International
Conference on Artificial Neural Networks - ICANN 2002, volume 2415 of Lec-
ture Notes in Computer Science, pages 254-256. Springer, 2002.
33. W. Maass. Neural computation: a research topic for theoretical computer
science? Some thoughts and pointers. In Rozenberg G., Salomaa A., and
Paun G., editors, Current Trends in Theoretical Computer Science, Entering
the 21th Century, pages 680-690. World Scientific Publishing, 2001.
34. W. Maass. Neural computation: a research topic for theoretical computer
science? Some thoughts and pointers. In Bulletin of the European Association
for Theoretical Computer Science (EATCS), volume 72, pages 149-158, 2000.
35. W. Maass and C. Bishop (Eds), Pulsed Neural Networks, MIT Press 1999.
336

36. W. Maass Networks of Spiking Neurons: The Third Generation of Neural

Network Models, Neural Networks, Vol. 10, No. 9, pp. 1659-1671, 1997.
37. T. Natschia, W. Maass, Dynamics of information and emergent computation
in generic neural microcircuit models, Neural Networks 18, 13011308, 2005.
38. W. Maass, T. Natschlager, and H. Markram, Real-time computing without
stable states: A new framework for neural computation based on perturbations,
In Neural Computation, volume 14, pages 25312560, 2002.
39. W. Maass, and P. Orponen, On the effect of analog noise on discrete-time
analog computations, Advances in Neural Information Processing 11 Systems
9, 1997, 218224; journal version: Neural Computation 10, 1998. detailed ver-
sion see http://www.math.jyu.fi/?orponen/papers/noisyac.ps .
40. J . Nagumo, S. Arimoto, and S. Yoshizawa, An active pulse transmission line
simulating nerve axon. Proceedings of the IRE, 50:20612070, 1962.
41. W.S. McCulloch and W. Pitts. A logical calculus of the ideas immanent in
nervous activity’, Bulletin of Mathematical Biophysics, 5:115-133, 1943.
42. C. Moore, Recursion theory on real-time language recognition by analog com-
puters. Theoretical Computer Science, 162:23-44, 1996.
43. Jerzy Mycka. mu-Recursion and infinite limits, Theoretical Computer Sci-
ence, 302:123-133, 2003.
44. M. Minsky, Computation: Finite and Infinite Machines, Prentice Hall, New
york, London, Toronto, 1967.
45. P. Odifreddi, Classical Recursion theory. Elsevier, 1989.
46. R. Penrose, The Emperor’s New Mind Concerning Computers, Minds, and
the Laws of Physics, Oxford: Oxford University Press, 1989.
47. R. Penrose, Shadows of the Mind: A Search for the Missing Science of Con-
sciousness, Oxford: Oxford University Press, 1994.
48. E. Post, “Recursively enumerable sets of positive integers and their decision
problems”, bulletin of the American Mathematical Society, 50:284-316.
49. H. Putnam, Renewing Philosophy, Cambridge, MA: Harvard University
Press, 1992.
50. H. Putnam, W a l and EWOT Predicates and the Solution of a Problem of
Mostowski, Journal of Symbolic Logic 30, pp. 4957, 1965.
51. F. Rieke, D. Warland, R.R. von Steveninck, and W. Bialek, Spikes: Exploring
the Neural Code, MIT Press, Cambridge, USA, 1997.
52. J.R. Searle, Minds, Brains, and Programs, Behavioral and Brain Sciences, 3,
417-457, 1980.
53. T.J. Sejnowski. Neural pulse coding. In W. Maass and C.M. Bishop, editors,
Pulsed Neural Networks, The MIT Press, 1999.
54. Daniel Silva Graa. Some recent developments on Shannon’s General Purpose
Analog Computer, Mathematical Logic Quarterly, 50(4-5):473-485, 2004.
55. W. Singer, Neuronal synchrony: A versatile code for the definition of rela-
tions, Neuron, 24, 49-65, 1999.
56. W. Softky, Simple codes versus eficient codes, Cum. Opin. Neurobiol., 5,
239-247, 1995.
57. C. Shannon, Mathematical theory of the differential analyser. J. Math. Phys.
MIT, 20:337-354, 1941.
 337

58. H.T. Siegelmann, Computation Beyond the T u r i n g Limit, Science, 268:545-

548, April 28 1995.
59. H.T. Siegelmann, Neural Networks and Analog Computation, beyond the Tur-
ing Limit, Birkhauser, 1999.
60. H.T. Siegelmann and E.D. Sontag, Analog computation via neural networks,
Theoretical Computer Science, 131:331-360, 1994.
61. H.T. Siegelmann and E.D. Sontag, O n computational power of neural net-
works, Journal of Computer and Systems Sciences, 50( 1):132.150, 1995.
62. Daniel Silva Graa and Jose Felix Costa, Analog computers and recursive
functions over the reals, Journal of Complexity, 19(5):644-664, 2003.
63. M. Stannett. X-machines and the halting problem: Building a super Turing
machine, Formal Aspects of Computing 2, 4, 331-341, 1990.
64. W. Rall, Theoretical significance of dendritic tree for input-output relation, in:
R.F. Reiss (Ed.), Neural Theory and Modeling, Standford University Press,
Stanford, pp. 73-97, 1964.
65. Christof Teuscher and Moshe Sipper, Communications of the ACM, Vol. 45,
No. 8, 2002.
66. S. Thorpe, F. Fize, and C. Marlot, Speed ofprocessing in the human visual
system, Nature, 381, 520-522, 1996.
67. A.M. Turing, O n computable numbers, with an application to the Entschei-
dungsproblem, Proc. London Math. SOC.42 (2) (1936) 230-265. A correction
in 43 (1937) 544-546 (reprinted in: M. Davis, (Ed.). The Undecidable: Basic
Papers on Undecidable Propositions. Unsolvable Problems and Computable
Functions, Raven, New York, 1965, pp. 115-154) Page numbers refer to the
1965 edition.
68. J. Thomson, 1954-55, Tasks and Super-Tasks, Analysis, XV, pp. 1-13;
reprinted in Salmon, 1970.
69. A.M. Turing, Systems of logic based o n ordinals, Proc. London Math. SOC.
45 (2) (1939) 161-228 (reprinted in: M. Davis (Ed.). The Undecidable: Basic
Papers on Undecidable Propositions. Unsolvable Problems and Computable
Functions, Raven, New York, pp. 154-222, 1995.
70. K. Wagner and T.M. Slagle. Optical competitive learning with V L S I liquid-
crystal winner-take-all modulators. Applied Optics, 32(8):14081435, 1993.
71. P. Wegner and D. Goldin, Interaction, Computability, and Church’s Thesis,
University of Massachusetts at Boston, 1999.
72. J. Wiedermann and J. van Leeuwen, Emergence of a Super-Turing Computa-
tional Potential in Artificial Living Systems, J. Kelemen and P. Sosik (Eds.):
ECAL 2001, LNAI 2159, pp. 5565, Springer-Verlag Berlin Heidelberg, 2001.
 HOW DOES COMPLEX MATHEMATICAL THEORY
ARISE? PHYLOGENETIC AND CULTURAL ORIGINS OF
ALGEBRA

HELEN DE CRUZ
Centre for Logic and Philosophy of Science
Vrije Universiteit Brussel
Pleinlaan 2, 1050 Brussels, Belgium
E-mail: hdecmz@vub.ac.be

Algebra has emergent properties that are neither found in the cultural context
in which mathematicians work, nor in the evolved cognitive abilities for m a t h e
matical thought that enable it. In this paper, I argue that an externalization of
mathematical operations in a consistent symbolic notation system is a prerequisite
for these emergent properties. In particular, externalism allows mathematicians
t o perform operations that would be impossible in the mind alone. By comparing
the development of algebra in three distinct historical cultural settings-China,
the medieval Islamic world and early modern Europe-I demonstrate that such
an active externalism requires specific cultural conditions, including a metaphysi-
cal view of the world compatible with science, a notation system that enables the
symbolic notation of operations, and the ontological viewpoint that mathematics
is a human endeavour. I discuss how extending mathematical operations from
the brain into the world gives algebra a degree of autonomy that is impossible to
achieve were it performed in the mind alone.

1. What Would Mathematics without Culture Look Like?

Even a concept as seemingly simple as the number two is a highly abstract
representation. There is no obvious perceptual similarity between two cows
and two bicycles, and yet both collections share the concept of ‘twoness’. A
large number of experimental studies nevertheless indicate that both non-
human animals and human infants have a robust understanding of number
(see Ref. 1 for an overview). For example, when presented with two dif-
ferent numerosities of food-items, red-backed salamanders go for the larger
quantity 2 . Human infants, likewise, can discriminate between small and
larger collections (e.g. two versus three, eight versus sixteen)-even when
non-numerical variables such as cumulative surface area are controlled for ’.
They can solve simple addition and subtraction problems involving both

338
 339

‘.
small and larger quantities Recently, single cell recordings in rhesus mon-
keys have identified number-sensitive neurons-individual neurons that
respond only to changes in number, while remaining insensitive to changes
in shape or size. Each neuron is tuned to a preferred quantity: a neuron
preferentially firing at, say two, will fire a bit less at one or three, and even
less when observing higher quantities. These neural tuning curves become
broader as quantities increase. Numerosities are thus not represented as a
linear mental number line, but more as a logarithmic ruler, in which the
psychological distance between one and two is considerably greater than
between fifty and one hundred.
Yet, mathematical systems without exact number representations can
only capture the most rudimentary of numerical relationships. Evidence
for this claim comes from two Amazonian cultures, the Pirahii and the
Munduruk6 7, where exact number words do not exist, such that the word
for ‘one’ can mean ‘two’ and vice versa. People in these cultures cannot
even discriminate a box with three fish from a box with four fish painted
on it. They do not possess a counting routine or any other cultural tool
which helps them to construct positive integer values. Conversely, despite
extensive training, no non-human animal has yet been able to learn com-
plex mathematical operations, such as exact positive integer representa-
tions. One long-term training programme extending over twenty years in-
volved teaching Ai, a female chimpanzee, to understand and produce Arabic
digits’. Ai never managed to count more than nine items, and never gen-
eralized to the counting procedure that children master with ease. These
lines of evidence combined suggest that although our mathematical abilities
build upon an evolved number sense which we share with other animals,
they are clearly more than that. It is as yet unclear how this cognitive
adaptation can account for the vast proliferation and complexity of cul-
tural mathematical concepts.
In this paper, I will argue that complex mathematical theory can emerge
if humans extend their minds into a symbolic notation system. Using al-
gebra as a case-study, I will demonstrate how humans can overcome their
cognitive limitations by externalizing operations that are difficult or impos-
sible to perform in the mind alone. I begin by outlining which cognitive
processes are involved when people learn and perform algebra. I show how
humans combine several evolved specialized neural circuits to solve equa-
tions, but that learning algebra still requires synaptic rewiring. I then go
on to explain how cultural factors, including religion, symbolic notation
and philosophy influence the development and level of abstraction in alge-
340

braic systems in China, the Islamic world and Europe. I argue that this
development critically depends upon the elaboration of a symbolic nota-
tion system, and the assumption that mathematics is a human endeavour,
which can be improved by individual mathematicians. Finally, I discuss
how mathematics can have emergent properties as a result of this external-
ism. Mathematics may be a hybrid system of knowledge, in that it contains
operations that are performable in the head, and operations that can only
exist as symbols.

2. Cognitive Processes Involved in Learning and

Performing Algebra
Classical algebra is a set of procedures and rules for solving equations with
one or more unknowns. How does one acquire this procedural knowledge?
It is important to note that the transmission of a concept such as ‘democ-
racy’ does not entail that this concept is downloaded from one brain to
another: rather, when I hear about democracy I reconstruct it in my own
mind, and it is quite likely that internal representations of democracy are
not identical ’. Some concepts are more easily transmitted than others
because they require minimal cultural input-in such cases the rest of the
concept is ’filled in’ by the recipient’s tacit assumptions. For example, some-
one who learns that a zombie does not move according to his own free will
tacitly assumes that such a being does obey the laws of physics. Thus, only
the counterintuitive part of this concept-a violation of intuitions about
agency-needs to be transmitted lo. Given the minimal transmission re-
quirements for such religious concepts, it is not surprising that they should
have evolved many times independently, occurring throughout a wide array
of physical and cultural settings. Neither their birth nor persistence seems
to require exceptional cultural conditions l l .
In stark contrast to religious concepts, algebra is non-intuitive: it re-
quires a co-optation of evolved cognitive strategies which are normally not
deployed together. This results in a new system with emergent properties
not present in any of the subsystems. We can observe this in other do-
mains as well: hunter-gatherers for instance, track down animals using a
combination of several domains of knowledge outside of biology, including
physics (like examining spoor and damage the animal caused to plants)
and psychology (imaging what one would do if one were in the position of
the prey) 12. As a result, the emergent properties of tracking enable the
hunter to work out solutions which would have been impossible in each of
 341

the subsystems. The drawback for these enhanced cognitive abilities is that
such skills are typically hard to learn. Male hunter-gatherers start learning
to track when they are adolescents, but only gain expertise-as measured
by their return-rates-when they are well in their thirties 13.
Likewise, fMRI studies (e.g. Ref. 14) found that people use different
brain circuits when they solve equations (Fig. 1):

i mantenor cingulatecortex
bilateral horizontalsegment of the intraparietal sulcu!
--.---- bilateralposteriorsuperior parietallobe
I

Figure 1. Brain areas involved in solving equations (only left hemisphere shown)

e the anterior cingulate cortex, which is otherwise active when people

reflect on other people's mental states l5
e the intraparietal sulci, which have been implicated in brain imaging
studies involving number l6
e the posterior parietal cortex, which is normally activated in vi-
suospatial tasks, including spatial working memory and attention
orienting l'.
One study investigating the difference between mathematical strategies in
normal and in precocious teenagers found that the latter deployed spatial
skills to solve algebraic problems by diagramming important relationships
apparent in the problem 18. Thus, solving equations depends on a successful
co-optation of several brain areas which are normally involved in ecolog-
ically relevant tasks such as detecting numerosities and assessing spatial
relationships. Perhaps the most surprising finding is that after a learning
period adolescents exhibited alterations in their cerebral blood flow when
solving equations 19. In other words, the brain rewires its synaptic connec-
tions in response to the task-demands. Thus, we cannot explain cultural
342

mathematical concepts solely as a result of the combination of specialized

neural circuits, since learning these produces permanent synaptic changes
in a few well-restricted areas of the brain.

3. Cultural Transmission of Algebra in Context

Like other domains of human culture, algebra has emergent properties
which cannot be traced back t o the innate cognitive architecture of our
brains. Human culture is cumulative, unlike social traditions in non-human
species. Our cultural systems build and elaborate upon the inventions of
previous generations, which has resulted in a vast proliferation of artefact
types, customs, religious beliefs and social systems that vary across cul-
tures. Why only humans can benefit from this ratchet effect of cultural
evolution is not entirely clear. One proposal, the extended mind thesis,
formulated by Clark and Chalmers 2o is that humans extend thought pro-
cesses into the external world. External memory devices such as books or
electronically stored documents, instruments such as computers or nautical
slide rulers, or simply pen and paper, all serve to extend computations be-
yond the brain. Clearly, the use of external media enables us t o accumulate
information beyond the scope of the individual memory. Moreover, work-
ing with external symbolic media may constitute an epistemic act in itself.
Some actions performed by use of external media solve problems more eas-
ily and reliably than if they had been solved in the mind alone. Take the
computer game Tetris: physically rotating the two-dimensional blocks by
means of a keyboard in order to fit them in the slots proves far more effi-
cient than mental rotation 21. Note that, in all these cases, cognition only
takes place within the human brain 22.
The use of external media makes computational solutions possible that
could not have been reached without them. In mathematics, a positional
system such as the Hindu numerals renders multiplications with large num-
bers easy and transparent, while multiplication in non-positional systems
such as the Roman numerals is a far more daunting task. As I shall demon-
strate, the absence or presence of symbols that represent operations influ-
ences the degree of abstraction within specific mathematical systems. Only
in early modern Europe did various cultural conditions shape the possibility
of a consistent externalism of mathematical ideas.
 343

3.1. Chinese Algebra

Chinese algebra first bloomed around the beginning of the Common Era
with the anonymous Nine Chapters on the Mathematical Arts (Chiu Chang
Suan Shu). It was organized around practical problems of public services,
such as taxation and digging canals 23. During the T’ang dynasty (618-
906), China exhibited an unusual openness to foreign influences, welcoming
Arab and Indian scholars. Thanks to this influx, by the end of the Sung
dynasty (900-1279), Chinese algebra had attained a level that would only
be surpassed in Europe in the course of the eighteenth century. During
the Ming dynasty (1368-1644), as a reaction to the Mongol invasions, in-
digenous culture and values were restored. This led to a stagnation and
decline in all sciences. Finally, during the Qing dynasty (1644-1911), there
was an increase in influence from European mathematics, which led to the
demise of Chinese algebra and other mathematical systems 24. Why did
Chinese algebra come to a standstill, and why was it eventually abandoned
in favour of Western algebra? A possible factor is the notation system,
shown in Fig. 2. To represent a system of equations, counting rods were
arranged in such a way that each column was assigned to each equation,
and each row corresponded to the coefficients of each unknown. This in
turn promoted the invention of matrix methods to solve simultaneous lin-
ear equations and higher-order equations. Figure 2, for example, represents
the following equations:
+
2~ - 3y 82 = 32

{ - 6 ~- 2y - z = 6 2
+
3x 21y - 3 z = 0
However, these rods were less useful to express general abstract rules other
than actual calculations, which preserved the concreteness of Chinese math-
ematics 23. Consequently, Chinese algebra textbooks never attempted to
give an abstract formulation of a general rule, but presented examples that
served as paradigms to solve similar problems 2 5 . Next to this, the venera-
ble status of the Nine Chapters on the Mathematical Arts impeded further
progress. Numerous mathematicians wrote commentaries on it, and as
time went by commentaries on those commentaries, and even those who
produced original work felt obliged to extensively refer to it. This attitude
stems from the Confucian point of view that only the wise sages of the past
have attained true wisdom; it was the duty of aspiring scholars to emulate
their mental state 26. The high status of mathematicians gradually eroded
as mathematics came to be perceived as a diligent and unquestioning appli-
344

II.. 1 coefficients

___
I
I
111 II
I

?quations
Figure 2. Representing simultaneous equations with counting rods. Red rods (here
shown in grey) indicate positive coefficients, black rods negative coefficients. Redrawn,
with permission, from Ref. 23, p. 146, Fig. 4.6.

cation of ancient wisdom, rather than an ongoing creative process 23. Chi-
nese mathematics never became an autonomous discipline: mathematicians
remained technical experts dealing with chronology, finances, taxation, ar-
chitecture, and the military 24.

3.2. Islamic Algebra

Islamic algebra developed as a combination of and elaboration upon cul-
tural inventions borrowed from other traditions, notably Greek geometry,
Indian numerals and arithmetic, and Nestorian astronomy. This was possi-
ble because of the patronage of the Abbasid caliphate (758-1258), centred in
Baghdad. It commissioned translations of major works by Aristotle, Galen,
and many Persian and Indian scholars into Arabic. This knowledge became
appropriated and reconstructed in Islamic mathematics 27. It built upon a
combination of Greek geometric visualization and ‘proofs’, and Indian and
Chinese arithmetic rules to solve equations quickly and efficiently. Through
this eclectic approach, Arabian mathematicians could formulate solutions
that they could not find through arithmetic alone. When, for example,
Omar Khayyhrn (1048-1123) failed to find an arithmetic solution to cubic
 345

equations, he resorted to a geometric method of intersecting cones to solve

them 23. After the fourteenth century, Islamic mathematics and other sci-
ences apparently stagnated and declined. A likely cause for this dwindling
was the growing incompatibility between Islamic metaphysics and science.
According to the Qu’ran, the world is governed by Allah’s uninterrupted
control of all events. This line of thought inhibits the possibility of a world
governed by rational, knowable and coherent laws of nature, being a prin-
ciple of Greek philosophy. Nevertheless, between the eighth and eleventh
century, Islam did adopt scientific thought in the form of philosophy, astron-
omy, medicine and mathematics. However, influential theologians such as
al-Ash’ari (873-935) and al-Ghazali (1058-1111) questioned the validity of
Greek philosophy which they saw as a menace to religious orthodoxy. They
became committed to Islamic occasionalism. According to this doctrine,
the basic building blocks of nature are individual indivisible ‘atoms’, each
of which lasts only a moment and then disappears. Allah creates the world
anew each moment of time, thereby giving it pattern and persistence. The
world by itself cannot exist without this continuous creation and is therefore
not ruled by laws of nature. Such a worldview is deeply incompatible with
the basic assumptions of science that the world is governed by rational, co-
herent and orderly laws 28. Mathematics and its application in astronomy
and other branches of science did not escape this devastating attack. The
fourteenth century theologian al-Iji wrote that “we can disregard mathe-
matical entities as they are more tenuous than a spider’s web”. From the
tenth century onwards, occasionalism became the dominant orthodoxy-
incidentally, it still is. Hampered by religious orthodoxy, mathematics and
other sciences could no longer flourish, with the exception of some applied
mathematics that had highly political or religious importance, such as the
construction of a round cupola on the square basis of mausoleums for polit-
ical rulers, or the calculation of the qibla, the relative position of Mecca 27.
Like in China, mathematics never became an autonomous discipline, but
remained ancillary to engineering, cartography, optics and geography.

3.3. European Algebra

European algebra starts in the late thirteenth century commercial cities
of the Catalonian, Provencal and northern Italian regions among mathe-
maticians with a background in teaching commercial calculation methods.
However, while showing substantial influence from Islamic mathematics,
this algebra went beyond it in its ambition to solve irreducible higher-degree
346

problems, starting with Jacopo da Firenze’s Tractatus algorismi (1307) 29.

Since the seventeenth century, mathematicians began to work at increas-
ingly abstract problems, including algebraic series and the calculus, culmi-
nating in collective conflicts among synthesizing schools with rival programs
on the foundations and nature of mathematics.
A disparate number of cultural factors have jointly contributed to this
evolution. In contrast to Islamic metaphysics, Christian metaphysics en-
dorsed the notions of natural laws and causality. The Scholastics saw ev-
idence of God’s harmonious creation, fully in line with Plato’s system of
nature, as a system of causal necessity. The metaphor of this machina
mundi is ubiquitous in their writings. The arrival of Arabic translations
of Aristotle in the thirteenth century gave this enthusiasm for naturalistic
enquiry another powerful boost. Although modern science has traditionally
been seen as a child of the Enlightenment, this development has actually
been more gradual. In particular, the emergence of Nominalism within
Scholasticism in the second half of the thirteenth century, was an impor-
tant factor in the development of science. The Nominalists introduced
empirical investigation, arguing that we can only discover the world by ob-
servation 30. Medieval universities institutionalized philosophical enquiry,
making it acceptable that knowledge acquired through scientific investiga-
tion, became a valid source of knowledge next to the Scriptures 28. Thus,
seventeenth century science can be seen as a second phase in an intellectual
movement that began in the thirteenth century. Nevertheless, it was not
until the seventeenth century that a fundamental shift in the perception
of science enabled turnovers such as the transformation from Ptolemaic to
Keplerian astronomy. Medieval mathematicians believed that mathematics
had deep roots in classical Greece: there always remained an unchanging
body of knowledge which could at best be preserved and at worst corrupted
by contemporary scholars. Gradually, however, they came to realize that
mathematicians could elaborate and even improve upon this body of knowl-
edge. In this newly emerged paradigm, science was no longer perceived as
ancient revealed knowledge, but as a human endeavour influenced by cul-
ture and circumstance. For example, John Wallis, a seventeenth century
historian of mathematics, not only wrote that the Muslims had developed
and improved classical texts, but proudly stated the further achievements of
his contemporaries 31. European mathematicians thus became increasingly
aware that they themselves created the levels of abstraction in mathemat-
ics. This realization led to many innovations, such as the introduction of
imaginary numbers. Take i, which symbolizes an operation that cannot
 347

actually be performed, namely extracting a square root from a negative

number. By denoting this operation it becomes possible to integrate roots
from negative numbers in operations, thus vastly extending the scope of
mathematics. Only European mathematics has consistently aimed at rep-
resenting such opaque operations by a constant set of symbols. Indian
mathematicians, for example, used the symbol 0 for both zero and the un-
known in equations 23. Such ambiguities, or in the case of Chinese algebra,
a lack of external symbols for operations, inevitably limit the degree of
abstraction in mathematics.

4. The Extended Mind and the Autonomy of Mathematical

Thought
External symbols can provide anchors for thoughts that are difficult to un-
derstand or represent. Without them, such thoughts would not survive long
in the competition for attention and for cognitive resources that character-
izes cultural evolution. Indeed, theoretical models that examine cultural
representations from an epidemiological perspective predict that concepts
that are hard to learn and hard to represent are quickly outcompeted in
favour of intuitive or attention-grabbing ideas. These predictions have been
experimentally confirmed in a variety of cultural settings. For example,
story recall experiments illustrate that both Hindu 32 and Christian 33 col-
lege students do not intuitively think about their respective gods as their
theologies require. Christians have difficulties representing God as an om-
niscient, omnipresent being and distort stories about him to fit intuitive
expectations they have about normal people, like that he can only attend
to one person or one event at the same time. External representations can
anchor non-intuitive concepts on a more permanent basis. Seeing mathe-
matical symbols facilitates the recall of previously stored knowledge. Once
mathematical concepts are nested outside of the brain, their evolution and
cultural transmission is less vulnerable to corruption by individual mathe-
maticians or to competition from ideas that are easier to learn, that speak
more to the imagination, or that pose less computational demands. They
gain a degree of autonomy that would be impossible to attain were they rep-
resented in the mind alone. In such distributed cognitive systems 34 both
the interaction of human minds with external notation systems, and the in-
teraction of many mathematicians’ minds enhance the cognitive capacities
of the brain.
The consistent striving of European mathematicians to externalize op-
348

erations has given mathematics a unique representational power. This ‘un-

reasonable’ efficiency of mathematics as an epistemic tool is, I argue, an
emergent property of its hybrid structure. Since mathematicians routinely
express operations that can never be performed in the mind alone (e.g. a
negative number, which results from the subtraction of a larger number
from a smaller one), mathematics can convey a range of ideas that are
actually opaque. Let us examine this distinction between transparent and
opaque more closely. A transparent concept is one to which we have seman-
tic access; we intuitively grasp its meaning. In mathematics, Arabic numer-
als which denote positive integers (1, 2, 3, . . .) are transparent concepts.
Studies of brain activation in adults and even five-year-olds have shown that
a number comparison task with Arabic numerals (say 5 and 3) activates
the same brain areas as one that involves arrays of dots (say five dots and
three dots)--even the speed of computation is identical 35. Thus, the brain
immediately translates a positive integer into a mental representation of its
quantity. In contrast, when confronted with i or -3, no such translation
takes place, because these operations cannot actually be performed in the
mind (it is impossible to imagine a negative quantity). The only way to
represent them is through an external set of symbols, the meaning of which
remains semantically inaccessible to us. Through this externalization, we
expand our representational abilities by delegating operations that are im-
possible to perform in the brain to our external environment, by denoting
them through a symbolic system. Mathematical concepts, in themselves
a hybrid structure composed of transparent and non-transparent symbolic
notations, can in its turn become an epistemic tool. We denote processes
in nature through mathematical formulations, because the latter have no
clear semantic content. Therefore, identical functions and equations can be
used in disparate contexts, including physics, biology, and anthropology.

5. Conclusion
In this paper, I have argued that algebra has emergent properties, resulting
from the co-optation of evolved mental abilities and the externalisation of
operations in a symbolic system. Such a system needs extensive cultural
support, both from the metaphysics of the culture it belongs to and the
ontological status of mathematics. It is crucial for mathematics and other
sciences that a metaphysical view is endorsed in which the world is gov-
erned by causal laws, and in which humans are capable of acquiring reliable
knowledge of the world. Once mathematicians realized their practice is a
 349

human endeavour, and that they can improve it, they could create higher
levels of abstraction by symbolically representing operations. Early mod-
ern European mathematics was not superior t o its Chinese and Arabian
contemporaries-rather, it could evolve because cultural conditions were
more favourable.

Acknowledgements
I gratefully acknowledge helpful comments by Johan De Smedt and Jean
Paul Van Bendegem on a n earlier draft. This research is supported by grant
OZR916BOF from the Free University of Brussels.

References
1. H. De Cruz, How do cultural numerical concepts build upon an evolved num-
ber sense? In: B. G. Bara, L. Barsalou and M. Bucciarelli (Eds.), Proceedings
of the X X V I I annual conference of the Cognitive Science Society. Mahwah,
Lawrence Erlbaum, 565-570 (2005).
2. C. Uller, R. Jaeger, G. Guidry and C. Martin, Salamanders (Plethodon
cinereus) go for more: rudiments of number in an amphibian. Animal Cog-
nition, 6,105-112 (2003).
3. F. Xu and E. S. Spelke, Large number discrimination in 6-month-old infants.
Cognition, 74, B1-B11 (2000).
4. K. McCrink and K. Wynn, Large-number addition and subtraction by 9-
month-old infants. Psychological Science, 15, 776-781 (2004).
5. A. Nieder, D. J. Freedman and E. K. Miller, Representation of the quantity
of visual items in the primate prefrontal cortex. Science, 297, 1708-1711
(2002).
6. P. Gordon, Numerical cognition without words: evidence from Amazonia.
Science, 306,496499 (2004).
7. P. Pica, C. Lemer, V. Izard and S. Dehaene, Exact and approximate arith-
metic in an Amazonian indigene group. Science, 306,499-503 (2004).
8. D. Biro and T. Matsuzawa, Chimpanzee numerical competence: cardinal and
ordinal skills. In: T. Matsuzawa (Ed.), Primate origins of human cognition
and behavior. Tokyo, Berlin, Springer, 199-225 (2001).
9. D. Sperber, Eqlaining culture. A naturalistic approach. Oxford: Blackwell
(1996).
10. P. Boyer, Evolution of the modern mind and the origins of culture: religious
concepts as a limiting-case. In: P. Carruthers and A. Chamberlain (Eds.),
Evolution and the human mind. Modularity, language and meta-cognition.
Cambridge, Cambridge University Press, 93-112 (2000).
11. R. N. McCauley, The naturalness of religion and the unnaturalness of sci-
ence. In: F. C. Keil and R.A. Wilson (Eds.), Explanation and cognition.
Cambridge, Ma., MIT Press, 61-85 (2000).
350

12. P. Carruthers, The roots of scientific reasoning: infancy, modularity and the
art of tracking. In: P. Carruthers, S. Stich and M. Siegal (Eds.), The cognitive
basis of science. Cambridge, Cambridge University Press, 73-95 (2002).
13. H. Kaplan, K. Hill, J. Lancaster and M. Hurtado, A theory of human life
history evolution: diet, intelligence and longevity. Evolutionary Anthropology,
9, 156-185 (2000).
14. Y. &in, C. S. Carter, E. M. Silk, V. A. Stenger, K. Fissell, A. Goode and J.
R. Anderson, The change of the brain activation patterns as children learn
algebra equation solving. Proceedings of the National Academy of Sciences of
the United States of America, 101,5686-5691 (2004).
15. U. Frith and C. Frith, The biological basis of social interaction. Current
Directions in Psychological Science, 10, 151-155 (2001).
16. S. Dehaene, E. S. Spelke, P. Pinel, R. Stanescu and S. Tsivkin, Sources
of mathematical thinking: behavioral and brain-imaging evidence. Science,
284,970-974 (1999).
17. 0. Simon, J.-F. Mangin, L. Cohen, D. Le Bihan and S. Dehaene, Topograph-
ical layout of hand, eye, calculation, and language-related areas in the human
parietal lobe. Neuron, 33, 475-487 (2002).
18. V. J. Dark and C . P. Benbow, Differential enhancement of working memory
with mathematical versus verbal precocity. Journal of Educational Psychol-
ogy, 83, 48-56 (1991).
19. B. Luna, Algebra and the adolescent brain. Trends in Cognitive Sciences, 8,
437-439 (2004).
20. A. Clark and D. Chalmers, The extended mind. Analysis, 58,7-19 (1998).
21. D. Kirsh and P. Maglio, On distinguishing epistemic from pragmatic action.
Cognitive Science, 18,513-549 (1994).
22. F. Adams and K. Aizawa, The bounds of cognition. Philosophical Psychology,
14,43-64 (2001).
23. G. G. Joseph, The crest of the peacock: non-European roots of mathematics.
Second edition. Princeton: Princeton University Press (2000).
24. S. Restivo, Mathematics in society and history. Sociological inquiries. Dor-
drecht: Kluwer Academic Publishers (1992).
25. K. Chemla, Generality above abstraction: the general expressed in terms of
the paradigmatic in mathematics in ancient China. Science in Context, 16,
413-458 (2003).
26. T. E. Huff, The rise of early modern science. Islam, China and the West.
Second Edition. Cambridge: Cambridge University Press (2003).
27. J. L. Berggen, Mathematics and her sisters in medieval Islam: a selective
review of work done from 1985 t o 1995. Historia Mathematica, 24,407-440
(1997).
28. T. E. Huff, Science and metaphysics in the three religions of the book. Zntel-
lectual Discourse, 8, 173-198 (2000).
29. J. Hoyrup, Jacopo d a Firenze and the beginning of Italian vernacular algebra.
Historia Mathematica, 33, 4-42 (2006).
30. I. Pyysiainen, ‘God’ as ultimate reality in religion and science. Ultimate Re-
ality and Meaning, 22,106-123 (1999).
 351

31. J. A. Stedall, Of our own nation: John Wallis’s account of mathematical

learning in medieval England. Historia Mathematica, 28, 73-122 (2001).
32. J. L. Barrett, Cognitive constraints on Hindu concepts of the divine. Journal
for the Scientific Study of Religion, 37, 608-619 (1998).
33. J. L. Barrett and F. C. Keil, Conceptualizing a nonnatural entity: anthro-
pomorphism in god concepts. Cognitive Psychology, 31, 21S247 (1996).
34. R. N. Giere, The problem of agency in scientific distributed cognitive systems.
Journal of Cognition and Culture, 4, 759-774 (2004).
35. E. Temple and M. I. Posner, Brain mechanisms of quantity are similar in 5-
year-old children and adults. Proceedings of the National Academy of Sciences
of the United States of America, 95, 7836-7841 (1998).