Professional Documents
Culture Documents
Emergence Explained Entities 07.02
Emergence Explained Entities 07.02
has probably run its course. know that they are not ex-
Just as it doesn't make sense pressed in terms of cold fronts
to talk about the hardness or or thunderstorms. …
temperature or intelligence of Every field of science operates
individual "elementary" by formulating and testing
particles, it is also not possible generalizations that are some-
to give a precise meaning to times dignified by being called
statements about particles be- principles or laws. … But there
ing composed of other are no principles of, [for ex-
particles. We do speak loosely ample,] chemistry that simply
of a proton as being composed stand on their own, without
of three quarks, but if you look needing to be explained re-
very closely at a quark you will ductively from the properties
find it surrounded with a cloud of electrons and atomic nuclei,
of quarks and anti-quarks and and in the same way there are
other particles, occasionally no principles of psychology
bound into protons; so at least that are free-standing, in the
for a brief moment we could sense that they do not need ul-
say that the quark is made of timately to be understood
protons. through the study of the hu-
Weinberg continues his explication of man brain, which in turn must
grand reductionism by using the weather ultimately be understood on
as an example. the basis of physics and chem-
[T]he reductionist regards the istry.
general theories governing air Thus the battle is joined: can all the laws
and water and radiation as be- that govern the functioning of higher
ing at a deeper level than the- level entities be derived from physics?
ories about cold fronts or thun-
derstorms, not in the sense 3 Epiphenomenal causation
that they are more useful, but In this section we examine the concepts
only in the sense that the lat- of epiphenomena and supervenience. We
ter can in principle be under- also define emergent as synonymous
stood as mathematical con- with epiphenomenal.
sequences of the former. The
reductionist program of phys- 3.1 Epiphenomena and Emer-
ics is the search for the com- gence
mon source of all explanations. If one doesn’t already have a sense of
… what it means, the term epiphenomenon
is quite difficult to understand. Here is
Reductionism … provides the the WordNet definition [WordNet],
necessary insight that there which is representative.
are no autonomous laws of
weather that are logically inde- A secondary phenomenon that
pendent of the principles of is a by-product of another phe-
physics. … We don't know the nomenon.
final laws of nature, but we
It is not clear that this definition pins or implementation of the systems that
much down. It’s especially troublesome realize them. System requirements are
because the terms secondary and by- written before systems are designed, and
product should not be interpreted to software specifications are intended to
mean that an epiphenomenon is separate be implementation-independent. Thus
from and a consequence of the state of system requirements and software spe-
affairs characterized by the “other” phe- cifications described properties that sys-
nomenon. tems and software are intended to exhibit
once implemented.
We suggest that a better way to think of
an epiphenomenon is as an alternative The primary difference between proper-
way of apprehending or perceiving a ties that we refer to as emergent and
given state of affairs. Consider Brownian those that result from designing a system
motion, which appears to be motion that to satisfy a specification is that we tend
very small particles of non-organic ma- to reserve the term emergent for proper-
terials are able to engage in on their ties that appear in systems that are not
own. Before Einstein, Brownian motion explicitly designed to produce them. But
was a mystery. How could inanimate this distinction is one of how we have
matter move on its own? We now know tended to use these terms rather than one
that Brownian motion is an epiphen- that distinguishes different kinds of
omenon of collisions of particles with properties.
atoms or molecules.
[Sidebar] Emergence and surprise
The key here is that we observed a phe- Emergence may sometimes seem like a
nomenon—the motion of inorganic magic trick: we see that it happens but
particles—without knowing what we didn’t anticipate it, and we don’t un-
brought it about. What we observed was derstand how it’s done. This may be why
an epiphenomenon of some underlying emergence is sometimes associated with
reality. With this usage as a guide we surprise. We believe it is wrong to rely
define the term epiphenomenon as fol- on surprise as a characteristic of emer-
lows. gence. An observer’s surprise or lack of
Epiphenomenon. A phenomenon that can surprise should have nothing to do with
be described in terms that do not depend how we understand a natural phenomen-
on the underlying phenomena that bring on.
it about.
[Sidebar] Two simple examples of
We define emergent as synonymous with
emergence
epiphenomenal. In other words, a phe- Even very simple systems may exhibit
nomenon is emergent if it may be char- emergence. Here are two examples.
acterized independently of its imple-
mentation. Consider a satellite in geosynchronous
orbit. It has the property that it is fixed
Defined in this way, emergence is syn- with respect to the earth as a reference
onymous with concepts familiar from frame. This property is emergent be-
Systems Engineering and Computer Sci- cause it may be specified independently
ence. System requirements and software of how it is brought about.
specifications are by intention written in
terms that do not depend on the design
A satellite tethered to the ground like a lower). The properties in H and L are all
balloon by a long cable (were that pos- presumed to be applicable to some com-
sible) would also be fixed. Of course mon domain of discourse.
that’s not how geosynchronicity works. H and L are each ways of characterizing
A geosynchronous orbit (a) circles the the state of affairs of the underlying do-
earth at the equator and (b) has a period main. For any particular state of affairs
that matches the earth’s period of rota- in the domain of discourse, the predic-
tion. ates in H and L will each be either true
If emergence is considered a property of or false (or perhaps not applicable).
a complex systems, such a two-element One says that H supervenes on (or over)
system is probably as simple a complex L if it is never5 the case that two states of
system as one can imagine.4 affairs will assign the same configura-
As a second example consider the fol- tion of values to the elements of L but
lowing code snippet. different configuration of values to the
elements of H. In other words when a
temp := x;
x := y; state of affairs assigns values to predic-
y := temp; ates in L, that fixes the assignments of
values to predicates of H.
This familiar idiom exchanges the values
in x and y. Since this property may be Consider the following simple example.
specified independently of the code we Let the domain be a sequence of n bits.
say that the exchange of x and y emerges Let L be the statements: bit 1 is on; bit 2
when this code is executed. is on; etc. Let H be statements of the
sort: exactly 5 bits are on; an even num-
Even though both of these examples are ber of bits are on; no two successive bits
extremely simple they both consist of are on; the bits that are on form the ini-
multiple independent actions. From our tial values in the Fibonacci sequence;
perspective for a system to be considered etc.
complex it must consist of two or more
independently operating elements. It is H supervenes on L since any configura-
generally the interaction of two or more tion of values of the statements in L de-
independent actions that produces effects termines the values of the statements in
that we refer to as emergent. H.
However, if we remove one of the state-
3.2 Supervenience ments from L, e.g., we don’t include in L
A term from the philosophical literature a statement about bit 3, but we leave the
that is closely related to emergence is statements in H alone, then H does not
supervenience. The intended use of this supervene on L.
term is to relate a presumably higher
level set of predicates (call such a set H To see why, consider the H statement
for higher) to a presumably lower level An even number of bits is on. (h1)
set of predicates (call such a set L for 5
Some definitions require that not only is
it never the case, it never can be the case. It does
4
Jonathan von Post (private communica- make a formal difference whether we base super-
tion) tells the story of how Arthur C. Clarke once venience on a logical impossibility or on empirical
applied for a British patent for geosynchronous or- facts. We finesse that distinction by adopting the
bits. It was rejected as impractical. Imagine the rule of thumb of fundamental particle physicists: if
lost royalties! something can happen it will.
For concreteness, let’s assume that there of bits. Consider H consisting of the fol-
are exactly 5 bits. Let’s assume first, as lowing single statement
in the first line of Figure 1, that all the The bits that are on are prime. (h2)
bits except bit 3, the one for which there
is no L statement, are on. Since there is Clearly H supervenes over L. Just as
no L statement about bit 3, all the L clearly, H does not supervene over any
statements are true even though bit 3 is proper subset of L, and certainly not
off. Since 4 of the 5 bits are on, h1 is over any finite subset of L—one needs
also true. to look at all of the bits to determine
whether it is exactly the prime bits that
Now, assume that bit 3 is on as in the are on.
second line of Figure 1. All the L state-
ments are still true. But since 5 bits are So even though we can conclude that H
now on, h1 is now false. supervenes over the infinite set of state-
ments in L, that information doesn’t buy
Since we have found an H statement that us much since we can’t discard H and re-
has two different values for a single con- place it in any practical way with all the
figuration of values of the L statements, members of L. A similar problem arises
H does not supervene over L. when we consider dynamic emergence.
The notion of supervenience captures the Returning to Weinberg and Fodor, pre-
relationship between epiphenomena and sumably both would agree that phenom-
their underlying phenomena. Epiphen- ena of the special sciences supervene on
omena supervene on underlying phe- phenomena in physics. A given set of
nomena: distinct epiphenomena must be phenomena at the level of fundamental
associated with distinct underlying phe- physics is associated with no more than
nomena, which is what one wants. You one set of phenomena at the level of any
can’t get two different sets of epiphen- of the special sciences.
omena from the same underlying phe-
nomena. Or looking top-down, two different
states of affairs in some special science
Note that the reverse is not true. Two dif- must be associated with two different
ferent states of the underlying phenom- states of affairs at the level of funda-
ena may result in the same epiphenom- mental physics. This is Weinberg’s petty
ena. In our bit example, there are many reductionism.
different ways in which an even number
of bits may be on. Where Weinberg and Fodor presumably
disagree is whether the principles of the
It would appear that the relationship special sciences can be derived from the
defined by supervenience will be useful principles of physics.
in analyzing multi-scale phenomena. To
some extent this is the case. But super-
4 Epiphenomenal causation
venience is not as useful as one might
Weinberg and Fodor have a second fun-
have hoped. One reason is related to the
damental area of agreement—which has
difficulty one encounters when using su-
some quite significant implications.
pervenience for infinite domains. Con-
Weinberg makes his case sarcastically.
sider our bit example again, but imagine
that we have a countably infinite number Henry Bergson and Darth
Vader notwithstanding, there is
cells (agents) are “alive;” the light cells one might be quite pleased by this result.
(agents) are “dead.” One can apply the It might merit a conference paper.
rules manually and satisfy oneself that
they produce the sequence as shown. 6.2 Gliders in our physics world
From our physics perspective, we note
Notice that the fifth configuration shows that the rules are the only forces in our
the same pattern of live and dead cells as Game of Life universe. Being epiphen-
the first except that the pattern is offset omenal, gliders are causally powerless.13
by one cell to the right and one cell The existence of a glider does not
down. change either how the rules operate or
If there are no other live cells on the which cells will be switched on and off.
grid, this process could be repeated in- Gliders may be emergent, but they do
definitely, producing a glider-like effect. not represent a new force of nature in the
Such a glider is an epiphenomenon of Game of Life universe. It may appear to
the Game of Life rules. If one thinks us as observers that a glider moves
about it—and forgets that one already across the grid and turns cells on as it
knows that the Game of Life can pro- reaches them. But that’s not true. It is
duce gliders—gliders are quite amazing. only the rules that turn cells on and off.
A pattern that traverses the grid arises A glider doesn’t “go to an cell and turn it
from very simple (and local) rules for on.” A Game of Life run will proceed in
turning cells on and off. exactly the same way whether one no-
tices the gliders or not.
We should be clear that gliders are epi-
phenomenal. The rules of the Game of This is a very reductionist position.
Life do nothing but turn individual cells Things happen only as a result of the
on and off. There is nothing in the rules lowest level forces of nature, which in
about waves of cells turning on and off this case are the rules.
sweeping across the grid. Such epiphen-
omenal gliders exemplify emergence. 6.3 The Game of Life as a pro-
gramming platform
• Gliders are not generated explicitly: Amazing as they are, gliders are also
there is no glider algorithm. There is trivial. Once one knows how to produce
no “code” that explicitly decides a glider, it’s a simple matter to make as
which cells should be turned on and many as one wants. If we look at the
off to produce a glider. Game of Life as a programming plat-
• Gliders are not visible in the rules. form—imagine that we are kids fooling
None of the rules are formulated, around with a new toy—we might exper-
either explicitly or implicitly, in iment with it to see whether we can
terms of gliders. make other sorts of patterns. If we find
some, which we will, we might want to
When looked at from our agent-based see what happens when patterns crash
modeling perspective, gliders may rep- into each other—boys will be boys.
resent epidemics or waves of births and
deaths. If one were attempting to demon- 13
All epiphenomena are causally power-
strate that such waves could be gener- less. Since epiphenomena are simply another way
of perceiving underlying phenomena, an epiphen-
ated by simple agent-agent interactions, omenon itself cannot have an effect on anything. It
is the underlying phenomena that act. This is a
point that Kim makes repeatedly.
After some time and effort, we might which cells will be off. No matter how
compile a library of Game of Life pat- real the patterns look to us, interaction
terns, including the API14 of each pattern, among them is always epiphenomenal.
which describes what happens when that So what are we talking about?
pattern collides with other patterns.15, 16, 17 What does one do to show that a Game
It has even been shown [Rendell] that by of Life implement of a Turing machine
suitably arranging Game of Life pat- is correct? One must adopt an operation-
terns, one can implement an arbitrary al perspective and treat the patterns and
Turing Machine.18 their interactions, i.e., the design itself,
as real—independently of the Game of
6.4 Designs as abstractions Life. It is the design, i.e., the way in
What did we just say? What does it which the patterns interact, that we want
mean to say that epiphenomenal gliders to claim implements a Turing Machine.
and other epiphenomenal patterns can be To show that we must do two things.
used to implement a Turing Machine?
1. Show that the abstract design con-
How can it mean anything? The patterns
sisting of patterns and their interac-
aren’t real; the Turing Machine isn’t
tions (epiphenomenal or not) actually
real; they are all epiphenomenal.
does produce a Turing Machine.
Furthermore, the interactions between
2. Show that the design can be imple-
and among patterns aren’t real either.
mented on a Game of Life platform.
They’re also epiphenomenal—and epi-
phenomenal in the sense described Note what this perspective does. It un-
above: the only real action is at the most shackles the design from its moorings as
fundamental level, the Game of Life a Game of Life epiphenomenon and lets
rules. Pattern APIs notwithstanding the it float free. (The protestors in the streets
only thing that happens on a Game of chanting “Free the design” can now
Life grid is that the Game of Life rules lower their picket signs and go home.)
determine which cells will be on and The design becomes an independent ab-
14
Application Programming Interface straction. Once we have such a abstrac-
15
Note, however, that interactions among tion we can reason about its properties,
patterns are quite fragile. If two patterns meet in i.e., (a) that it accomplishes what we
slightly different ways, the results will generally
be quite different. want, namely that it performs the func-
16
Since its introduction three decades ago, tionality of a Turing Machine and (b)
an online community of Game of Life program- that it can be reattached to its moorings
mers has developed. That community has created and be implemented on a Game of Life
such libraries. A good place to start is Paul Calla-
han’s “What is the Game of Life?” at platform. In other words, emergence is
http://www.math.com/students/wonders/life/life.ht getting epiphenomena to do real (func-
ml. tional) work.
17
See the appendix for a sketch of how
such a pattern library may be produced. Implementing new functionality by us-
18
The implementation of a Turing Ma- ing mechanisms from an existing library
chine with Game of Life patterns is also an ex- is, of course standard practice in com-
ample of emergence. There is no algorithm. The
Turing Machine appears as a consequence of epi-
puter science. As we hinted earlier, this
phenomenal interactions among epiphenomenal technique is also used by biological or-
patterns. The appearance of gliders and Turing ma- ganisms. In section XXX we explore the
chines is what we refer to in [Abbott, If a Tree] as
non-algorithmic programming.
general applicability of this technique in had thought. In fact they have an Internet
nature. connection. Hacka had learned not only
that she was a Game of Life system but
For now let’s reiterate that the imple-
that the Game of Life can emulate a Tur-
mentation of a Turing Machine on a
ing Machine. She had decided to pro-
Game of Life platform exemplifies
gram herself to do just that. Her parents
emergence (according to our definition)
disapproved, but girls just want to have
because we characterize Turning Ma-
fun.
chines in terms that are independent of
their implementation. No wonder we felt uncertain about our
results. Even though the Game of Life
[Sidebar] Game of Life anthropo- rules explained every light that went on
logists and off on Hacka’s face, it said nothing
Let’s pretend that we are anthropologists about the functionality implemented by
and that a previously unknown tribe has Hacka’s Turing Machine emulation.
been discovered on a remote island. It is
The rules explained everything about
reported that their grid-like faces are
how the system worked; they said noth-
made up of cells that blink on and off.
ing about what the system did. The rules
We get a grant to study them. We travel didn’t have a way even to begin to talk
to their far-off village, and we learn their about the functionality of the system—
language. They can’t seem to explain which is logically independent of the
what makes their cells blink on and off; rules. The rules simply have no way to
we have to figure that out for ourselves. talk about Turing Machines.
After months of study, we come up with A Turing machine is an autonomous
the Game of Life rules as an explanation functional abstraction that we (and
for how the grid cells are controlled. Hacka) built on top of the rules of the
Every single member of the tribe oper- Game of Life. Our reductive explana-
ates in a way that is consistent with those tion, that a certain set of rules make the
rules. The rules even explain the unusual cells go on and off, had no way to cap-
patterns we observe—some of them, ture this sort of additional functionality.
glider-like, traverse the entire grid.
Thrilled with our analysis, we return 7 Implications of emer-
home and publish our results. gence
But one thing continues to nag. One of This section explores the implications of
the teenage girls—she calls herself the sort of emergence illustrated by our
Hacka—has a pattern of activities on her Game of Life Turing Machine.
grid that seems somehow more complex
than the others. The Game of Life rules 7.1 Non-reductive regularities
fully explain every light that goes on and Recall Weinberg’s statement that
every light that goes off on Hacka’s there are no autonomous laws
pretty face. But that explanation just of weather that are logically in-
doesn’t seem to capture everything that’s dependent of the principles of
going on. Did we miss something? physics.
To make a long story short, it turns out Clearly there are lots of autonomous
that the tribe was not as isolated as we “laws” of Turing Machines (namely
computability theory), and they are all know that the halting problem for Turing
logically independent of the rules of the Machines is unsolvable.
Game of Life. In other words, epiphenomena are down-
The fact that one can implement a Tur- ward entailing. Properties of epiphenom-
ing Machine on a Game of Life platform ena are also properties of the phenomena
tells us nothing about Turing Machines from which they spring. This is not quite
—other than that they can be implemen- as striking as downward causation19
ted by using the Game of Life. would be, but it is a powerful intellectual
tool.
An implementation of a Turing Machine
on a Game of Life platform is an ex- Earlier, we dismissed the notion that a
ample of what might be called a non-re- glider may be said to “go to a cell and
ductive regularity. The Turing Machine turns it on.” The only things that turn on
and its implementation is certainly a Game of Life cells are the Game of Life
kind of regularity, but it is a regularity rules. But because of downward entail-
that is not a logical consequence of (i.e., ment, there is hope for talk of this sort.
is not reducible to and cannot be de- Once we establish that a Turing Machine
duced from) the Game of Life rules. can be implemented on a Game of Life
Facts about Turing Machines, i.e., the platform, we can then apply results that
theorems of computability theory, are we derive about Turing Machines as ab-
derived de novo. They are made up out stractions to the Game of Life. We can
of whole cloth; they are not based on the do the same thing with gliders. We can
Game of Life rules. The fact that such establish a domain of discourse about
abstract designs can be realized using gliders as abstract entities. Within that
Game of Life rules as an implementation domain of discourse we can reason about
platform tells us nothing about comput- gliders, and in particular we can reason
ability theory that we don’t already about how fast and in which direction
know. gliders will move.
Having developed facts and rules about
7.2 Downward entailment
gliders as independent abstractions, we
On the other hand, the fact that a Turing
can then use the fact that gliders are epi-
Machine can be implemented using the
phenomena of the Game of Life and—by
Game of Life rules as primitives does
appeal to downward entailment—apply
tell us something about the Game of Life
those facts and rules to the Game of Life
—namely that the results of computabil-
cells that gliders traverse. We can say
ity theory can be applied to the Game of
that a glider goes to a cell and turns it
Life. The property of being Turing com-
on.
plete applies to the Game of Life pre-
cisely because a Turing Machine can be
7.3 Reduction proofs
shown to be one of its possible epiphen- Consider in a bit more detail how we can
omena. Similarly we can conclude that conclude that the Game of Life halting
the halting problem for the Game of Life problem is unsolvable. Because we can
—which we can define as determining implement Turing Machines using the
whether a game of Life run ever reaches
a stable (unchanging or repeating) con- 19
See, for example [Emmeche] for a num-
figuration—is unsolvable because we ber of sophisticated discussions of downward
causation.
Game of Life, we know that we can re- the computation is easy to do at the epi-
duce the halting problem for Turing Ma- phenomenal level of billiard balls. We
chines to the halting problem for the know that the computation we do at the
Game of Life: if we could solve the billiard ball level applies to the real
Game of Life halting problem, we could world because of downward entailment:
solve the Turing Machine halting prob- billiard balls are epiphenomena of the
lem. But we know that the Turing Ma- underlying reality.
chine halting problem is unsolvable. Downward entailment is, in fact, a reas-
Therefore the Game of Life halting prob- onable description of how we do sci-
lem is also unsolvable. ence: we build models, which we then
This sort of downward entailment reduc- apply to the world around us.
tion gives us a lot of intellectual leverage We are not saying that there are forces in
since it’s not at all clear how difficult it the world that operate according to bil-
would be to prove “directly” that the liard ball rules or that there are forces in
halting problem for the Game of Life is the Game of Life that operate according
unsolvable. to glider rules. That would be downward
Thus another consequence of downward causation, a form of strong emergence,
entailment is that reducibility cuts both which we have already ruled out. What
ways. One can conclude that if we are saying is that billiard balls,
something is impossible at a higher level gliders, Turing Machines, and their inter-
it must be impossible at the lower (im- actions can be defined in the abstract.
plementation) level as well. But the only We can reason about them as abstrac-
way to reach that conclusion is to reason tions, and then through downward entail-
about the higher level as an independent ment we can apply the results of that
abstraction and then to reconnect that ab- reasoning to any implementation of
straction to the lower level. Logically in- those abstractions whenever the imple-
dependent higher level abstractions mat- mentation preserves the assumptions re-
ter on their own. quired by the abstraction.
7.4 Downward entailment as sci- 7.5 The reality of higher level ab-
ence stractions
A strikingly familiar example of down- In “Real Patterns” [Dennett ‘91], Den-
ward entailment is the kind of computa- nett argues that when compared with the
tion we do when determining the effect work required to compute the equivalent
of one billiard ball on another in a New- results in terms of primitive forces, one
tonian universe. It’s a simple calculation gets a “stupendous” “scale of compres-
involving vectors and the transfer of kin- sion” when one adopts his notion of an
etic energy. intentional stance [Dennett ‘87]. Al-
though “Real Patterns” doesn’t spell out
In truth there is no fundamental force of
the link explicitly, Dennett’s position ap-
physics corresponding to kinetic energy.
pears to be that because of that intellec-
If one had to compute the consequences
tual advantage, one should treat the on-
of a billiard ball collision in terms of
tologies offered by the intentional stance
quantum states and the electromagnetic
as what he calls “mildly real”—although
force, which is the one that applies, the
he doesn’t spell out in any detail what
task would be impossibly complex. But
regarding something as “mildly real” in- ents held together by forces acting
volves. among those components.
Our claim is that the entities (such as bil- Once one has defined an abstract struc-
liard balls) about which higher level ab- ture of this sort, one can derive proper-
stractions are formulated are real in an ties of matter having this structure. One
objective sense even though interactions can do so without knowing anything
among those entities remain epiphenom- more about either (a) the particular ele-
enal. (We discuss entities in [Abbott-part ments at the lattice nodes or (b) how the
2].) binding forces are implemented. All one
needs to know are the strengths of the
In a recent book [Laughlin], Laughlin ar-
forces and the shape of the lattice.
gues for what he calls collective prin-
ciples of organization, which he finds to From our perspective, both Newton’s
be at least as important as reductionist laws and the solid state of matter are ab-
principles. In discussing Newton’s laws stract organizational designs. They are
he concludes from the fact that (p. 31) abstractions that apply to nature in much
the same way as a Turing Machine ap-
these [otherwise] overwhelm-
plies to certain cell configurations in the
ingly successful laws … make
Game of Life. Laughlin calls the imple-
profoundly wrong predictions
mentation of such an abstraction a pro-
at [the quantum] scale
tectorate.
that
Laughlin points out that protectorates
Newton’s legendary laws have tend to have feasibility ranges, which are
turned out to be emergent. often characterized by size, speed, and
They are not fundamental at temperature. A few molecules of H2O
all but a consequence of the won’t have the usual properties of ice.
aggregation of quantum mat- And ice, like most solids, melts when
ter into macroscopic fluids and heated to a point at which the attractive
solids. … [M]any physicists re- forces are no longer able to preserve the
main in denial. To this day they lattice configuration of the elements.
organize conferences on the Similarly Newton’s laws fail at the
subject and routinely speak quantum level.
about Newton’s laws being an The existence of such feasibility ranges
“approximation” for quantum does not reduce the importance of either
mechanics, valid when the sys- the solid matter abstraction or the New-
tem is large—even though no tonian physics abstraction. They just
legitimate approximation limit the conditions under which nature
scheme has ever been found. is able to implement them.
A second example to which Laughlin The more general point is that nature im-
frequently returns is the solid state of plements a great many such abstract
matter, which, as he points out, exhibits designs. As is the case with computabil-
properties of rigidity and elasticity. The ity theory, which includes many sophist-
solid state of matter may be character- icated results about the Turing machine
ized as material that may be understood abstraction, there are often sophisticated
as a three dimensional lattice of compon- theories that characterize the properties
is, an entity will at any time consist of For functionalism, entities are every-
physical elements arranged in a particu- where: mice and cans are good exam-
lar way. With this decision we are ex- ples. The higher level sciences speak of
cluding from our notion of entity strictly all sorts of entities, including biological
mental constructs such as sets, numbers, entities (e.g., you and me) and social,
concepts, propositions, relationships, political, and economic entities, such as
designs, abstractions, etc. families, states, and corporations.
If the preceding does not formally ex- We propose to characterize an entity as
clude instants, events, and durations, we either atomic or as any demarcatable re-
will explicitly exclude them too. Entities gion that exhibits a persistent and self-
for us will be required to persist in time, perpetuating reduced level of entropy.
but they will not be aspects of time, i.e., Since we are not prepared to define the
instants or durations, or events, whatever term demarcatable region, perhaps de-
an event is. fining entity in terms of a notion as
An entity for us will be either atomic loosely defined as demarcatable region
(not in the sense of being a chemical ele- doesn’t get one very far.20 But the notion
ment but in the more generic sense of of an entity always seems to imply a
having no constituents—if indeed there boundary that distinguishes the entity
are atomic physical elements in nature), from its surroundings. Entities in our
or, if an entity has constituents, it will be sense always have an “inside.”
an epiphenomenon of its constituents. We discuss two kinds of entities: entities
Thus for us non-atomic entities will rep- at an energy equilibrium and entities that
resent one of the most common forms of are far from equilibrium.21
emergence.
It is important to note that since non-
Our purpose in this section is not to atomic entities have a reduced level of
settle the grand philosophical question of entropy, they always have an internal
what one should mean by the terms structure, i.e., a design. Furthermore, the
thing, object or entity but to sketch out design of an entity often allows it to as-
what it means to be an entity in our sume one or more states. A good ex-
sense. Of course we hope that the frame- ample is the design of an atom: a nucleus
work we develop will offer a useful way along with associated electrons in vari-
of thinking about some of the uses to ous orbitals. Among the states of an
which we commonly put the terms atom are those differentiated by the dif-
thing, object, and entity. fering energy levels of its electrons.
8.1 Entities, entropy, designs, It seems pretty clear that we (and other
and functionality animals) have evolved the ability to per-
The standard model of physics includes ceive entities in this sense. Our intuitive
fundamental particles such as electrons, sense of entity seems to map fairly well
photons, quarks, etc. These are entities onto the notion of a persistent demarcat-
which have no constituents. Beyond able region that displays some special
these, one has atomic nuclei, atoms, and 20
As Varzi [Varzi Spring ‘04] points out,
molecules, all of which we want to in- the notion of a boundary is itself quite difficult to
clude in our notion of entity. pin down. Some boundaries, Mt. Everest’s, for ex-
ample, are quite vague.
21
We first proposed this in [Abbott].
order that distinguishes it from its envir- The answer has to do with the strong
onment, i.e., an area that has an internal nuclear force, which implements what is
design. called the “binding energy” of nucleons
within a nucleus. Without going into de-
We use the term design deliberately. En-
tails, the bottom line is that the mass of,
tities implement abstract designs in
say, a helium nucleus (also known as an
much the same way as abstract designs
alpha particle, two protons and two neut-
such as Newtonian mechanics are imple-
rons), which is one of the products of
mented on a larger scale. Because an en-
hydrogen fusion, is less than the sum of
tity implements a particular design, it ex-
the masses of the protons and neutrons
hibits the functionality that its design
that make up an alpha particle when not
produces. One of the tasks of science,
bound together as an alpha particle.22
then, is to decide for any entity (or cat-
egory of entities), what design it embod- The same entity-mass relationship holds
ies and what the implications of that for all physical and chemical entities.
design are for the behavior of that entity The mass of an atom or molecule is
(or those entities). (negligibly) less than the sum of the
masses of its components taken separ-
When nature implements an abstract
ately. The mass of the solar system is
design such as solid matter or Newtonian
(negligibly) less than the mass of the sun
mechanics it is the functionalities that
and the planets when taken separately.
come along with that design—what the
design implies about how matter that im- This fact implies that the entropy of
plements it behaves—that make us inter- these entities is lower than the entropy of
ested in it. These larger scale abstract the components taken separately. In oth-
designs are typically embodied by sub- er words, an entity at an energy equilib-
stances or by arbitrary collections of rium is distinguishable by the fact that it
things. In contrast, the designs that entit- has lower mass and lower entropy than
ies implement produce a particular kind its components taken separately.
of functionality in a constrained and These entities are trivially self-perpetuat-
bounded region. ing in that they are in what is often
called an energy well and require energy
8.2 Entities at an energy equilib-
to pull their components apart. This
rium
gives us a nice metric of entityness for
The entities of physics and chemistry are
at-equilibrium entities: the amount of en-
at an energy equilibrium. A distinguish-
ergy required to pull it apart.
ing feature of these entities is that the
mass of any one of them is strictly smal-
8.3 Entities and emergence are
ler than the sum of the masses of its
fundamental
components. This may be seen most The mechanisms (gravity, the strong
clearly in nuclear fission and fusion, in nuclear force, and the electromagnetic
which one starts and ends with the same force) that expel entropy from at-equilib-
number of atomic components, i.e., elec-
trons, protons, and neutrons—which 22
It turns out that the atomic nucleus with
raises the obvious question: which mass the least mass per nucleon is iron. Energy from fu-
was converted to energy? sion is possible for elements lighter than iron; en-
ergy from fission is possible for elements heavier
than iron. (See [Nave] for a discussion of these
matters.)
rium entities and that hold these entities equilibrium entities to autonomous enti-
together are the fundamental forces of ties.
nature. Intuitively, a dissipative structure typic-
One can say that these mechanisms in ally manifests when energy is pumped
some sense run for free. To the extent into a bounded region. D
that we understand how they work at all, issipative structures typically involve
we attribute their operation to virtual structured activities internal to the re-
particles that pop into and out of exist- gion. A standard example consists of the
ence and that do the work of the force— Bénard convection cycles that form in a
with no extra effort expended anywhere liquid when one surface is heated and
else. the opposite surface is kept cool. (See
There really is a free lunch. Atomic nuc- Figure 4.)
lei form, atoms form, solar systems and A number of interesting phenomena may
galaxies form—all without depleting any be understood as dissipative structures.
energy reservoirs. We are so used to this Consider the distribution of water over
fact that we hardly notice it. But if one the earth. Water is transported from
stands back and observes that at-equilib- place to place via processes that include
rium entities exemplify emergence at its evaporation, atmospheric weather sys-
most basic—an atom is emergent from, tem movements, precipitation, ground-
it is an epiphenomenon of, and it super- water flows, ocean current flows, etc.
venes over its components—we may Taken as a global system, these cycles
conclude that spontaneous emergence is may be understood as a dissipative struc-
fundamental to how nature works. ture that is driven primarily by solar en-
Even so, one might suppose that beyond ergy, which is pumped into the earth’s
combining in these basic ways (as atom- atmosphere and surface structures. All of
ic nuclei, atoms, and astronomical ag- this is played out against a static frame-
gregations held together by gravity), at- work defined and held in place by the
equilibrium entities are not very interest- earth’s surface and its gravitational field.
ing. Standing back again makes it clear We note that our definition of a dissipat-
that this is not the case. Given what we ive structure is quite broad. It includes
have learned during the past half century virtually any energy-consuming device
(and what we still don’t know)—espe- that operates according to some design.
cially about condensed matter physics Consider a digital watch. It converts an
and including, as we said earlier, the inflow of energy into an ongoing series
startling fact that the same matter is cap- of structured internal activities. Does a
able of implementing multiple abstrac- digital watch define a dissipative struc-
tions with radically different properties ture? One may argue that the design of a
—at-equilibrium entities are far from digital watch limits the ways in which it
boring. can respond to an energy inflow. There-
fore the structured activity that arises as
8.4 Dissipative structures
energy is pumped into it should not be
In [Prigogine] (and elsewhere) Prigogine
characterized as a dissipative structure.
discussed what he called a dissipative
structure. We see dissipative structures But any bounded region has only a lim-
as the essential stepping stone from at- ited number of ways in which it can re-
ergy wells that exist in environments that So, what do we mean by autonomy?
include energy flows. Certainly, we no longer believe in any-
thing like vitalism, i.e., that there is such
8.6 Autonomous entities a thing as a “life force” the possession of
The notion of an autonomous entity which differentiates the animate from
seems central to how we look at the the inanimate. But when we speak of
world. autonomous entities, have we done much
• For millennia we have found it con- more than substitute the word autonom-
ous for other words? Do we have a ser-
venient to partition the world into
viceable definition of what it means to
two realms: the animate and the in-
be autonomous?
animate. The inanimate world is
ruled by external forces; the animate In non-political contexts, the term auton-
world is capable of autonomous ac- omous is generally taken to mean
tion. Recall that this is why Browni- something like self-directed or not con-
an motion posed such a problem: trolled by outside forces.24 But defini-
how can inanimate particles look so tions of this sort don’t help much. Per-
much like they are moving autonom- haps self-directed is what we mean by
ously? autonomous. But what do we mean by
self-directed?
• For the past half-millennium western
civilization (and more recently civil- Furthermore any entity (in our sense of
ization world-wide) has pursued, an entity as having some material aspect)
with significant success, the dream is subject to outside, i.e., physical,
of creating autonomous sources of forces. Nothing is free from the laws of
action. We have built machines about physics. So it may not make any sense to
which it can be said that in varying demand that to be autonomous an entity
degrees they act on their own. We must not be controlled by outside forces.
do not yet confuse our machines with The intuition behind self-directed and
biological life, and we have not yet the connection to outside forces may
managed to construct biological life give us a clue, however. Perhaps one can
“from scratch.” But the differences require that an autonomous entity con-
between human artifacts and natural trol—at least to some extent and in what
biological life are becoming more may be considered a self-directed way,
and more subtle—and they are likely although without implying willfulness—
to disappear within the lifetimes of how it is affected by outside forces.
many of us.
Putting these ideas together, we suggest
• Most people will acknowledge that that a useful way to think about
the kinds of entities that the biologic- autonomy may be that an entity is
al and social sciences deal with seem autonomous to the extent that it shapes
somehow different from those of the way it is affected by outside forces.
physics and chemistry. A major part
of that difference is the apparent But this is pretty much how we have
ability of the entities in those sci- defined a dissipative structure. A dissipa-
ences to act on their own, i.e., their
autonomy. 24
See, for example, the American Herit-
age® Dictionary definition. URL as of 9/15/2005:
http://www.bartleby.com/61/86/A0538600.html.
tive structure results from the operation [NASA]), the internal design of a hur-
of an energy-driven abstract design. In ricane involves a greater than normal
other words, a dissipative structure res- pressure differential between the ocean
ults when an energy-driven abstract surface and the upper atmosphere. That
design shapes the way outside forces op- pressure differential causes moist surface
erate within a bounded region. air to rise. When the moisture-laden air
reaches the upper atmosphere, which is
Because this seems to be such a nice fit
cooler, it condenses, releasing heat. The
with our intuition of what it means for an
heat warms the air and reduces the pres-
entity to be autonomous, we will define
sure, thereby maintaining the pressure
an autonomous entity as an entity that is
differential—a marvelous design for a
implementing the abstract design of a
self-perpetuating process.
dissipative structure.25
In effect, a hurricane is a heat engine in
In other words, we define an autono-
which condensation, which replaces
mous entity as a self-perpetuating region
combustion as the source of heat, occurs
of reduced entropy that is implementing
in the upper atmosphere.27 Thus, al-
a dissipative structure’s abstract design.
though physically very large, a hurricane
By definition, autonomous entities con-
has a relatively simple design, which
sume energy and are far from equilibri-
causes it to consume energy and which
um. We suggest that most if not all of the
allows it to perpetuate itself as an area of
entities of the higher level sciences satis-
reduced entropy.
fy our definition of an autonomous en-
tity.
8.8 Natural and artificial
Note that most biological, social, and autonomous entities
economic autonomous entities are even Most of our energy consuming machines
more autonomous than our definition also qualify as autonomous entities. The
suggests. Most of these entities acquire primary difference between human pro-
energy in some “frozen” form such as duced autonomous entities and naturally
food or money26 and convert it to energy occurring ones is that the naturally oc-
according to their internal designs. Thus curring autonomous entities use at least
they do more than simply shape how some of the energy they consume to per-
“raw” energy that they encounter affects petuate themselves as entities. In con-
them. They are often able to save energy trast, human-produced autonomous entit-
and to chose in some sense when to use ies are almost always at-equilibrium en-
it. tities through which energy flows. In
other words, the nature of human-pro-
8.7 A naturally occurring autono- duced autonomous entities is that their
mous entity that is neither persistence as entities tends to be inde-
biological nor social pendent of their use of the energy that
We suggest that a hurricane qualifies as flows through them. This tends not to be
an autonomous entity. (See Figure 3.) In the case with naturally occurring
simple terms (paraphrased from autonomous entities.
25
This intuitive fit may be one reason that
the notion of a dissipative structure generated as 27
A characterization of hurricanes as “ver-
much enthusiasm as it has. tical heat engines” may be found in Wikipedia.
26
The maxim follow the money is really URL as of 9/1/2005:
advising to follow the energy. http://en.wikipedia.org/wiki/Hurricane
One of the senses of the word natural is As we suggest later, the fact that paral-
to have properties characteristic of ele- lels exist between autonomous and at-
ments found in nature. We suggest that equilibrium entities leads to the sugges-
the distinction between entities that rely tion that one might be able to integrate
on an at-equilibrium frame and those the two and see at-equilibrium entities as
that more actively construct their frame- one end of a continuum that includes
work is one of the central intuitive dif- both at-equilibrium and autonomous en-
ferences between what we call artificial tities.
and what we call natural. A hurricane
would thus be considered a naturally oc- 8.10 Autonomous entities and en-
curring autonomous entity which is ergy flows
neither biological nor social. Autonomous entities require energy
flows for survival. But the kinds of en-
As an example of a naturally occurring ergy flows available are limited. The
at-equilibrium entity that becomes most familiar (at least here on earth) is
autonomous, consider an atom that is be- the flow of energy from the sun. Plants
ing excited by a photon stream. Because exploit it. We are also familiar with arti-
of its design it captures the energy of the ficial energy flows, as in the flow of
photons, which it releases at some later electricity to a device when the switch is
time in what may be a slightly different turned on. Other than these, what other
form. This is the basis of the laser. flows of energy support autonomous en-
tities?
8.9 Autonomous entities and
phase transitions Thermal vents in the ocean are one pos-
Many autonomous entities exhibit the sibility. Yet the primary food producers
equivalent of phases—and phase trans- in thermal vents are bacteria that convert
itions. Such phases differ from phases in chemicals from the vents to more
at-equilibrium entities in that they reflect useable forms of energy.28 It is not clear
different ways in which the autonomous what role, if any, is played by the flow of
entity makes use of the energy that is thermal energy itself. It would be signi-
flowing through it. Examples include ficant if a life-form were found that used
gaits (walking, running, etc.), heart beats thermal energy directly to power an in-
(regular and fibrillation), and possibly ternal process in a way that paralleled
psychological conditions such as mania, the way plants use energy from the sun.
depression and psychosis. It may be that some of the chemical re-
actions that occur in inhabitants of vent
The primary concern about global warm-
ecologies depend on a high ambient tem-
ing is not that the temperature will rise
perature. But that seems to be a different
by a degree or two—although the melt-
sort of dependency than using a direct
ing of the ice caps resulting from that is
energy flow.
potentially destructive—but the possibil-
ity that if the temperature warms suffi- Most biological autonomous entities ac-
ciently, a phase transition will occur, and quire their energy in a packaged form,
the global climate structure, including at- e.g., as “food” of some sort rather than
mospheric and oceanic currents, will as a direct energy flow. Once the energy
change abruptly—and possibly dis- resource has been ingested, energy is ex-
astrously. 28
See, for example, Comm Tech Lab and
University of Delaware.
tracted from it. This is even the case So our answer to the question of whether
with our hurricane example. The energy “the same ship” is in the harbor from
of condensation is produced within the year to year is “No” if we are thinking
hurricane after warm moist air is “inges- about the material ship and “Yes” if we
ted.” are thinking about the larger ship-plus-
maintenance entity.
This seems to be another distinction
between naturally occurring and artifi- By our definition, the larger ship-plus-
cial autonomous entities. No artificial maintenance entity would be considered
entities procure their own energy re- natural rather than artificial because it as
sources. Other than plants, all naturally a social process and is not at-equilib-
occurring autonomous entities do. rium; it uses some of the energy it con-
sumes to perpetuate itself. We would
8.11 Theseus’s ship consider most social entities to be natur-
The distinction between natural and arti- al in this sense even though they are con-
ficial entities sheds some light on the structed and maintained by people.
paradox of Theseus’s ship, a ship that
was maintained (repaired, repainted, 8.12 Autonomous entities may act
etc.) in a harbor for so long that all of its in the world
original material had been replaced. As we know, hurricanes can cause signi-
Does one say that it is “the same ship” ficant damage. So far we haven’t talked
from year to year? about how that might happen.
We would like to distinguish between Since energy flows through autonomous
two ways of looking at Theseus’s ship. entities, part of that flowing through in-
One way is to consider the material ship volves flowing out. In other words,
as it exists at any one moment. By our autonomous entities may include as part
definition, this is an entity—although it of their designs means for projecting
is not an autonomous entity—since it is force into the world by directing outward
at an energy equilibrium. It is held to- flows of energy.29
gether by a large number of relatively Furthermore, the internal design of most
shallow energy wells. Entities of this autonomous entities enable them (a) to
sort are particularly vulnerable to every- store energy, (b) to move it about intern-
day weathering and wear and tear. It ally, and (c) to tap it as needed.
doesn’t take much to push some of the
energy wells beyond their limits. 8.13 Autonomous entities tend
A second way to look at Theseus’s ship not to supervene over their
is to include the maintenance process as static components
part of a larger autonomous ship entity. As we said earlier, an at-equilibrium en-
The ship along with its maintenance pro- tity consists of a fixed collection of com-
cess is an entity because it is a self-per-
petuating region of reduced entropy. It is 29
This solves a problem that concerned
Leibniz with respect to monads: how do they inter-
a relatively simple example of a social act. Leibniz’s answer was that they don’t. Our
autonomous entity. Both materials and autonomous entities interact with each other and
people cycle through it, but the process with the rest of the world though energy flows
over which they have the ability to exert some
perpetuates itself by using energy from control. Of course our autonomous entities can ex-
the society in which it is embedded. ert that control because they have internal designs;
Leibniz’s monads didn’t.
other than the collection of all mater an agent is a software object that has an
with which they may come in contact internal thread.30
during their lifetimes. Given these definitions of object and
It appears that the concept of superven- agent, we suggest that to a first very
ience may not be as useful as one might rough approximation31 objects are the
have hoped for thinking about epiphen- software equivalent of at-equilibrium en-
omena and emergence—at least in the tities and agents are the software equi-
case of autonomous entities. valent of autonomous entities.
• The instruction execution cycle itself generally leading to what is called a cli-
is epiphenomenal over the flow of max or (less dramatically) a steady state.
electrons through gates—which Our notion is closer to that of bricolage,
knows no more about the instruction a notion that originated with the structur-
execution cycle than the instruction alism movement of the early 20th century
execution cycle knows about al- [Wiener] and which is now used in both
gorithms. biology and the social sciences. Bricol-
In all of the preceding examples it is rel- age means the act or result of tinkering,
atively easy to identify the agent(s), the improvising, or building something out
environment, and the resulting epiphen- of what is at hand.
omena. In genetics bricolage refers to the evolu-
tionary process as one that tinkers with
9.2 Design and evolution an existing genome to produce
It is not surprising that designs appear in something new. [Church].
nature. It is almost tautologous to say
that those things whose designs work in John Seely Brown, former chief scientist
the environments in which they find for the Xerox Corporation and former
themselves will persist in those environ- director of the Xerox Palo Alto Research
ments. This is a simpler (and more ac-
curate) way of saying that it is the fit— Center captured its sense in a recent talk.
entities with designs that fit their envir- [W]ith bricolage you appropri-
onment—that survive. ate something. That means
you bring it into your space,
9.3 The accretion of complexity you tinker with it, and you re-
An entity that suits its environment per- purpose it and reposition it.
sists in that environment. But anything When you repurpose
that persists in an environment by that something, it is yours.34
very fact changes that environment for
everything else. This phenomenon is Ciborra [Ciborra] uses bricolage to char-
commonly referred to as an ever chan- acterize the way that organizations tailor
ging fitness landscape. their information systems to their chan-
ging needs through continual tinkering.
What has been less widely noted in the
complexity literature is that when This notion of building one thing upon
something is added to an environment it another applies to our framework in that
may enable something else to be added anything that persists in an environment
latter—something that could not have changes that environment for everything
existed in that environment prior to the else. The Internet provides many inter-
earlier addition. esting illustrations.
This is an extension of notions from eco-
logy, biology, and the social sciences. A 34
In passing, Brown claims that this is how
term for this phenomenon from the eco- most new technology develops.
logy literature, is succession. (See, for [T]hat is the way we build almost all
example, [Trani].) Historically succes- technology today, even though my law-
yers don't want to hear about it. We
sion has been taken to refer to a fairly ri- borrow things; we tinker with them; we
gid sequence of communities of species, modify them; we join them; we build
stuff.
• Because the Internet exists at all, ac- who buy and sell these assets for
cess to a very large pool of people is their own accounts even though they
available. This enabled the develop- do not intend to play the game. (Per-
ment of websites such as eBay. haps they already exist.) Someone
may develop a service that tracks the
• The establishment of eBay as a per- prices of these assets. Perhaps fu-
sistent feature of the Internet envir- tures and options markets will devel-
onment enabled the development of op along with the inevitable invest-
enterprises whose only sales outlet ment advisors.
was eBay. These are enterprises with
neither brick and mortar nor web The point is that once something fits
storefronts. The only place they sell well enough into its environment to per-
is on eBay. This is a nice example of sist it adds itself to the environment for
ecological succession. everything else. This creates additional
possibilities and a world with ever in-
• At the same time—and again be- creasing complexity.
cause the Internet provides access to
a very large number of people—oth- In each of the examples mentioned
er organizations were able to estab- above, one can identify what we have
lish what are known as massively been calling an autonomous entity. In
multi-player online games. Each of most cases, these entities are self-per-
these games is a simulated world in petuating in that the amount of money
which participants interact with the they extract from the environment (by
game environment and with each selling either products, services, or ad-
other. In most of these games, parti- vertising) is more than enough to pay for
cipants seek to acquire virtual game the resources needed to keep it in exist-
resources, such as magic swords. Of- ence.
ten it takes a fair amount of time, ef- In other cases, some Internet entities run
fort, and skill to acquire such re- on time and effort contributed by volun-
sources. teers. But the effect is the same. As long
as an entity is self-perpetuating, it be-
• The existence of all of these factors
comes part of the environment and can
resulted, though a creative leap, in an
serve as the basis for the development of
eBay market in which players sold
additional entities.
virtual game assets for real money.
This market has become so large that 9.4 Increasing complexity in-
there are now websites dedicated ex- creasing efficiency, and his-
clusively to trading in virtual game torical contingency
assets. [Wallace] The phenomenon whereby new entities
• BBC News reported [BBC] that are built on top of existing entities is
there are companies that hire low- now so widespread and commonplace
wage Mexican and Chinese teen- that it may seem gratuitous even to com-
agers to earn virtual assets, which are ment on it. But it is an important phe-
then sold in these markets. How long nomenon, and one that has not received
will it be before a full-fledged eco- the attention it deserves.
nomy develops around these assets? Easy though this phenomenon is to un-
There may be brokers and retailers derstand once one sees it, it is not trivial.
After all, the second law of thermody- another—to which we add, and nature is
namics tells us that overall entropy in- a bricolage. We repeat the observation
creases and complexity diminishes. Yet Anderson made more than three decades
we see complexity, both natural and man ago.
made, continually increasing. For the The ability to reduce
most part, this increasing complexity
everything to simple funda-
consists of the development of new
mental laws [does not imply]
autonomous entities, entities that imple-
the ability to start from those
ment the abstract designs of dissipative
laws and reconstruct the uni-
structures.
verse.
This does not contradict the Second
Law. Each autonomous entity maintains 10 Entities, emergence, and
its own internally reduced entropy by us- science
ing energy imported from the environ-
ment to export entropy to the environ- 10.1 Entities and the sciences
ment. Overall entropy increases. Such a One reason that the sciences at levels
process works only in an environment higher than physics and chemistry seem
that itself receives energy from outside somehow softer than physics and chem-
itself. Within such an environment, com- istry is that they work with autonomous
plexity increases. entities, entities that for the most part do
Progress in science and technology and not supervene over any conveniently
the bountifulness of the marketplace all compact collection of matter. Entities in
exemplify this pattern of increasing physics and chemistry are satisfyingly
complexity. One might refer to this kind solid—or at least they seemed to be be-
of pattern as a meta-epiphenomenon fore quantum theory. In contrast, the en-
since it is an epiphenomenon of the pro- tities of the higher level sciences are not
cess that creates epiphenomena. defined in terms of material boundaries.
These entities don’t exist as stable
This creative process also tends to exhib- clumps of matter; it’s hard to hold them
it a second meta-epiphenomenon. Over- completely in one’s hand—or in the grip
all energy utilization becomes continu- of an instrument.
ally more efficient. As new autonomous
entities find ways to use previously un- The entities of the special sciences are
used or under-used energy flows (or objectively real—there is some objective
forms of energy flows that had not exis- measure (their reduced entropy relative
ted until some newly created autonom- to their environment) by which they
ous entity generated them, perhaps as a qualify as entities. But as we saw earlier,
waste product), more of the energy the processes through which these entit-
available to the system as a whole is put ies interact and by means of which they
to use. perpetuate themselves are epiphenomen-
al. Even though the activities of higher
The process whereby new autonomous level entities may be described in terms
entities come into existence and perpetu- that are independent of the forces that
ate themselves is non-reductive. It is cre- produce them (recall that this is our
ative, contingent, and almost entirely a definition of epiphenomenal), the funda-
sequence of historical accidents. As they mental forces of physics are the only
say, history is just one damn thing after
forces in nature. There is no strong Let’s imagine that no hurricane ever ex-
emergence. All other force-like effects isted—at least not anywhere that an
are epiphenomenal. earthbound scientist could observe it.
Under those circumstances no scientist
Consequently we find ourselves in the
would hypothesize the possibility of
position of claiming that the higher level
such a weather system. Doing so just is-
sciences study epiphenomenal interac-
n’t part of the scientific agenda; it is not
tions among real if often somewhat eth-
the kind of task that scientists set for
ereal entities.
themselves.
10.2 Science and emergence Why waste one’s time thinking about
The idea that one can use heat and the something so strange—a weather system
expansion of gases that it produces to that not only contains its own built-in
implement a particular function is not a power plant but one in which the heat is
concept of fundamental physics. Of generated by condensation rather than
course the Carnot engine is a conse- combustion and the “furnace” in which
quence of fundamental physics, but it is the heat is generated is located in the up-
not a concept of fundamental physics. per atmosphere. Thinking through such a
The idea of using a force to implement possibility might make interesting sci-
new functionality is simply not within ence fiction.
the realm of fundamental physics.
In a galaxy far away, on a
Physics, like most science, does not con- planet of a medium size star
sider new functionality. It examines ex- near the edge of that galaxy, a
isting phenomena, and it asks how they planet that had storms with
are brought about. It does not ask how their own built-in heat engines,
knowledge gained from such an analysis ….
can be used to implement something
new. Certainly nothing so bizarre could ever
occur naturally. It would not be consid-
Here is a representative definition of the ered science.
term science.
Imagine also how bizarre phase transi-
• The observation, identifica- tions would seem if they weren’t so
tion, description, experi- common—matter sometimes obeying
mental investigation, and one set of rules and sometimes obeying
theoretical explanation of another set. It wouldn’t make any sense.
phenomena. [American Her- What would happen at the boundaries?
itage] How would transitions occur? If phase
Science is thus the study of nature, how transitions didn’t happen naturally, sci-
it is designed, i.e., organized, and how ence almost certainly wouldn’t invent
its designs work. Science does not have them.
as part of its charter to take what is If we conceive of science as the study of
known about nature and to create existing phenomena, science is reduc-
something new. tionism. To paraphrase Weinberg, the
Recall our discussion of hurricanes. Ap- goal of science is
parently they are the only kind of weath-
er system with an internal power plant.
to find simple universal laws The obvious place to look for sciences
that explain why nature is the building upon other sciences is the hier-
way it is. archy of the sciences. To take the most
concrete case, chemistry is built on the
When science is understood in this way,
abstraction of the atom as an entity with
mathematics, computer science, and en-
an internal structure. Yet we certainly
gineering, all of which create and study
don’t wonder about whether chemistry is
conceptual structures that need not exist,
a real science.
are not science. Indeed scientists and
mathematicians are often surprised when Molecules are the (emergent) entities of
they find that a mathematical construct chemistry. They form when combina-
that had been studied simply because it tions of atoms in are in a lower energy
seemed mathematically interesting has a state than the atoms would be in isola-
scientific application. tion. How does nature implement this?
It does it in terms of abstract structures
Fortunately for us, nature is not a scien-
known as orbitals.
tist. Like computer scientists and engi-
neers, she too creates things that need Molecules form when orbitals from pairs
not exist—people and hurricanes, for ex- of atoms merge. What is an orbital? It is
ample. part of the abstract design—the design
that determines how electrons and pro-
What about this paper? We would cate-
tons relate to each other—that matter im-
gorize this paper as science because one
plements by following the rules of
of its goals is to help explain, i.e., to
quantum mechanics. Thus molecular
provide some intellectual leverage for
bonds are implemented by nature
understanding, why the nature is the way
through the quantum mechanical mech-
it is.
anisms of orbitals. Like the formation of
This immediately raises another ques- atoms themselves, chemical bonding is
tion: if this is science, are we happy with part of the free lunch that nature sets out
it? Let’s assume that the simplest and for us—and another illustration that
most universal way to understand nature emergence is a fundamental aspect of
is in terms of multilevel abstractions. Is nature.
this satisfactory? Is this approach to sci-
Of course, this is just one example. As
entific explanation as real and as con-
we shall saw above, the complexity that
crete as explaining nature in terms of
we see around us is a direct result of the
more absolute single-level laws? Isn’t
fact that new abstractions may be built
there something unreal about explaining
on top of the functionalities provided by
nature at least in part as implementations
existing abstractions.
of abstractions?
One way to argue for the reality of these 11 Varieties of Emergence
abstractions is to show that they build In this section we stand back and review
upon each other. When a new abstraction the kinds of emergence we have dis-
is implemented in terms of the function- cussed. In particular we discuss two cat-
alities embodied in existing abstractions, egories of emergence: static emergence
there seems little choice but to acknow- and dynamic emergence. We also sug-
ledge the reality of the implementing ab- gest that these categories correspond to
stractions.
Weinberg’s notion of petty and grand re- make sense to talk about the
ductionism. hardness … of individual ‘ele-
As in the case of Weinberg’s petty and mentary’ particles.
grand reductionism, static emergence, Since statically emergent phenomena
while of great importance, is of lesser in- must be implemented in terms of some
terest. It is dynamic emergence, and es- underlying model, and since time is by
pecially stigmergic dynamic emergence definition excluded from that imple-
that is central to complex systems. mentation, static emergence is equival-
Recall that we defined a phenomenon as ent to Weinberg’s petty reductionism.
emergent over a underlying model if
11.2 Dynamic emergence
(a) it has an independent conceptualiza-
Properties or phenomena of a model are
tion and (b) it can be implemented in
dynamically emergent if they are defined
terms of elements of that model.
in terms of how the model changes (or
11.1 Static emergence doesn’t change) over some time. Dy-
An emergent phenomenon is statically namic emergence occurs either with or
emergent if its implementation does not without autonomous entities. We call the
depend on time. former stigmergic emergence, but we
look at non-stigmergic dynamic emer-
As an interesting example of static emer- gence first.
gence, consider cloth as a collection of
threads woven together. Cloth has the 11.3 Non-stigmergic dynamic
emergent property that it is able to cover emergence
a surface. This property is implicitly two Interactions among at-equilibrium entit-
dimensional. The components of cloth, ies result in non-stigmergic dynamic
i.e., threads, do not have (or at least are emergence. Two examples are: (a) ob-
not understood in terms of) that property. jects moving in space and interacting ac-
A thread is understood in terms of the cording to Newtonian mechanics and
property length. Yet when threads are (b) the quantum wave function.
woven together the resulting cloth has
What appears to be distinctive about
this new property, which effectively con-
such systems is that they are not charac-
verts a collection of one dimensional
terized in terms of discrete states that
components to a two dimensional object.
their elements assume. Elements do not
Many human manufactured or construc- transition from one state to another. Such
ted artifacts exhibit static emergence. A systems may be defined in terms of con-
house has the statically emergent prop- tinuous equations.
erty number-of-bedrooms. More gener-
The quantum wave function is an espe-
ally, a house has the emergent property
cially interesting example. As long as it
that it can serve as a residence. Static
does not undergo decoherence, i.e., in-
emergence also occurs in nature. As
teraction with an environment, the wave
Weinberg points out,
function encompasses all possibilities,
[A] diamond is hard because but it realizes none of them.
the carbon atoms of which it is
Since quantum states are discrete (hence
composed can fit together
the term quantum), objects cannot trans-
neatly [even though] it doesn't ition smoothly from one quantum state
to another. So how does that transition model (a) whose elements include air,
occur? Quantum theory turns these water vapor, and radiation and
transitions into probabilities. As Hardy (b) whose elements interact according
points out [Hardy], by making such to the principles of physics. When that
transitions probabilistically continuous, model is run it will generate the emer-
gent phenomena that we would recog-
quantum theory offers us a
nize as cold fronts and thunderstorms.
way to have the advantages of
discreteness and continuity at Grand reductionism is thus the explana-
the same time. tion of phenomena at one level in terms
of phenomena at a more fundamental
For all practical purposes, actually to as-
level. One shows that when the laws at
sume a state requires something more,
the more fundamental level are applied,
the so-called collapse of the wave func-
the result will be the phenomena of in-
tion. That happens stigmergically. At the
terest at the less fundamental level.
quantum level, stigmergy is equivalent
Since these sorts of models are inevit-
to decoherence.
ably dynamic this is dynamic emergence
11.4 Dynamic emergence and but expressed in other terms.
grand reductionism Nor should the equating of grand reduc-
As static emergence corresponds to tionism with dynamic emergence sur-
Weinberg’s petty reductionism, dynamic prise anyone in the field of complex sys-
emergence seems to correspond nicely to tems. After all, the presumed reason to
Weinberg’s grand reductionism. Wein- build a model is to show that a set of
berg explains grand reductionism as fol- lower level rules will produce higher
lows level results—which is exactly the grand
[T]he reductionist regards the reductionist agenda.
general theories governing air Of course the terminology that we would
and water and radiation as be- use is not just that lower level rules pro-
ing at a deeper level than the- duce higher level results, that lower level
ories about cold fronts or thun- rules may implement a higher level ab-
derstorms, not in the sense straction.
that they are more useful, but But no matter how it is expressed,
only in the sense that the lat- without the lower level substrate, the
ter can in principle be under- higher level phenomena would not exist.
stood as mathematical con-
sequences of the former. The 11.5 Stigmergic emergence
reductionist program of phys- Stigmergic emergence is dynamic emer-
ics is the search for the com- gence that involves autonomous entities.
mon source of all explanations. What tends to be most interesting about
… autonomous entities are (a) they may as-
We hope that Weinberg would not object sume discrete states and (b) they change
to the following paraphrase. state as they interact with their environ-
ments.35
The reductionist goal (with respect to
reducing weather terminology to the 35
One might liken an isolated quantum
terminology of physics) is to build a wave system to the inside of an autonomous entity.
It assumes a state (i.e., collapses) when it interacts
It had once been hoped that software de- capabilities. Agriculture and animal hus-
velopment could evolve to a point at bandry use both plant reproduction and
which one need only write down a form- such animal capabilities as locomotion
al specification of what one wanted the or material (i.e., skin) production for our
software to do. Then some automatic own purposes. The exploitation of exist-
process would produce software that sat- ing capabilities for our own purposes is
isfied that specification. not a new idea.
That dream now seems quite remote. Be- An interesting example of this approach
sides the difficulty of developing (a) a to engineering involves recent develop-
satisfactory specification language and ments in robotics. Collins reported
(b) a system that can translate specifica- [Collins] that a good way to make a ro-
tions written in such a language into ex- bot walk is by exploiting gravity through
ecutable code, the real problem is that it what he called passive-dynamic motion
has turned out to be at least as difficult —raise the robot’s leg and let gravity
and complex to write formal specifica- pull it back down—rather than by direct-
tions as it is to write the code that pro- ing the robot’s limbs to follow a pre-
duces the specified results. defined trajectory.
Even if one could write software by This illustrates in a very concrete way
writing specifications, in many cases— the use of an existing force in a design.
especially cases that involve large and Instead of building a robot whose every
complex systems, the kinds of cases for motion was explicitly programmed,
which it really matters—doing so does- Collins built a robot whose motions were
n’t seem to result in much intellectual controlled in part by gravity, a pre-exist-
leverage, if indeed it produces any at all. ing force.
This illustrates quite nicely that we often
12.3 Infrastructure-centric devel-
find ourselves in the position of wanting
opment
to produce epiphenomena (epiphenom- Building new capabilities on top of ex-
ena, which may be very important to us), isting ones is not only good design, it is
whose formalization as an abstraction highly leveraged design. But now that
we find to be either infeasible or not par- we are aware of this strategy a further
ticularly useful. lesson can be drawn. New systems
should be explicitly designed to serve as
12.2 Bricolage as design
a possible basis for systems yet to come.
The process of building one capability
Another way of putting this is that every
on top of another not only drives the
time we build a new system, it should be
overall increase in complexity, it also
built so that it becomes part of our envir-
provides guidance to designers about
onment, i.e., our infrastructure, and not
how to do good design work. Any good
just a piece of closed and isolated func-
designer—a developer, an architect, a
tionality.
programmer, or an engineer—knows that
it is often best if one can take advantage By infrastructure we mean systems such
of forces and processes already in exist- as the Internet, the telephone system, the
ence as part of one’s design. electric power distribution system, etc.
Each of these systems can be character-
But even before engineering, we as hu-
ized in isolation in terms of the particu-
man beings made use of pre-existing
lar functions they perform. But more im- 12.4 Service refactoring and the
portant than the functional characteriza- age of services
tion of any of these individual systems is The process whereby infrastructure ser-
the fact that they exist in the environ- vices build on other infrastructure ser-
ment in such a way that other systems vices leads not only to new services, it
can use them as services. also leads to service refactoring. The
corporate trend toward outsourcing func-
We should apply this perspective to all
tions that are not considered part of the
new systems that we design: design them
core competence of the corporation illus-
as infrastructure services and not just as
trates this. Payroll processing is a typical
bits of functionality. Clearly Microsoft
example.
understands this. Not only does it posi-
tion the systems it sells as infrastructure Because many organizations have em-
services, it also maintains tight owner- ployees who must be paid, these organ-
ship and control over them. When such izations must provide a payroll service
systems become widely used elements of for themselves. It has now become feas-
the economy, the company makes a lot ible to factor out that service and offer it
of money. The tight control it maintains as part of our economic infrastructure.
and the selfishness with which it controls This outsourcing of internal processes
these systems earns it lots of resentment leads to economic efficiencies in that
as well. Society can’t prosper when any many such processes can be done more
important element of its infrastructure is efficiently when performed by special-
controlled primarily for selfish purposes. ized organizations. Such specialized or-
The US Department of Defense (DoD) is ganizations can take advantage of eco-
currently reinventing itself [Dick] to be nomies of scale. They can also serve as
more infrastructure-centric. This requires focal points where expertise in their spe-
it to transform what is now a cialized service can be concentrated and
the means of providing those services
huge collection of independent
improved.
“stovepipe” information sys-
tems, each supporting only its As this process establishes itself ever
original procurement specifica- more firmly, more and more organiza-
tion, to a unified assembly of tions will focus more on offering ser-
interoperating systems. vices rather than functions, and organiz-
ations will become less stovepiped.
The evocative term stovepipe is intended
to distinguish the existing situation—in We frequently speak of the “service in-
which the DoD finds that it has acquired dustries.” For the most part this term has
and deployed a large number of func- been used to refer to low level services
tionally isolated systems (the —although even the fast food industry
“stovepipes”)—from the more desirable can be seen as the “outsourcing” of the
situation in which all DoD systems are personal food preparation function. With
available to each other as an infrastruc- our more general notion of service in
ture of services. mind, historians may look back to this
period as the beginning of the age of ser-
vices.
races we are talking about not just popu- ran its own internal genetic program-
lation sizes but actual evolutionary ming model. But we are unaware of any
changes. such work.37
Imagine a situation in which a plant spe- Finally, consider the fact that geckos
cies comes under attack from an insect climb walls by taking advantage of the
species. In natural evolution the plant Van der Walls “force.” (We put force in
may “figure out” how to grow bark. Can quotation marks because there is no Van
we build a computer model in which this der Walls force. It is an epiphenomenon
solution would emerge? It is very un- of relatively rarely occurring quantum
likely. To do so would require that our phenomena.) To build a model of evolu-
model have built into it enough informa- tion in which creatures evolve to use the
tion about plant biochemistry to enable it Van der Walls force to climb walls
to find a way to modify that biochem- would require that we build quantum
istry to produce bark, which itself is physics into what is presumably inten-
defined implicitly in terms of a surface ded to be a relatively high-level biolo-
that the insect cannot penetrate. gical model in which macro geckos
Evolving bark would require an enorm- climb macro walls
ous amount of information—especially It’s worth noting that the use of the Van
if we don’t want to prejudice the solu- der Walls force was apparently not an
tion the plant comes up with. extension of some other gecko process.
The next step, of course, is for the insect Yet the gecko somehow found a way to
to figure out how to bore through bark. reach directly down to a quantum-level
Can our model come up with something effect to find a way to climb walls.
like that? Unlikely. What about the The moral is that any base level that we
plant’s next step: “figuring out” how to select for our models will be arbitrary,
produce a compound that is toxic to the and by choosing that base level, we may
insect? That requires that the model in- miss important possibilities. Another
clude information about both plant and moral is that models used when doing
insect biochemistry—and how the plant computer security or terrorism analysis
can produce a compound that interferes —or virtually anything else that includes
with the insect’s internal processes. This
the possibility of creative adaptation—
would be followed by the development
37
by the insect of an anti-toxin defense. Genetic programming is relevant be-
cause we are assuming that the agent has an arbit-
To simulate this sort of evolutionary pro- rarily detailed description of how the it functions
and how elements in its environment function.
cess would require an enormous amount
Notice how difficult it would be implement such a
of low level detail—again especially if system. The agent’s internal model of the environ-
we don’t want to prejudice the solution ment would have to be updated continually as the
in advance. environment changed. That requires a means to
perceive the environment and to model changes in
Other than Tierra (see [Ray]) and its suc- it. Clearly that’s extraordinarily sophisticated. Al-
cessors, which seem to lack the richness though one could describe such a system without
recourse to the word consciousness, the term does
to get very far off the ground, as far as come to mind.
we know, there are no good computer Nature’s approach is much simpler: change during
models of biological arms races. A seem- reproduction and see what happens. If the result is
ingly promising approach would be an unsuccessful, it dies out; if it is successful it per-
sists and reproduces. Of course that requires an en-
agent-based system in which each agent tire generation for each new idea.
will always be incomplete. We will only The moral is the same as before. Models
be able to model effects on the levels for will always be incomplete. We will only
which our models are defined. The ima- be able to model effects on the levels for
ginations of any agents that we model which our models are defined. The ima-
will be limited to the capabilities built ginations of any agents that we model
into the model. will be limited to the capabilities built
into the model.
12.7 Modeling: the difficulty of
looking upward 13 Observations
We noted earlier that when a glider ap- Our fundamental existence depends on
pears in the Game of Life, it has no ef- taking energy and other resources from
fect on the how the system behaves. The the environment. We must all do it to
agents don’t see a glider coming and stay in existence. Raises fundamental
duck. More significantly we don’t know ethical questions: how can taking be
how to build systems so that agents will condemned? Supports stewardship no-
be able to notice gliders and duck. tions since we are all dependent on en-
It would be an extraordinary achieve- vironment.
ment in artificial intelligence to build a Dynamic entities consist are composed
modeling system that could notice emer- of static and dynamic entities (bodies
gent phenomena and see how they could and societies). That’s what makes them
be exploited. Yet we as human beings do solid. But those static entity components
this all the time. The dynamism of a are frequently replaced.
free-market economy depends on our
ability to notice newly emergent patterns Competition for energy and other re-
and to find ways to exploit them. sources justifies picture of evolution as
survival of the meanest. Also justifies
Al Qaeda noticed that our commercial group selection since groups can ensure
airlines system can be seen as a network access to resources better than individu-
of flying bombs. Yet no model of terror- als.
ism that doesn’t have something like that
built into it will be able to make that sort 14 Concluding remarks
of creative leap. Our models are blind to
emergence even as it occurs within 14.1 Computer Science and Philo-
them. sophy
Notice that this is not the same as the • For centuries after Newton, nature
difficulty of looking downward. In the was seen as a perfect clockwork
Al Qaeda example one may assume that mechanism.
one’s model of the airline system in- • After the invention of the telephone,
cludes the information that an airplane the brain was likened to a switch-
when loaded with fuel will explode board.
when it crashes. The creative leap is to
notice that one can use that phenomenon Science and technology tends to shape
for new purposes. This is easier than the our view of the world. It’s not surprising
problem of looking downward. But it is that the perspectives developed in this
still a very difficult problem. article reflect those of Computer Sci-
ence, the predominant perspective of this
age. Is this parochialism? It’s difficult to
tell from so close. One thing is clear. Be- programs execute without even being
cause Computer Science has wrestled— observed by people—programs in cars
with some success—with many serious and other machinery, for example. And
intellectual challenges, it is not unreas- even programs that perform strictly sym-
onable to hope that the field may con- bolic transformations, do work that may
tribute something to the broader intellec- be understood as constraining the forces
tual community. of nature (the motion of electrons) to
some formal shape.
It is useful to think of computers as re-
ification machines: they make the ab- To make the computer a useful reifica-
stract concrete. As such they are similar tion device—to make it possible for pro-
to the drawing tools of an animator— grammers to write text that causes a
who draws anything that crosses his or computer to convert programmers’
her mind, no matter how wild or fanci- fantasies to some concrete form—Com-
ful. In the hands of a skilled and creative puter Science has had to deal with some
programmer,38 a computer and an associ- of philosophy’s thorniest issues.
ated software development environment
• Computer Science has created lan-
invite virtually unlimited creativity.
guages that are both formally defined
But there is a major difference between —with formal syntax and semantics
the product produced by an animator and —and operational, i.e., they actually
that produced by a programmer. An an- function in the real world.
imator produces images that have mean-
ing only when they enter the mind of an- • Computer Science has figured out
other human being. They do not stand on how to represent information in data-
their own. They are meant strictly as bases in ways that allow that inform-
communication from one human being ation to hang together meaningfully.
to another. Nor are either the physical • Computer Science has faced—and to
medium on which the images are stored a significant extent resolved—the
or the mechanisms that causes the im- problem of working on many levels
ages to be displayed of much signific- of abstraction and complexity simul-
ance. The only thing that matters is that taneously.
the images arrive as intended in the
minds of the viewers. If insights gained from these and other
intellectual wrestling matches can be ap-
A computer program is stuck in the real plied in a wider context, it is only Com-
world. It’s work is to shape the activity puter Science paying back the debt that
of a computer, a physical device. Many it owes to the engineers, scientists, math-
ematicians, and philosophers who set the
38
We use the term programmer deliber- stage for and participated in its develop-
ately. Fancier terms like software developer (or
software engineer or worse information techno-
ment.
logy specialist) lose the focus of what is really
happening when one creates software. A program- 14.2 Constructive science
mer (writes text that) programs, i.e., shapes, the For most of its history, science has pur-
way a computer behaves. That’s really all that mat-
ters: what does the text tell the computer to do? sued the goal of explaining existing phe-
In the preceding, note the stigmergy and down- nomena in terms of simpler phenomena.
ward entailment. A program doesn’t tell a com- That’s the reductionist agenda.
puter anything. Successful programmers work
from that perspective all the time.
The approach we have taken is to ask Science has produced some specialized
how new phenomena may be construc- answers to this question. The biological
ted from and implemented in terms of evolutionary explanation involves ran-
existing phenomena. That’s the creative dom mutation and crossover of design
impulse of artists, computer scientists, records. The cosmological explanation
engineers—and of nature. It is these new involves falling into energy wells of
phenomena that are often thought of as various sorts. Is there any more to say
emergent. about how nature finds and then explores
new possibilities? If as Dennett argues in
When thinking in the constructive direc-
[Dennett ‘96] this process may be fully
tion, a question arises that is often under-
explicated as generalized Darwinian
appreciated: what allows one to put ex-
evolution, questions still remain. Is there
isting things together to get something
any useful way to characterize the search
new—and something new that will per-
space that nature is exploring? What
sist in the world? What binding forces
search strategies does nature use to ex-
and binding strategies do we (and
plore that space? Clearly one strategy is
nature) have at our disposal?
human inventiveness.
Our answer has been that there are two
sorts of binding strategies: energy wells 15 Acknowledgement
and energy-consuming processes. En- We are grateful for numerous enjoyable
ergy wells are reasonably well under- and insightful discussions with Debora
stood—although it is astonishing how Shuger during which many of the ideas
many different epiphenomena nature and in this paper were developed and re-
technology have produced through the fined.
use of energy wells.
We also wish to acknowledge the fol-
We have not even begun to catalog the lowing websites and services, which we
ways in which energy-consuming pro- used repeatedly.
cesses may be used to construct stable,
self-perpetuating, autonomous entities. • Google (www.google.com);
• The Stanford Encyclopedia of Philo-
Earlier we wrote that science does not sophy (plato.stanford.edu);
consider it within its realm to ask con-
• OneLook Dictionary Search
structivist questions. That is not com-
pletely true. Science asks about how we (onelook.com).
got here from the big bang, and science
References
asks about biological evolution. These
Abbott, R., “Emergence, Entities, En-
are both constructivist questions. Since
tropy, and Binding Forces,” The Agent
science is an attempt to understand
2004 Conference on: Social Dynamics:
nature, and since constructive processes
Interaction, Reflexivity, and Emergence,
occur in nature, it is quite consistent with
Argonne National Labs and University
the overall goals of science to ask how
of Chicago, October 2004. URL as of
these constructive processes work. As far
4/2005: http://abbott.calstatela.edu/Pa-
as we can determine, there is no sub-dis-
persAndTalks/abbott_agent_2004.pdf.
cipline of science that asks, in general,
how the new arise from the existing. American Heritage, The American Her-
itage® Dictionary of the English Lan-
guage, 2000. URL as of 9/7/2005:
http://www.bartleby.com/61/67/S014670 as of 6/2005:
0.html. http://www.nature.com/msb/journal/v1/n
1/full/msb4100007.html.
Anderson, P.W., “More is Different,”
Science, 177 393-396, 1972. Ciborra, C. "From Thinking to
Tinkering: The Grassroots of Strategic
BBC News, “Gamer buys $26,500 virtu-
Information Systems", The Information
al land,” BBC News, Dec. 17, 2004.
Society 8, 297-309, 1992.
URL as of 2/2005: http://news.b-
bc.co.uk/1/hi/technology/4104731.stm. Clarke, A. C., "Hazards of Prophecy: The
Failure of Imagination,” Profiles of The
Bedau, M.A., “Downward causation and
Future, Bantam Books, 1961.
the autonomy of weak emergence”.
Principia 6 (2002): 5-50. URL as of Cockshott, P. and G. Michaelson, “Are
11/2004: http://www.reed.edu/~mab/pa- There New Models of Computation:
pers/principia.pdf. Reply to Wegner and Eberbach.” URL as
of Oct. 10, 2005: http://www.dc-
Boyd, Richard, "Scientific Realism",
s.gla.ac.uk/~wpc/reports/wegner-
The Stanford Encyclopedia of Philo-
25aug.pdf.
sophy (Summer 2002 Edition), Edward
N. Zalta (ed.), URL as of 9/01/2005: Colbaugh, R. and Kristen Glass, “Low
http://plato.stanford.edu/archives/sum20 Cognition Agents: A Complex Networks
02/entries/scientific-realism/. Perspective,” 3rd Lake Arrowhead Con-
ference on Human Complex Systems,
Brown, J.S., Talk at San Diego State
2005.
University, January 18, 2005. URL as of
6/2005: http://ctl.sdsu.edu/pict/jsb_lec- Collins, Steven, Martijn Wisse, and
ture18jan05.pdf Andy Ruina, “A Three-Dimensional
Passive-Dynamic Walking Robot with
Carroll, S.B., Endless Forms Most
Two Legs and Knees,” The International
Beautiful: The New Science of Evo Devo
Journal of Robotics Research Vol. 20,
and the Making of the Animal Kingdom,
No. 7, July 2001, pp. 607-615, URL as
W. W. Norton, 2005.
of 2/2005:
Chaitin, G. Algorithmic Information http://ruina.tam.cornell.edu/research/top-
Theory, reprinted 2003. URL as of Sept. ics/locomotion_and_robotics/papers/3d_
6, 2005: passive_dynamic/3d_passive_dynam-
http://www.cs.auckland.ac.nz/CDMTCS/ ic.pdf
chaitin/cup.pdf.
Comm Tech Lab and the Center for Mi-
CFCS, Committee on the Fundamentals crobial Ecology, The Microbe Zoo, URL
of Computer Science: Challenges and as of Oct 10, 2005: http://commtech-
Opportunities, National Research Coun- lab.msu.edu/sites/dlc-
cil, Computer Science: Reflections on me/zoo/microbes/riftiasym.html
the Field, Reflections from the Field,
Comte, A. “Positive Philosophy,” trans-
2004. URL as of 9/9/2005:
lated by Harriet Martineau, NY: Calvin
http://www.nap.edu/books/0309093015/
Blanchard, 1855. URL: as of 7/2005:
html/65.html.
http://www.d.umn.edu/cla/faculty/jham-
Church, G. M., “From systems biology lin/2111/ComteSimon/Comtefpin-
to synthetic biology,” Molecular tro.html.
Systems Biology, March, 29, 2005. URL
an LCG are determined only by other More formally, if l is an LCG, let Game
cells in the same LCG.39 of Life(l) be the set of LCGs that are
formed by applying the Game of Life
Suppose that an LCG contains the only rules to the cells in l. For any particular
live cells on a Game of Life grid. Con- l, Game of Life(l) may be empty; it may
sider what the mapping of that LCG by be contain a single element; or it may
the Game of Life rules will produce. contain multiple elements. If l’ is a
There are three possibilities. member of Game of Life(l) write l -> l’.
1. The live cells may all die.
For any LCG l 0, consider a sequence of
2. The successor live cells may consist successor LCGs generated in this man-
of a single LCG—as in a glider or ner:
still life.
3. The successor live cells may parti- l0 -> l1 -> l 2 -> l3 -> … .
tion into multiple LCGs—as in the
so-called bhepto pattern, which starts Extend such a sequence until one of
as a single LCG and eventually sta- three conditions occurs.
bilizes as 4 still life LCGs and two 1. There are no successor LCGs, i.e.,
glider LCGs. Game of Life(li) is empty—all the
live cells in the final LCG die. Call
In other words, the live cells generated these terminating sequences.
when the Game of Life rules are applied
to an LCG will consist of 0, 1, or mul- 2. There is a single successor LCG, i.e.,
tiple successor LCGs. Game of Life(li) = {lk}, but that suc-
cessor LCG is in the same state as an
LCG earlier in the sequence, i.e., lk =
39
In particular, no LCG cells have live neighbors lj, j < k. Call these repeating se-
that are outside the LCG. Thus no cells outside quences.
the LCG need be considered when determining
the GoL transitions of the cells in an LCG. A 3. The set Game of Life(li) of successor
dead boundary cell may become live at the LCGs contains more than one LCG,
next time-step, but it will do so only if three of i.e., the LCG branches into two or
its neighbors within the LCG are live. Its more LCGs. Call these branching se-
neighbors outside the LCG are guaranteed to
be dead.
quences.
Note that some LCG sequences may
If a boundary cell does become live, the next- never terminate. They may simply pro-
state LCG of which it is a member will include
cells that were not part of its predecessor LCG. duce larger and larger LCGs. The so-
called spacefiller pattern, which actually
consists of multiple interacting LCGs,
one of which fills the entire grid with a
single LCG as it expands,40 is an amazing
example of such a pattern. I do not know
if there is an LCG that expands without
40
See the spacefiller pattern on http://www.math.-
com/students/wonders/life/life.html or
http://www.ibiblio.org/lifepatterns.
limit on its own. If any such exist, call far backwards.41 When run forward such
these infinite sequences. fuse or wick LCGs converge to a single
LCG. Yet given the original definition of
For any LCG l0, if the sequence
BP even these LCG sequences are in-
l0 -> l1 -> l2 -> l3 -> … . cluded in it. Each of these unbounded
predecessor chains is included in BP
is finite, terminating in one of the three starting at each predecessor LCG.
ways described above, let seq(l0) be that
sequence along with a description of Clearly BP as defined includes many re-
how it terminates. If dundant pattern descriptions. No attempt
is made to minimize BP either for sym-
l0 -> l1 -> l2 -> l3 -> … . metries or for overlapping patterns in
which one pattern is a suffix of another
is infinite, then seq(l0) is undefined. —as in the fuse patterns. In a computer
program that generated BP, such effi-
Let BP (for Basic Patterns) be the set of ciencies would be important.
finite non-branching sequences as
defined above. That is, 16.3 BP is recursively enumerable
The set BP of basic Game of Life pat-
BP = {seq(l0) | l0 is an LCG} terns may be constructed through a
formal iterative process. The technique
Note that it is not necessary to extend
employed is that used for the construc-
these sequences backwards. For any
tion of many recursively enumerable
LCG l0, one could define the pre-image
sets.
of l0 under the Game of Life rules. Game
of Life-1(l) is the set of LCGs l’ such that 1. Generate the LCGs in sequence.
Game of Life(l’) = l. 2. As each new LCG is generated, gen-
erate the next step in each of the se-
For any chain seq(l0) in BP, one could
quences starting at each of the LCGs
add all the chains constructed by prefix-
generated so far.
ing to seq(l0) each of the predecessors l’
of l0 l’ as long as l’ does not appear in 3. Whenever an LCG sequence termin-
seq(l’). But augmenting BP in this way ates according to the BP criteria, add
would add nothing to BP since by defini- it to BP.
tion seq(l’) is already defined to be in BP
for each l’. The process sketched above will effect-
ively generate all members of BP. Al-
We noted above that we do not know if though theoretically possible, such a pro-
there are unboundedly long sequences of cedure will be so inefficient that it is
LCGs beginning with a particular l0. useless for any practical purpose.42 The
With respect to unboundedly long prede-
41
cessor chains, it is known that such un- A simple fuse pattern is a diagonal configuration
of live cells. At each time step, the two end cells
bounded predecessor chains (of unboun- die; the remaining cells remain alive. A simple
dedly large LCGs) exist. The so-called fuse pattern may be augmented by adding more
fuse and wick patterns are LCG se- complex features at one end, thereby building a
pattern that becomes active when the fuse exhausts
quences that can be extended arbitrarily itself. Such a pattern can be built with an arbitrar-
ily long fuse.
42
Many much more practical and efficient
programs have been written to search for patterns
only reason to mention it here is to es- using a technique similar to that used for
tablish that BP is recursively enumer- generating BP itself, one can (very tedi-
able. Whether BP is recursive depends ously) enumerate all the possible BP in-
on whether one can in general establish teractions.
for any LCG l0 whether seq(l0) will ter- More formally, let Pf(BP) be the set of
minate.43 all finite subsets of BP. For each member
of Pf(BP) consider all possible (still
16.4 Game of Life patterns: com-
binations of basic patterns only a finite number) relative configura-
Many of the interesting Game of Life tions of its members on the grid so that
patterns arise from interactions between there will be some interaction among
and among basic patterns. For example, them at the next time step. One can then
the first pattern that generated an unlim- record all the possible interactions
ited number of live cells, the glider gun, among finite subsets of BP.
is a series of interactions among combin- These interactions would be equivalent
ations of multiple basic patterns that cyc- to the APIs for the basic patterns. We
lically generate gliders. could call a listing of them BP-API.
Since BP is itself infinite, BP-API would
To characterize these more complex pat- also be infinite. But BP-API would be
terns it is necessary to keep track of how effectively searchable. Given a set of
basic patterns interact. In particular, for elements in BP, one could retrieve all the
each element in BP, augment its descrip- interactions among those elements. BP-
tion with information describing API would then provide a documented
a) its velocity (rate, possibly zero, and starting point for using the Game of Life
direction) across the grid, as a programming language.
b) if it cycles, how it repeats, i.e., which As in traditional programming lan-
states comprise its cycle, and guages, as more complex interactions are
developed, they too could be docu-
c) if it branches, what the offspring ele- mented and made public for others to
ments are and where they appear rel- use.
ative to final position of the terminat-
ing sequence.
Two or more distinct members of BP
that at time step i are moving relative to
each other may interact to produce one
or more members of BP at time step i+1.
The result of such a BP “collision” will
generally depend on the relative posi-
tions of the interacting basic patterns.
Even though the set BP of basic patterns
is infinite, since each LCG is finite, by
in the GoL and related cellular automata. See
http://www.ics.uci.edu/~eppstein/ca/search.html
for a list of such programs.
43
This is not the same question as that
which asks whether any Game of Life configura-
tion will terminate. We know that is undecidable.