Simon Design Process

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 10

Invited Paper: X-v -v -v -v -__,.

,
Proceedings of the 2nd Annual.
Style in Design Design Research Association
Conference, 197TI
by Herbert A. Simon
Camegie-Mellon University

Abstract objects with the particular manufacturing pro-


A style is some one way of doing things, cesses that produced then. Thus, observing
chosen from a number of alternative ways. different styles of flint arrowheads, we nay
Since design problems generally do not have distinguish between those produced by percus-
unique optimal solutions, style may enter in sion snd those produced by pressure. Or we
choosing any one of many satisfactory solu- nay distinguish pottery made on the wheel
tions. Components of style can arise from from coiled pottery.
three sources: direct specification of the At one further remove from the final ob-
final object, the processes used to manufac- ject, styles may also arise from differences
ture it, and the processes used to design it. in the process of design. Buildings designed
This paper examines the dynamic aspects of primarily from the inside out, are different
style, and particularly those aspects that a- from those designed primarily from the outside
rise from the design process itself. The anal- in. Thus, components of style can arise from
ysis suggests a new interpretation of the es- at least three sources: the direct specifica-
thetic significance of functionalism. tions of the final object, the nature of the
The following passage from E. H. Comb- processes used to manufacture it, and the na-
rich's remarkable book, ART AND ILLUSION (page ture of the processes used to design it. Of
21), will terve as my thesis: course, the latter two sources--manufacturing
"The history of taste and fashion and design processes--also manifest themselves
is the history of preferences, of in features of the final object, but only in-
various acts of choice between directly as a consequence of the interplay of
given alternatives . . . [But] - dynamic forces. It is the second rurpose of
an act of choice is only of symp- this paper to examine.the dynamic aspects of
tomatic significance, is expres- style, and particularly those aspects that a-
sive of something only if we can rise from the design process itself.
really want to treat styles as
symptomatic of something else, Satisficing and the One Right Way
. . . , we cannot do without There is a proverb to the effect that the
some theory of the alternatives." best is enemy of the good. A little reflec-
A Style, in this view, is some one way of tion shows that the proverb can be read in two
doing things, chosen from a number of alterna- diametrically opposed ways, and that it is not
tive ways. If Frederick Taylor had been cor- clear which reading is intended--or is the
rect in insisting that there is "one right way more defensible. By one interpretation, the
of doing anything," there would be an inevita- proverb means that if we are willing to settle
ble clash between function and style--we could for the good enough--to satisfice--we will
exercise stylistic freedom only by sacrificing never attain the best. By the other interpre-
function. One purpose of this paper is to tation, it means that our striving to reach an
argue that we are not in fact faced with this unattainable best may prevent us from reaching
particular dilemma--that design problems gen- an achievable good-enough. It is this latter
erally do not have unique optimal solutions, interpretation that describes the stern real-
hence that we may choose any one of many ities of the design process in the world we
satiafactory solutions. know in any world, for that matter, that is
Styles are generally identified by fea- even moderately complex.
tures of the objects designed. We may char-
acterize (not very profoundly) the Gothic Since I have made the satisficing argu-
style in architecture by its pointed arches, ment at length elsewhere, I will review it
and the Romanesque style by its round arches. ( only briefly here, and then apply it specifical-
Sometimes, however, at one step removed from ly to the problem of design (1). Host problem
the objects themselves, we connect styles in solving can be represented as a tfearch through
large space of possibilities. For real-world
f rob lens, the spaces are not merely large, but the size of the space searched. In contrast,
fmnense, and there is not the slightest chance the cost of a search to find a satisfactory
for either man 'or computer to search them ex-
haustively for the solution that is absolutely ^solution depends only on the density of dis-
tribution of solutions of varying quality
Best solutions sre therefore only attain- through the space, and is more or less inde-
able in those situations where there exist pendent of the size of the total space. The
apme systematic procedures, or algorithms, for familiar needle-in-haystack problem illus-
going more or less directly to the optimum trates the point. In a haystack throughout
With pnly a moderate amount of highly selective which needles of varying sharpness 'are dis-
pearch, or, in favorable instances, with no tributed randomly, finding the sharpest needle
search at all. Thus, to find the local maxima requires a search of the entire stack. Finding
pf f real'Valued, continuous and differenti- needle sharp enough to sew with requires a
ate function of a variable defined on a closed search of a pile of hay big enough to contain
interval, simply execute the algorithm for find- one such needle--a pile whose size does not de-
ing the first derivative of the function, set pend at all on the size of the whole stack.
the first derivative equal to zero, and solve Fqr most practical purposes, the real world is
fpr (he unknown. TbU the function consists an infinite haystack, hence a place where we -
e>f a npn«-denunerable infinity of points causes are well advised to satisfice.
ttp trouble, for only a small finite set of The second point often ignored by optim-
these is ever examined, and the algorithm leads izers is that, if the general'magnitude of the
VP directly to them. available search effort is known in advance,-
In ether kinds of problem spaces, the then following an optimizing procedure until
available algorithms are not powerful enough the marginal cost of additional search exceeds
to eliminate entirely the need for search, but the expected gain is not equivalent to employ-
re sufficiently selective to reduce search to ing the best satisficing procedure. Given
tolerable proportions. An important class of resource limit, there may exist many satis-
example* are the situations that can be formu- ficing procedures that will, on average, find
lated as linear programming problems. The better solutions than a truncated optimizing
available algorithms for seeking the optimum procedure. This proposition can again, of
in a linear programming problem-- for example, course, be translated into optimizing terms,
the »imp lex algorithm--aearch only certain but as a practical matter, it has often been
boundary points of the space, are guaranteed overlooked by optimizers, who have consequent-
to climb monotonically toward the optimum, and ly Invested too little of their effort In the .
reach that optimum after a finite number of sesrch for powerful heuristic procedures.
ftept. Although it is easy to find real-life Some of us would argue that this has resulted,
problems that demand more computation in a in the past two decades, in a serious imbalance
linear programming formulation than man or com- between the amount of research devoted to im-
puter can provide, a good many problems become proving the kit of optimizing tools and the
manageable when stated in linear programming amount devoted to improving the kit of satls-
terms. ficing tools; and a consequent imbalance in the
Where optimizing algorithms are unavail- kits themselves. But pursuing this point would
able, or are impractical for problems of the take me away from my main topic, which is style.
size we muat solve, settling for a satisfactory What has the distinction between satisficing
folution instead of seeking the best one is and optimizing to do with style?
generally an excellent way out. Techniques Optimizing techniques generally produce
for sstisficing are often called "heuristic 'unique solutions or small sets of similar solu-
programming techniques." tions. Although one can construct all sorts
Sttisficing is sometimes dismissed as of more or less pathological counterexamples
obvious snd not very interesting common sense: to this generalization, they have little rele-
"If you can't do the best, do the best you vance to the real world. Hence, an optimizer
can." Alternatively, economists sometimes faces no question of style, but simply a ques-
tion of finding the best solution. If we In-
argue that satiificing is simply optimizing sist on according the optimizer a style, we
under constraint on the resources available can only equate it with an extreme function-
for search, The optimizing rule, in this view, al ism, in which function determines form, ut-
if: Continue searching until the expected Im- terly and completely. Few designers who sub-
provement in the solution from investment of scribe to the functionalist view would feel
sdditional search effort is just worth that comfortable with the doctrine that they exer-
effort; then halt. cise no choice and that no questions of style
Because they overlook two important enter their design. But that is certainly
points, the arguments from common sense and where an attempt to combine functionalism with
economies underestimate the importance of tak- optimizing leads us.
ing a satisf icing viewpoint. First of all, For the satisflcer, the unique solution
when no selective optimizing algorithm is Is the exception rather than the rule. Mush-
available, the cost of a search to find the rooms can be found in many places in the for-
best tends to increase at least linearly with est, snd the time it takes us to fill a sack
with them nay not depend much on the direction lie hybrid procedurea that satisfy a aet of
ve wander. We may feel free, then, to exer- constraint!, with one or more variables left
ciae tome choice of path, and even to intro- undetermined, then optimize some function of
duce additional choice criteria (for example, the remaining free variablea. Whether we re-
the pleasantness of the walk or the avoidance gard these as optimizing or satiaficing pro-
of wet placet) over and above the pragmatic cedures depends on two things: how severely
one of bringing back a full tack. the aatiafactlon of the constraints restrict!
There are really two caaea to be dii- the solution! that remain; and whether the
tinguiahed here. In the one caae, two diatinct function to be optimized corresponds to a cen-
aearch paths may be generated by two different tral goal of the design or to a relatively un-
heuriatic procedure! each of which it deaigned important aide condition.
only to find mufhrooma. In the other caae, If the constraints do most of the work
two diatinct aearch patha may be generated by of determining the solution, while the optimiz-
two different heuriatic procedure! becauae on* ation only modifiea it marginally, we often
of them incorporate! criteria beyond the apeak of "suboptimizatlon.-" Suboptimizing i!
basic one of finding mushrooms--or both incor- most profitably viewed aa latiaficing, for
. porate auxiliary criteria, but different one!. thla view calls our attention to the all-impor-
In the former caae, the choice of one or the tant constrainta rather than to the relatively
other of the heuriatic procedure! ia a choict uninfluential criteria. Suboptimizing reminds
of atyle; in the latter caae, the choice of one of the joke about the household in which
auxiliary criteria ia a choice of atyle. In the husband made all the important deciaions
both caaei, it ia the -non-uniqueness of aatia- (about international affalra) and the wife all
'factory solutions that permit! us the choice. the unimportant onea (about the family budget).
Hence, in a world where the beat la the enemy In many auboptimizing procedure!, the con-
of the good, there la almost unlimited room straint! make the really important selections.
for the expression of preference! in atyle In. the remainder of this paper, I shall
even for the functionaliat. be concerned with situatlona where the goals
and aide condition! of the design are all ex-
Criteria and Constraint! pressed aa constraints to be satiafied. Sym-
The requirement! for a design or prob- metrizing the design requirements in this way
lem aolution are usually expressed by some has the further advantage that it allows us to
combination of criteria and constraints. In treat identically both the requirements that
a typical linear programming problem, for ex- are Imposed by the designer and hia client
ample, the criterion ia a function that ia to and the requlrementa that are imposed by nature.
be maximized or minimized, and the constralnta For example, the compressive strength of con-
are bounds on the magnitude of certain of the crete can be introduced into the design prob-
variable!: e.g., minimize the cost of the bas- lem as a constraint of the aame form aa the
ket of food (criterion) that containa at least minimum dimensions of a room.
. 2,000 caloriea (constraint) and not teas than
a', a-, a., .....milligrams of vitamin! and Generators and Test!
mineral! 1, 2, 3, .... (constraint!). Having put all constraints on an equal
Notice that in thia example the diatinc- footing, we now discriminate among them along
tion between criterion and constraint! li not a new dimension we distinguish between con-
the diatlnction we commonly make between the straints imbedded in generatora and constraint!
goala pf an action and the aide condition! it expreased by teat!. The simplest paradigm for
must satisfy. For the goal of buying a baaket a design process contains two subprocesses.
of food ia not to aave money, but to meet The first of these generates, sequentially,
nutritional requirements. Here the goal is a certain objects that are candidate solutions
set of constraints, while a aide condition l! or components of solutions. The second proceaa
the- criterion--the minimization of coat. The teats whether a candidate satisfies a set of
distinction in linear programming between constraints. Any object generated that then
criterion and constraint! ia a technical dia- aatiafiea the test process is guaranteed to
tinct ion between the variablea that are to be aatiafy all the design requirements (2).
aubjected to an optimizing procedure and thoae Most commonly, in a generate-and-teat
that are to be held within bound!. Either may proceai the design is assembled component by
be goal; either may be aide condition. In the component. Each component generated is added
problem of maximizing the return on an inveat- to the previous assembly, and the new struc-
ment portfolio, subject to limit! on riak, the ture then tested. If the test succeed!, the
main goal and the criterion function become process continues; if it falls, the new com-
identical. ponent is discarded and another generated.
When heuriatic technique! are used for Here the deaign process involve! not Just a
aatiaficlng, the asymmetry between criteria cycle of generating and testing, but a whole
and conatrainta disappears. The task i! to aequence of such cycles.
find an object that aatiafiea all of a (po!- The monkey! in the British Museum who
aibly large) aet of constraint!. On the were trying to reproduce the works of Shake-
borderline between optimizing and aati!ficing speare on their typewritera employed the moat
primitive torn of generate-and-test process, out*. The generstor, in this case, guarantee*
ainply typing random sequence! of letter* un- sstlsfsctlon of certsin topological constraint*;
til a sequence* «hould emerge that duplicated the tests deal with other topological con-
fhakeapeare'a. Generste-and-teat processes of trslnts and with metrical properties.
'practical inter eat proceed more subtly and
Selectively than thla. They are constructed The Creation of Design Alternatives
fp (hat any .object they generate la already Up to this point, I have made no dis-
guaranteed to satisfy s certain subset of the tinction between design and sny other kind of
4tfiga requirements. problem-solving sctivity. In ordinary language,
The distribution of design constrsints however, we apply the term "design" only to
between generators and tests is well illua- problem solving that alms at aynthealzlng new
(rated fey theorem-proving programs (3). The object*. If the problem ia imply to choose
prPPf pf * theorem in some branch of mathemat- among given set of *lternatives--to choose
ica is a sequence of well-formed expreaaions, the-location or site for plant, aay-we do
(I) the final one of which is the theorem, to not usually call it a design problem, even if
be proved, and (2) each one of which either ( ) the set of available alternatives is quite
if directly derivable by definite and simple large, or possibly infinite.
eperationa from one or two of the expressions The slgorithm for finding the maximum
that precede it in the sequence or (b) it an of function by setting its first derivative
axiom Off previously-proved theorem. Consider equal to zero select* solution from
now a theorem prover that works backward from non-denumersble infinity of poaaibllitle*.
the theorem to expressions from which it can However Isrge this set of possibilities, It 1*,
be derived» and from these expressions to in s certain sense, given in advenes by the
I ethera, until connection is made with the ax- definition of the function and the range of
|ems and previously-proved theorema. The gen- its independent vsriable. If we don't call
erator in such a program is guaranteed to pro* the maximizing algorithm a design algorithm
' duce only well-formed expressions. Moreover, (and we certainly don't), our terminological
it if guaranteed that in the tree of expres- decision baa nothing to do with the number of
ipns that ia generated, the later members sre alternatives, but possibly with their "given-
i\\ directly derivable in one or more ways ness." But what it means for an alternative
from earlier members. Hence, all that remains to be given requires further discussion,
for the test process is to determine when ex- An architect seeking to produce the room
prepaions produced by the generator belong to Isyout for a house would ordinarily be thought
the axiomp and previously-prrved theorems. to be engaged in design. But the number of
Other divisions of labor between the gen- alternative floor plans that can be drawn on
trftPr and the test process of a theorem prover a sheet of paper Is exactly the same nor-
re conceivable. For example, the generator denumerable infinity the Infinity of the con-
night produce expressions that are not guaran- tinuum--as the number of points on a contin-
teed tP be well formed, but grammaticality of uous function, neither more nor less. If the
theie expressions could be determined by s possible solutions of the maximum problem are
tClt» and the ungrammatical ones weeded out. given, why not the possible solutions of the
An analogue to this can be found on certain architectural problem?
«hepp-playing programs that generate moves not The difference between these two situ-,
futrantead tp be legal, and then test the moves atlons csnnot be divorced from the avail-
for legality if they have other desirable pro- ability of solution processes. There are two
• pcrtitf. reasons, one negative and one poaitive, why
The division of labor between generators we call the layout problem, but not the max-
ni tMtf if pragmatic. It depends in consider- imizing problem, a design problem. The nega-
able measure upon what constraints are easily tive reason is that we do not have .a simple
embedded in generators and what onea are not. finite algorithm for going directly to a solu-
The theprem-prpver example suggests thst when tion of the layout problem. The poaitive
the generator ia producing a continually grow- reaaon is that the processes we do have for
ing afpetnbly of potential design components, attacking It involve synthesizing the solu-
''iecfl" epnptraints--that is, relations among tion from component decisions that are selec-
neighboring components--are easily incorporated tive, cumulative and tentative.
in generaterp, while more global conatraints The possible slternatives for solving a
«y hive to be left for tests. design or synthesis problem sre also given in
J, Craion S (1970) program for laying out c certain sense. Synthesis does not make
fleer planf provides another example of a div- things out of whole cloth, but assembles them
ilien of labor between generators and tests.. from components. What is given In the case of
. Ihf program employs a planat graph representa- synthesis are a set of elementary component*
tion ef the room layout problem, thereby guar- and a grammar defining the admissible ways of
anteeing that any arrangements generated will combining components into larger structures.
lit in the plane. Satisfaction of apecified In the layout problem, the components are pos-
Size constraints for the rooms, however, is sible walls, each of which can be defined by
guaranteed by tests spplied to proposed lay- the four coordinates of it* two endpoint*.
(For simplicity, I am denying the architect We can illustrate the proceas by refer­
the privilege of curving hi* walls t but the ence again to theorem proving, which ia not
generalization to permit this would not change usually thought of aa design, but whose pro­
the argument.) The baaic grammatical rule it cesses do not differ from design processes in
Chat valla nay not overlap—that the entire get any aigniflcant way. There are two way* of
of walla auat constitute a planar graph. viewing theorem proving. It can be thought of
Why do we think the architect la synthe- a* aearch for the theorem through a apace of
aicing, or even "creating" when he make* the theorem*, in which each *tep involve* applica­
layout? Because he *olve* hi* problem by mov­ tion of the rule* of inference to point* of
ing through a large combinatorial apace in the Space previously visited. (This ia a de-
which he adda one element after another to hia acription of forward search; with a little
deiign—of course, sometime* revising or even generalization, backward aearch can be brought
deleting those that are already in it. In thla within the paradigm.) But it can also be
reapect, hi* process 1* not different from that thought of aa a process for generating an ob­
of a chemist who is trying to synthesize a com* ject called a proof—that 1*, a sequence of
plex molecule having certain desirable proper- theorems having the propertiea we mentioned
tie* (for example, a broad-band, biodegradable earlier. At each atage In the aearch, one or
inaecticide .to replace DDT). To be sure, the more partially completed proof* are augmented
chemist starta with only a finite number of or modified until a proof of the desired theo­
building blocks--the 92 natural elements, say— rem haa been found.
while the architect atarts with a four-parameter
non-denumerable infinity of possible straight Now aelectivity in thi* design process 1*
derived from several *ource*. First, the gen-
llnea. The chemist, we might think, haa far erator producea only legitimate proof steps.
fewer alternativea open to him than the archi­ Second, the *y*tem incorporate* a number of
tect. intermediate tests thst guide the generation,
But we have already aeen that number of by measuring progress toward the desired proof.
alternatives is a misleading criterion. "Large"
and "small" mean one thing in mathematics, but For example, the final expression in the par-
something quite different in the empirical tial aequence can be compared with the theorem
world. For the problem solver, any number of to be proved,- and the differencea between them
alternatives is immense (and, for practical used to determine the next atep of generation.
purpose*, equal to any other immense number) (This is the procedure, Incorporated in the
if it is too large to be enumerated and ex- General Problem Solver, known aa means-end*
amined exhaustively within a reasonable time. analyaia.)
In design, 1,000 is an intnense number, for de- Hence, we ahould not conceive of the de-
sign* must generally be produced by examination ign generator as a process that spews out in
of fewer than 1,000 alternatives. a fixed aequence possible designs that are ac-
The richness of the combinatorial apace in cepted or rejected by final tests. The gener-
ator ia itself a parametrized procesa whose
which the problem solver moves, then! rather output at any given moment 1* determined both
than the number of elementary components he haa by it* structure and by the information it ob-
at hia disposal, la the hallmark of design act- tains from it* previous output. The testing
ivity. By virtue of this richness the designer processes include both tests for conformity
must search selectively; for among all the im- with the problem constrainta and tests for
menae--and possibly infinite--number of combin- measuring progress toward satisfying thoae
ations potentially available to him, he can ex- constraint*.
amine only a tiny number. The designer exer- Each of the partially completed designs
cises choice, hence introduces style into hi* generated during search corresponds, of course,
design. to a whole class of final designs. The testa
Notice that the problem apace through that are applied during the design process do
which the designer searches is not a space of not merely weed out unsatisfactory design*,
designs, but a apace of design components and but instead, they choose promising classes of
partially-completed designs. Even this is a designs for further particularization.
simplified picture, for the designer may oper-
ate sometimes in the space of design components, The Emergence of Style
sometimes in one or another abstract planning The satisficing process we have been dis-
apace that represents only certain aspects of cussing can be described, *lightly *implified,
the entire problem. In all eventa, in the aa follow*:
course of the design process very few complete The design program consists of one or more
designs are generated, compared and evaluated generators and a number of teat*. When the
in the typical case, only one. The generator program is furnished a se.t of constraints
guarantees that a number of the constrainta that ia, a set of criteria for determining
will be satisfied automatically at each atage when a satisfactory solution to a design prob-
of the design, and the tests guarantee that * lem has been obtained it proceeds to gener-
partial designs violating the remaining con- ate elements for consideration as component*
strainta will be rejected or modified a* the in the design. After each proposed addition
earch proceed*. to the design, two kinds of testa are applied.
Tests of the first kind determine whether the
proposed addition satisfies certain of the de­ program derive much more from the architect's
sign constraints—if so, it i« added to the •tore of criteria than from those provided
design. Tests of the second kind inform the by the client. * *
generator whether progress is being made and To view the matter in a little different
what constraints remain unsatisfied, and there­ light, there are of course strong*positive
by modify the next stage of generation. reasons for incorporating in design programs
The relation between the constraints de­ both prefabricated assemblies snd autonomous
fining a particular design problem and the constraints. The former are a natter of ef­
design program is not simple. Not all of the ficiency—they make design less costly--but
constraints are incorporated in test routines they are more than that. A subproblem solution
for, as we have already noted, the generator OBJ that is to be used repeatedly as a component
be, and usually will be, so constructed as to in the-solution of larger problems can be
satisfy certain ot the constraints automatical* brought to a level of high "polish11—that if,
ly. Moreover, not all of the tests that are can be designed to severe constraints--beyond
applied by the program come from the problem what would be justified if -it were to be used
constraints. We need to distinguish, there­ only in a single design.
fore, those characteristics of the generator Autonomous constraints in design programs
that simply guarantee satisfaction of design may also be highly desirable features. If the
constraints from its other characteristics-­ designer is a professional expert, serving a
let us call them autonomous—which are not lay client, then the client nay want the ex­
implied by the initial problem requirements. pert to take care of all the considerations
that will become important to him in his actual
Similarly, we need to distinguish autonomous use of the-object designed, but which he ia un­
tests from tests that simply apply design able to anticipate in advance.
constraints. Both generators and tests of In conceptualizing a design program that
the design program will generally contain both makes use of prefabricated subproblem solutions
problem-determined and autonomous components. and supplementary autonomous constraints, it is
One of the simplest, and most obvious, useful to distinguish between the program
autonomous components of a design generator proper (the generators and test processes) and
is the set of processes that determines the a memory in which the autonomous supplements
order in which elements will be considered. are stored. Only a small subset of the latter
Seldom do the specifications of the object to may come into active use in the course of de­
be designed place restrictions on this order. signing any single object. They will be e-
Thus, for example, the architect's generator voked by particular situations that arise in
is free to begin work on a problem from the the course of the design, retrieved from memory,
outside in or from the inside out. But in a and applied. Obviously, the contents of this
satisficing process, seeking the first object memory will make an important contribution to
that will satisfy s specified set of con­ the style of the design program having access
straints, the order in which possibilities to it.
are examined may have a major influence on In summary, we can speak of the distinc­
which solution is discovered. Hence, auto­ tive style of a.design program because the con­
nomous characteristics of the generator that straints that define design problems for it
determine the order of search are an important are never so limiting as to point to a unique
aspect of style. solution. All of these characteristics of the
A second set of autonomous elements in program that determine which of the many pos­
most design programs are "prefabricated" solu­ sible satisfactory designs will actually be
tions to subprobleajs that arise repeatedly created define the program's style. Among the
*in different contexts. By using such assemb­ most important of these autonomous, style-
lies as components in the design instead of determining characteristics are (1) the pro­
synthesizing it from simpler elements, the cesses that determine the order of search, (2)
program is able to operate at a more aggre­ stored prefabricated solutions to recurrent
gate level and reduce its search effort. subproblems, and (3) stored autonomous con­
Since, again, the prefabricated assemblies straints to be added to the explicitly given
are not uniquely determined, they provide • problem constraints.
second source of idiosyncratic style. In ordinary discourse, style is predi­
The design program may add to the speci-. cated both of objects and of persons who create
fication of a design problem yet another set them. We speak of a building as Gothic in
of autonomous constraints. If the program style, but we also speak of Mozart's style of
resides in the head of a hunwn designer, music. In the former case, a collection of
these additional specifications may simply objects possessing certain common character­
represent characteristics the designer would istics is set off and labelled as belonging to
like to see included in the object, over and a particular style. In the later case, the
above those mentioned in the problem specifica­ collection is defined by the design program
tions. It is frequently suspected by the from which the objects issued (4).
clients of architects that the constraints Mozart, in a celebrated letter, puzzled
actually applied by the architect's design about the connection between design process '
•nd product: "Now, how does it happen, that, of the object, end by features of the program
while I am at work, my compositions assume tht that designed it, respectively. In this case
form or the style which characterize Mozart the designer was Nature. Kepler discovered
•nd arc not like anybody else's? Just as it s beautifully succinct characterization-of the
happens that my nose is big and hooked, Moz­ style of the planets, by determining that each
art's nose and not another man's. I do not moved in sn ellipsoidal orbit, with the Sun
•in at originality and X should be much at a st one focus of the ellipse. But Newton showed
loss to describe my style. It is quite natural that the style could be characterized even
that people who really have something particu­ more fundamentally by a gravitational process
lar about them should be different from each that accelerated each planet toward the Sun
ether on the outside as well as on the inside." in inverse ratio to the square of its distance.
To determine whether an object belongs to The style of the process described by Newton
• particular style, we may test whether cer­ implied the style of the orbits generated by
tain constraints, obvioua or subtle, apply to that process.
it. If its arches are pointed, we aay, "That's
Gothic," if round, "That's Romanesque." Or Style from the Dynamics of Design
we nay reject the former classification on In the remainder of this paper, I should
discovering that the arches do not bear the like to give particular attention to aspects
weight of the vault, but hide a steel frame. of style that reveal characteristics of the
Whatever the criteria, obvious or subtle, design process. In doing this, I shall have
tests spplied to the characteristics of the ob­ to stray perilously close to realms of esthet­
ject tell us whether or not it belongs to the ics in which I have only amateur status, and
style in question. to draw examples from painting, music, and
We determine whether a particular object architecture. If my examples do no more than
was produced by a particular design program in irritate professionals into correcting or im­
exactly the same way--by applying tests to proving them, they will have served their
characteristics of the object. One reaaon we purpose. • .
do this is because we ordinarily do not have The Kepler-Newton example given above il­
direct access to the program, hence can iden­ lustrates the central esthetic premise from
tify it only by its products. We cannot (yet) which I shall proceed. Kepler found pattern
look inside Mozart to determine what it is in the planetary orbits, but Newton disclosed
about his program that makes his music Moz­ the deeper pattern by demonstrating that it
art's. But the cues we detect in the design was not.a brute fact of Nature, but followed
to identify its source may also be obvious or inexorably from the working of a simple gen­
subtle, superficial or essential. erator. The Newtonian explanation gave us
Some painters employ obvious trademarks— the more parsimonious characterization of th'e
the usually conspicuous X, for example, that pattern.
can be found on Stuart Davis's canvases. The Functionalism in design offers the same
.''trademark is an explicit and conscious symbol kind of parsimony. When we notice that some
' of the author's identity, almoat a signature.
The tests range from these very obvious ones feature of a design follows from the function
to others that are difficult or impossible to the design is supposed to serve, that feature
no longer requires its own independent con­
verbalize. As an exercise for the reader, I straint—its special flat—but derives from
propose that he try to write a characteriza­ conditions central to the design problem It­
tion of the paintings of Cezanne that would self. Functional design is parsimonious of
permit someone who had never seen one to iden­ constraints, producing objects that are more •
tify them. than the additive outcomes of multitudes of
A more sophisticated approach to charac­ detailed design specifications.
terizing the style of a design program is to
define it by statistical measures. This ap­ At one time or another, both obedience
proach has been farthest developed in the to function and simplicity have been erected
into basic esthetic principles. What our
identification of literary style from the analysis suggests, instead, is that the key
Statistics of vocabulary and syntax, and the term is parsimony, and that we value function­
identification of music from the statistics of al ism and classicism precisely when they dis­
tonal or chordal sequences (5). The technique close economy of pattern. Functionalism be­
is sometimes successful, and undoubtedly can comes an important source of parsimonious
be developed further, but in some important design precisely because Nature la cleverer
sense we find it artificial. The identifying than Man in devising dynamic principles that
statistics are clearly symptoms, not causes. have rich sets of consequences for the objects
They are the accidental byproducts of the they generate. Simplicity warns us against
working out of the underlying characteristics multiplying constraints as a cheap source of
of the design program, and we should expect to superficisl richness in pattern. It urges us
find no direct one-to-one relation between the . to omit detail when that detail is arbitrary
statistics and the program characteristics. and not derivable from constraints or compon­
One important episode in the history of ents of the design process that are already
modern science illustrates dramatically these
two ways of characterizing style--by features preaent and expressed in other waya. It
urges us, also, to be sparing of detail when
that detail vould make it difficult to per­ line, harmony, tonsl quality, and others. Goal-
ceive the. dynamic source* of pattern. position cannot proceed simultaneously with sll
Let BC leave these general principles be-*f these dimensions, but must tske one, or s small
fore I strain them beyond credibility, and number as the point of departure, with conse­
turn to some examples of how style emerges quences for the style of the completed work.
from (and can be characterized by) dynamic Roger Sessions (THE MUSICAL- EXPERIENCE,
features of the design process. Princeton University Press, Princeton, N. J.,
A crude architectural example vas already 1950, pp. 46-47) has given a description of
suggested earlier in this paper. The feneetrs- the process of composition that accords veil
tion of a building can be entered into the de­ with this view of style.
sign at an early stage, or it can be postponed **I vould say that a musical idea
until many aspects of the internal layout have is simply that fragment of music
been determined. If we see a building with a which forms the composer's point
symmetric facade, we can be reasonably sure .of depsrture, either for s whole
that that facade was generated at an early composition or for sn episode or
point in the deaign. If, on the other hand, even single aspect of s composi-
ve see one with many asymmetries, we will con­ 'tion.. I ssy 'fragment 1 knowing
jecture with some confidence that these asym­ full veil that it can get me into
metries are external expressions of decisions difficulties. For in my exper-
about how to meet internal requirements. ience, in which I include observ-
This does not mean, of course, that asym- . atlon and analysis ss well' ss
metry cannot be introduced deliberately as one composition, a 'musical idea'
of the design constraints. A designer can the starting point of a vital
Introduce any constraint that whimsy or esthe- musical 'train of thought'--
tic sensibilities suggest, provided it does caa be virtually anything which
not interact too strongly and contrarily with strikes a composer's imagin-
the other constraints already present. When .
stion. It may, certainly, be a >. .*
he does so, he generally succeeds in producing * motif, a small but rhythmically *'
something that the viewer finds "quaint" or self-sufficient fragment of
"ornamental." He does not produce a Mont-St- melody or of harmony; but I am
Michel, whose wonderful forms sre almost by- fairly certain that by no means
products of the demands of site and history, sll motifs can be called 'musical
but which, far from appearing accidental or ideas'. On the other hand, I
adventitious, are integral expressions of those could cite many examples where
fundamental demands. the most essentisl musical ideas,
This is not simply another way of saying the elements that give the music
that the forms are functional. However, dif- its real character, consist not
ficult it is for us to imagine it now, the rock in motifs st sll, but in chords,
of Mont-St-Michel might have been" crowned by in sonorities, in rhythmic fig-
many other masses of masonry performing the ures, or even in single notes of
same function. Perhaps some of these alterna- a particularly striking context.
tives would have excited as much wonder over Sometimes--and this occurs, I
the years ss the particular one that emerged. think, more often though not
But each of them would have a style reflecting always in works of composers
not just the demands of the site and its use of great maturity, in 'late
(identical for all), but also the process for works,' as we call them--one of
'generating possible ways of meeting those de- the most important musical
mands . ideas, in a fundamental and
Of .course an element of style that a- motivating sense, may be not
rises in this way--from interaction between de- even a thematic fragment at sll
sign, requirements and design processes--may but some feature of the large
become so valued that it is incorporated dir- design, such ss a recurring
ectly into the design constraints. A modern relationship between two har-
Cothic or Gothicized church, built on a steel monica or keys, or even s lin-
frame, deliberately satisfies visual require- ear relationship embodied in
ments that in no way follow from other design different aspects of the music
constraints or processes, but are simply orna- at different moments."
mental. The objection to such ornamentation The differences among the major styles of
is not its lack af functionality, but its classical Western music further illustrate
failure to communicate something sbout the de- the dependence of style upon process. The com-
. sign process. position of homophonic music, with its emphasis
I will return later to the topic of or- on a prominent melodic line harmonized by the
namentation. Before I do so, I should like to other parts, proceeds in quite different ways
look to music for other illustrations of the from the composition of polyphonic music, with
basic thesis. Music has a complex structure its interweaving of independent voicea. View-
built on the dimensions of rhythm, melodic ed from a harmonic atandpoint, polyphonic
eiusic txhibits oany unuaual chorda or even tional painting, the role of the proceaa in
diaaonancea (e.g., of paaaing tonta) that art forming the atyle ia especially clear. Here,
timply byproductt of the working out of con- the painter may put an initial deaign upon the
trapuntal problema; and theae harmonica and canvaa for ita auggeative value aa to what ht
diaaonancea are an important tlement of poly- night do next. Such painting becomes even
phonic ttyle at experienced by the liatener. more than painting with other methoda--a
But polyphonic muaic i* not written by intro^ feedback proceaa of interaction between paint-
dueIng diaaonancea.
Perhaps the.clearest expreaaiona of thit er and partially completed canvaa.
notion of the tourcet of ttyle come from the Duane Palyka haa uaed the computer aa
txponentt and practitionera of compoaition vith Inatrument to create viaual compositions with
twelve tonea--the idea of Arnold Schoenberg the aid of thia kind of feedback. Hia com-
that vnderliea ao much Twentieth Century muaic. poaing program employed a number of aeparate .
(Set, e.g., Joteph Rufer, Composition with aubroutinea. The initial tubroutihea produced
Twelve Notea. London: Barne and Rockliff, a aimple compoaition that was then stored In
reviaed edition, 1961.) The twelve-tone pro- memory--generated aome pointa, for example,
cedure may be eunmarized crudely aa followa: that then served aa centera for circlea of
The chromatic acale containa twelve tonea in varying diametera; or generated pointa
"each octave (correaponding to the twelve ktyt that aerved aa vanishing pointa for aubsequent-
per octave of the piano keyboard). The com- ly-generated aeta of linea. Other aubroutinea
poser selects, on the baaia of aome aet of then examined and modified the atored initial
criteria, a particular tone row, a permuta- compositions, distorting figurea, shading
tion of the twelve tonea of the chromatic them, aharpening or blurring boundariea, and
acale. .He then proceeda to deaign a compoai- to on. In principle, there waa.no limit to
tion, uaing thia ordering of the tones to the number of atagea in the compoaing proceaa,.
define both the vertical (harmonic) and but- in practice, two or three atagea sufficed
horizontal (melodic) relationa. There are to produce compositions of aubatantial eathe-
many waya in which thia can be done, and tic intereat.
correspondingly many aub-atylea that are all
covered by the broad label of "twelve-tone Function. Process. Ornament
music." But all twelve-tone muaic recelvea If the designer's task it to choose and
t strong anylietic imprint from its method of aelect, then he must hsve aome baaia for
generation--from taking the tone row at the choosing what he doea. In worlds having Im-
initial input to the generator. mense numbers of alternative possibilities,
The quotation from Combrich with which I it would seem that he' would need a great many
introduced thia paper equates style with criteria in order to narrow down theae poa-
choice, but the choices of which Combrich aibilitiea to a aingle deaign. He would bt
speaks pertain mainly to perception rather confronted with comblnatorica in the apace of
r than to design process. Gombrich's choice it possible criteria almost sa fearsome aa the
'.' a choice among the many perceptual cuea that combinatorics in the space of possible de-
'are available for representing, or creating aigna, and hia deaign activity-would have to
the Illusion of, the external reality. The be preceded by a pre-design activity to aelect
painter chooses a amall aubaet of theae cuet the criteria.
at the baaia of hia ttyle and omita or de- However, the deaign process synthesizes
emphaaizea the reat. The artiat cannot meet only one (or a few) objecta rather than a large
the totality of demands that an attempt to number, and proceeda iteratively to build the
reproduce nature literally would impose on object cumulatively from an initial kernel.
him, hence selects out aa hia design con- Under these circumstancea, a few conatraintt
atrainta Just a few of these demands. embedded in the atructure of the design pro-
In actual fact, perceptual aelection and cess itself, together with a few design re-
telection of deaign proceaa are not at all quirements applied again and again at each
Independent. For example, the choice between atage of the cumulation, may auffice to de-
a linear and a painterly mode of representa- termine the final deaign. To the extent that
tion carries with it all aorta of commitmentt they do, the object designed will exhibit deep
at to how the painting ahall proceed. Witneat unity in all ita aspects, which express the
Leonardo's prescription for (linear) painting working out in interaction of a amall number
la Chapter IV of hia AKT OF PAINTING: of processes and parameter!.
"Painting is divided into two Dealgners are most fortunate when nature
principal parta. The firat it and technique impose severe constraints upon
the figure, that la, the linea them. If we did not understand it, we might
which distinguish the forma of view the cathedral at Rheims aa highly
bodies, and their component decorated and decorative. But with any feel
parta. The aecond ia the for the Gothic deaigner'a problem, we aee the
colour contained within thoat flying buttreasea and their pinnacles aa
limitt." functional meana for turning the great aide-
Ia tome modea of modern non-representa- ward thrusts of the vault downward and con-
ducting then safely ttf the ground. The
elaborateness of the atructure it acceptable Notes
because it arises from applyiog a few baaic * This work was supported by Public Health
processes to an equally few imperative con- Service Reaearch Crant MI-07722, from the
straints. National Institute of Mental Health. The
When a act of initial deaign conacralnte ideas set forth here begsn to urge themselves
ia too eaaily satisfied, the deaign proceaa on me when I vas preparing my Karl Tavlor.
vill preaent us with aolutiona too eaaily. Compton Lectures, now published as Th<
Ve vill obtain more than one solution, and Sciences of the Artificial (M.I.T.'Press,
vill feel constrained to choice among them-- 1969). For s discussion of design, see Chap-
that is, to impose new constraints that bear ter 3 of* that volume, snd for s brief state-
no particular relation to the initial act. vent of its relation to style, p. 75.
Since these will frequently be constraints on
characteristics of the final design, rather 1. The reader wishing to pursue this point
than on its suitability to function or on the vill find s fuller discussion in my Models 'of
search process, they will lead us to super- Man (Wlley, 1957), particularly the introduc-
ficiality and mannerism. tion to Part IV, and Chapters 14, "A Behav-
An alternative is to retain the Initial ioral Model of Rational Choice," and 15,
constraints but set more severe acceptance "Rational Choice snd the Environment."
levels for them. Again, the Gothic cathedral
is a case in point. As long as the deaigners 2. The organization of problem solving sys-
made extravagant demands on the structure to tems is discussed more fully in Newell, A., 6
admit as much light as possible and, as a Simon, H. A., Human Problem Solving (Prentice-
means to doing this, sought to raiae the Hall, forthcoming, 1971),especially Chapters
height of the vault, all of their ingenuity 4 snd 14.
vas spent in finding £ .way to achieve their •tv»
goal. There was great need for elaboration, 3. Newell, A., & Simon, H. A. (1971), . /
but * little room for irrelevant elaboration. Chapter*4.
Conclusion 4. Defining the boundaries of s style by a
My main purpose here, however, is not to period in time'and a geographical location
preach a particular esthetic doctrine. Rather, really amounts to the same thing, for such
I have sought to show that those differences boundaries only make sense if there sre com-
among designs that we call "style"' arise from munal ities among the design programs thst
s variety of sources, and that the design operated inside them.
process is one of the most important of these.
The imprints of design process upon style are 5. See the extensive discussion of these mat-
fundamental in the same sense that the imprints ters in Killer & Isaacson, Experimental Music
of function, of material, and of manufacturing (McCraw-Hill, 1959), Chapters 5 and 7.
process are.
Ve have now entered an era where, for the 6. See Killer & Isaacson, 1959.
first time in history, the design process
itself is becoming directly accessible to
examination and description. Increasingly,
ve are able to characterize the method of
creating explicitly, instead of being forced
to indicate it indirectly, in descriptions
of the things created. Already, this new
capability is being reflected in novel forms
of artistic activity, in which the human
artist's task is to create a deaign process
that vill, in turn, create objects of esthetic
value. I have cited above the example of
Falyka's program for producing non-representa-
tional "prints" (1 don't know what else to call
them); and many of you are familiar with com-
puter-composed music like Lejaren Killer's
ILIIAC SUITE a.id COMPUTER CANTATA (6).
This shift in attention from the work of
srt to the design process is bound to have
consequences for style and design as signifi-
cant as the transformation from Kepler's kine-
matic pattern descriptions to Newton's dynam-
ic ones.

You might also like