You are on page 1of 8

Artificial Intelligence, Computer Simulation

and Theory Construction in the Social Science∗


Rainer Schnell
1991

Abstract
The current use of Artificial Intelligence methods as tools for the-
ory construction in sociology is reviewed. The main use seems to be
the modelling of cognitive models processes of individual actors. Other
approaches to simulation in sociology uses very simple models of indi-
vidual actors, especially cellular automata. The theoretical importance
of this kind of model for sociology is neglectable. The most promising
application of simulation methods are explications of rational choice
based theories.
keywords: Artificial Intelligence, Computer Simulation, Theory Con-
struction in the Social Science

1 Artificial Intelligence Approaches to


Theory Construction
In the social sciences with very few exceptions (for example: Schrodt 1985)
most articles concerning the uses of artificial intelligence methods1 are only

This is a reprint of a lecture given 1991 and published in 1992 in the conference reader
(Frank Faulbaum (ed.) (1992): SoftStat ’91. Advances in Statistical Software 3, Stuttgart,
p.335-342). Only a few typos have been corrected. The lecture itself was an extension of
an earlier review of computer simulations in sociology (Schnell 1990).
1
A critical account of the achievements of artificial intelligence, which in the domi-
nant reception mostly and wrongly boils down to expert systems, can be found a recent
collection of articles by Stephen Graubart (1988).

1
demonstrations of an extended vocabulary of the author. An example of this
kind of work is Garson’s article (1989). The main current real use of artificial
intelligence methods in social science for theoretical work is the development
of explicit theoretical models of cognitive processes2 . For Cognitive Sciences,
theoretical based simulations are essential (H.A. Simon and C.A. Caplan
1989, Z.W. Pylyshyn 1989). The astonishing progress of Cognitive Science
(in sharp contrast to the stagnation of sociological theory) seems to be im-
possible without computer simulations. Computational cognitive science is
directly relevant for sociological work: For example as a theory basis of data
collection (Computational Memory Models, see D.L. Hintzman 1990), for the
theory of personal identity (M. Minsky 1987, Hill and Schnell 1990), emo-
tions (P. Johnson-Laird 1988) and for the theory of interaction structures
(Doran 1985, Banerjee 1986). Data structures like frames and scripts can be
directly imported and included as part of the working model of the actors of
a sociological simulation. If the explanandum requires an extended micro-
theory of an actor, elements of it can be found in cognitive science. Therefore,
the possible applications of artificial intelligence for theory construction in
the social sciences may be more in the domain of using data structures of
AI-simulations for the organization of every day knowledge required by in-
dividual actors3 . The frame problem in artificial intelligence (see Pylyshyn
1987) has a direct correspondence to the aims of ethnomethodology: What
kind of knowledge is required for every day actions? Because of the refusal
of most ethnomethodologists to use any formal method, the striking simi-
larities of both problems has not been noticed in the literature. There is
only one major exception: The work of Kathleen Carley (1989). She uses a
detailed model of cognitive processes of individual actors for the explanation
of the generation of every day social knowledge. Her theory (named "con-
structuralism") is used for the explanation of social processes like diffusion
of innovation and the genesis of invisible colleges. Unfortunately, it seems
to be, that she doesn’t use a computer model. Her work shows, that the
accumulated knowledge of sociologists is not sufficient for the explanation of
2
Other uses are the of logical consistency of a theory with expert systems. Such is
demonstrated by Edward E. Brent (1986) under the strange label "qualitativ formalism".
An earlier similar application is David Sylvan’s und Barry Glassner’s (1985) formalization
of a theory by Simmel as LOGLISP-program. These applications are no simulations (the
generating mechanism is not specified) and are therefore not considered further.
3
Social scientists are just beginning to realize the theoretical power of these concepts
for their own work. See for example, Esser (1990).

2
individual knowledge structures. Of course the results of ethnomethodology
are of no use for the solution of practical modeling problems. If you want to
simulate actors on this level of detail, you have to do your own fieldwork. But
very often, this is not necessary. The currently most promising explanation
methodology in the social science seems to be the rational choice approach,
specially of the kind of Lindenberg’s approach: His "method of decreasing
abstraction" ("As simple as possible and as complex as necessary", Linden-
berg 1989a:11) uses detailed models of individual actors only if no simple
model is sufficient. This approach will be illustrated below.

2 Models of individual actors and the rational


choice approach
The most simple version of models with individual "actors" are variations
of the so called "Checkerboard Model of Social Interaction" (Sakoda 1971),
or if you prefer, simple cellular automata. Most of these simulations are
extensions of the segregation model of Schelling (1971). There are no theo-
retical advances since 1971 in the application of this kind of models in the
social sciences4 . Due to the easy programming of stunning graphical effects
with these models, a lot of students is busy with this kind of model, but
they are only reinventing the wheel. There is a very interesting, partly even
formalized, literature concerning cellular automata (Burks 1966, 1970, Kauff-
man 1984, Toffoli and Margolus 1987 and expecially fascinating: Poundstone
1987) but I was unable to locate any theoretically relevant idea for the devel-
opment of social science theory. As always, there is at least one exception:
Phillip Schrodt (1981) has shown by such a simulation, that three simple
assumptions concerning the behavior of individual actors are sufficient for
explanation of territorial units.
An example of non-sociological work on computer simulations of individual
actors for sociological problems consists of two papers of the Second Confer-
ence on Artificial Life, which was held in 1990. MacLennan (1990a, 1990b)
has simulated a population of simple machines, which reacted to their envi-
ronment and were able to modify it. Due to a selection pressure in favor of
cooperating behavior in the simulated environment, it was shown, that the
possibility of communication among the machines increased average fitness
4
For an example, see the simple model of Nowak, Szamrej and Latane (1990).

3
by 84%. Furthermore, a significant association between symbols and situa-
tion was observed. Therefore, the simulation has shown a possible mechanism
of the development of communication and language.
A beautiful example for a theoretical based simulation of individual actors
is a waiting queue model of Norman P. Hummon (1990)5 . Without any as-
sumptions concerning exchanges, rewards or sanctions, his model explains
the development of division of labor and organization on in a population of
individual "bcrats".
It is remarkable, that a sociologist, who doesn’t share the rational choice
perspective, has reached very similar conclusions concerning the structure
of a simulation of social actors6 . Bo Anderson (1989:214) has described the
elements of such a simulation. He does not mention rational choice theory
at all, but his description of simulation models comes sufficiently close to
an economic explanation: His simulated actors have interests, they try to
maximize, they live in an environment with economic constraints. Further-
more, his actors have world-views representing the input received from other
actors and the environment. I would prefer to call this "subjective expected
utilities". At last, he writes: "For each actor, there are also production rules
that connect the interpretations of the input from other actors and the events
in the environment with the available action options. The actors cognitive
structures, in combination with the inputs, produce beliefs about causes,
about what other actors did, and what the actor himself can do." If we use
the means of such production rules as simulation tool for the formalization
of the production functions of the rational choice theories7 , we have the basis
for a complete explanation. For the explanation of preferences Lindenberg
(1989a, 1989b) uses social production functions, which vary according to so-
cial positions. Production functions are the knowledge of the technical means
to reach the ultimate goals of human beeings: physical well being and so-
cial approval (Lindenberg 1989a:14, concerning norms and social production
5
This is the only theoretical based waiting queue model I’m aware of.
6
See also the multi-actor system of the TEAMWORK project described by Jim Doran
(1985:163 ): "(...) a multi-actor system requires several actors independently interacting
with a common task environment. Within a computer program this is achieved by setting
up ’clones’ of the basic actor program each with its own knowledge repertoire and tempo-
rary plan structures. The simulated task environment is a separate part of the program
capable of being inspected or manipulated by any of the actors."
7
Other, and more simple tools than production rules are available, if we use a subjective
expected utility framework for decisions.

4
functions, see Lindenberg 1989b:190-194).
Although the application of explicitly rational choice based simulations of
individual actors for the explanation of macro-phenomena is a new field, a
few simulations have been completed recently in Germany. Frank Kalter
(1990) has done a computer simulation of the so called theory of the "Spi-
ral of Silence." He showed the ill-definition of the theory and the extremely
small range of possible applications due to the very specific initial conditions
implicitly required by the theory. Christof Wolf (1991) is currently working
on the simulation of various theories of friendship choices. He has tried to
study the implications of various constraints on friendship choices and com-
pares his results to empirical data of the GSS. Wolfgang Sodeur (University
of Essen) has studied by computer simulation the impact of demographic
constraints such as the number of persons within an acceptable spatial range
and an acceptable age range on friendship choices of children. Karl-Dieter
Opp (1991) has used a BASIC-program with simulated simple individual
actors for exploration of a rational choice theory of collective action.

3 Final remarks
Finally, I would like to add three technical remarks. First, currently, the-
oretical relevant simulations in the social sciences do not need expensive
machines or special hardware. For psychology, Broadbent (1987:171) sum-
marizes: "You should be able to replicate the results by borrowing from your
nearest teenager the machine usually used for playing space invaders." Sec-
ond, because simulations should be used as a tool for theory construction,
which only is possible by a public discussion of proposed theories, it is essen-
tial to use a language, which is widely available and as much self-documenting
as possible. I believe, only PASCAL without any extensions fulfill all these
points. Furthermore, it should by now be clear, that a simulation without
easily public available source code is scientifically worthless. That is much
more a psychological problem than a technical: Many (european) authors
don’t like to publish programs, in which they had invested a large amount of
work. But for a public discussion, this attitude has to be changed8 .
8
Technically it may be necessary to establish a fileserver in a research network, which
can hold such files. Examples for such servers are the BITNET TRICKLEs, see Schnell
(1989).

5
4 References
Anderson, B. (1989): On artificial intelligence and theory construction in
sociology; in: Journal of mathematical sociology, 14, 2-3, p.209-216.
Banerjee, S. (1986): Reproduction of Social Structures: An Artificial Intelli-
gence Model, in: Journal of Conflict Resolution, 30, 2, p.221-252.
Brent, E.E. (1986): Knowledge-Based Systems: A Qualitative Formalism,
in: Qualitative Sociology, 9, 1986, 3, p.256-282.
Broadbent,D. (1987): Simple models for experimental situations; in: Mor-
ris,P. (ed.): Modelling Cognition, Chichester, p.169-185.
Burks, A.W. (ed.) (1966): Theory of Self-Reproducing Automata, Urbana.
Burks, A.W. (ed.) (1970): Essays of Cellular Automata, Champaign.
Carley, K. (1989): The Value of Cognitive Foundations for Dynamic Social
Theory; in: Journal of Mathematical Sociology, 14, 3, 1989, p.171-208.
Doran, J. (1985): The computational approach to knowledge, communica-
tion and structure in multi-actor systems; in: Gilbert,G.N./Heath,C. (eds.):
Social Action and Artificial Intelligence, Aldershot, p.160-171.
Esser, H. (1990): Habits, Frames und Rational Choice; in: Zeitschrift für
Soziologie, 19, 4, p.231-247.
Garson, G. D. (1989): Computer Simulation, Artificial Intelligence, and Po-
litical Science; in: Garson,G.D./Nagel,S.S. (eds.): Advances in Social Science
and Computers, Vol. 1, Greenwich, p.25-45.
Graubart, S. R. (ed.) (1988): The Artificial Intelligence Debate: False Starts,
Real Foundations, Cambridge/Mass.
Hill, P.B./Schnell, R. (1990): Was ist "‘Identität"’ ?; in: Esser,H./Friedrichs,J.
(Hrsg.): Generation und Identität, Opladen, p.25-42.
Hintzman, D.L. (1990): Human Learning and Memory: Connections and
Dissociations; in: Annual Review of Psychology, p.109-139.
Hummon, N.P. (1990): Organizational Structures and Network Processes;
in: Journal of Mathematical Sociology, 15, 2, p.149-161.
Johnson-Laird,P. (1988): The Computer and the Mind, Cambridge, Mass.
Kalter, F. (1990): Dynamische Prozesse der öffentlichen Meinung, unpub-
lished manuscript, Institut für angewandte Sozialforschung, Universität zu
Köln, 1990.
Kauffman,S.A. (1984): Emergent Properties in Random Complex Automata;
in: Physica 10D, p.145-156.
Lindenberg, S. (1989a): Toward the Construction of Interdisciplinary Theo-
retical Models to Explain Demographic Behavior; A Comment, Unpublished

6
Manuscript, Groningen.
Lindenberg, S. (1989b): Choice and Culture: The Behavioral Basis of Cul-
tural Impact on transactions; in: Haferkamp,H. (ed.): Social Structure and
Culture, Berlin/New York, p.175-200.
MacLennan, B. (1990a): Evolution of Communications in a Population of
Simple Machines, Technical Report, Computer Science Department, Univer-
sity of Tennessee, Knoxville.
MacLennan, B. (1990b): Synthetic Ethology: An Approach to the Study of
Communication, Technical Report, Computer Science Department, Univer-
sity of Tennessee, Knoxville.
Minsky, M. (1987): The Society of Mind, London.
Nowak,A., Szamrej,J., Latane,B. (1990): From Private attitude to public
opinion: A Dynamic Theory of Social Impact; in: Psychological Review, 97,
1990, 3, p.362-376.
Opp, K.-D. (1991): Processes of Collective Political Action. A Dynamic
Model And The Results Of A Computer Simulation; in: Rationality And
Society, 3, 2, p.215-251.
Poundstone, W. (1987): The Recursive Universe. Cosmic Complexity and
the Limits of Scientific Knowledge, Oxford.
Pylyshyn, Z.W. (eds.) (1987): The Robot’s Dilemma: The Frame Problem
in Artificial Intelligence, Norwood.
Pylyshyn, Zenon W. (1989): Computing in Cognitive Science; in: Posner,
M.I. (ed.): Foundations of Cognitive Science, Cambridge (Mass.), 1989, p.51-
91.
Sakoda,J.M. (1971): The Checkerboard Model of Social Interaction; in: Jour-
nal of Mathematical Sociology, 1, 1971, p.119-132.
Schelling, T.C. (1971): Dynamic Models of Segregation, in: Journal of Math-
ematical Sociology, 1, p.143-186
Schnell, R. (1990): Computersimulation und Theoriebildung in den Sozial-
wissenschaften; in: Kölner Zeitschrift für Soziologie und Sozialpsychologie,
42, 1, p.109-128.
Schnell, R. (1989): Möglichkeiten der Nutzung von BITNET in den Sozial-
wissenschaften; in: ZA-Information 24, 1989, p.101-115.
Schrodt, P.A. (1981): Conflict as a determinant of territory; in: Behavioral
Science, 26, 1981, p.37-50.
Schrodt, P.A. (1985): Adaptive Precedent-Based Logic and Rational Choice:
A Comparison of Two Approaches to the Modeling of International Behavior,
in: U.Luterbacher und M. D. Ward (eds.): Dynamic Models of International

7
Conflict, Boulder/Colorado, p.373-400.
Simon, H.A., Kaplan, C.A. (1989): Foundations of Cognitive Science; in:
Posner, M.I. (ed.): Foundations of Cognitive Science, Cambridge (Mass.),
p.1-47.
Sylvan, D.J., Glassner,B. (1985): A Rationalist Methodology for the Social
Sciences, Oxford.
Toffoli,T., Margolus,N. (1987): Cellular Automata Machines, London.
Wolf, C. (1991): Unpublished dissertation manuscript, University of Cologne.
[now published as: Gleich und gleich gesellt sich: individuelle und struk-
turelle Einflüsse auf die Entstehung von Freundschaften, Hamburg, Kovac,
1996]

You might also like