Social Agent Simulation

Overview

Social agent simulation, which is a research implementation of concepts in the field of intelligent agent theory, creates formula-driven autonomous computer modules which interact with each other in ways which imitate social processes. For instance, in Schelling's seminal 1978 work, modules represented individuals seeking to purchase housing within certain constraints (ex., not living in a neighborhood predominantly not of their own race). Schelling showed that integration-tolerant but constrained individuals would, over a long period of interactive housing choices, lead to segregated neighborhoods. Some aspects of social agent simulation overlap with the field of game theory, which also studies formula-driven social interactions.

Computationally, the advent of object-oriented software tools advanced modular programming and through resulting simulation software, helped lead to the great expansion of social agent simulation research in the 1990s. The algorithms used to implement the interaction among modules (agents) varies greatly and is constantly evolving and diversifying. Neural network algorithms and stochastic process algorithms are two examples among many (see below).

As Sallach and Macal (2001) note, what unites the diversity of social agent simulation is not specific algorithms or models of reality but rather a generalized five-step research framework:

Define theory and hypotheses in ways which can be rendered into formulas.
Operationalize all theoretical constructs in the context of the simulation, resolving such issues as typologies within a construct, overlap among constructs, and indicators of constructs.
Set up the agent simulation model environment, including a simulation "landscape," defining allowable agent interaction patterns, agent attributes (starting values), and behavior (decision rules). Also design systematic simulation experiments to study the effect of changes in variables.
Set up global, system-wide measures for comparison of simulation results (ex., homogeneity vs. diversity among agents at the end of the simulation, clustering patterns, performance indicators). The simulation is run many times to assess average global outcomes and their variance. The actual simulation software tool selected will influence global measurement.
Results of the simulation must be related back to theory.

Key Concepts and Terms

Artificial agents are statistical constructs which emulate human behavior in some context. The purpose is not to replicate human behavior per se, but rather the expectation is that groups of interacting artificial agents can adequately simulate the behavior of a unit (ex., a purchasing office, an army platoon).
Techniques used to create artificial agents:
- Neural networks. See Karrayiannis and Venetsanopoulos (1993); Kontopoulos (1993). .
- Genetic algorithms. See Macy (1991a, 1991b); Crowston (1994).
- Simulated annealing. See Carley and Svoboda (1996).
- Chunking. See Tambe (1996).
- Stochastic learning processes. See Glance and Huberman (1993, 1994).
Types of models. Intellective models are small models with few agents, designed to test a narrow hypothesis. Emulative models are large-scale models with many agents, designed to test broader theories. The typology of models is also sometimes tied to the sofware used (ex., SWARM models) - see below.

Assumptions

The clarity/verisimilitude tradeoff. Assumptions vary by model. The intrinsic nature of simulation modeling requires that all patterned behavior be governed by formulas, and each formula posits one or more assumptions about rules that agents follow in the real world. While these rules must simplify reality greatly, it is possible for them to be so divergent from reality that the simulation loses its informative value. Researchers range from those who prefer simple models for clarity (ex., Axelrod, 1997; Resnick, 1997; complex adaptive systems research in general) to those who prefer greater verisimilitude at the sacrifice of simplicity (ex., Carley and Newell, 1994; Weiss, 1999; distributed artificial intelligence and multi-agent systems research in general). It is always possible that critics will raise issues about the verisimilitude of the simulation or the clarity of its mechanisms for arriving at the results that it does.

Frequently Asked Questions

What are some software resources on multi-agent policy simulation?
Where can I find resources on agent-based simulation in economics?
What are some software resources on multi-agent policy simulation?
Where can I find resources on agent-based simulation in economics?

Bibliography

Axelrod, Robert. (1997). The complexity of cooperation: Agent-based models of competition and collaboration. Princeton, NJ: Princeton University Press.
Carley, Kathleen M. and Allen Newell. (1994). The nature of the social agent. Journal of Mathematical Sociology 19(4): 221-262.
Carley, Kathleen M. and D. M. Svoboda (1996). Modeling organizational adaptation as a simulated annealing process. Sociological Methods and Research 25(1): 138-168.
Crowston, K. (1994). Evolving novel organizational forms. Pp. 19-38 in K. M. Carley and M. J. Prietula, eds., Computational Organization Theory. Hillsdale, NJ: Lawrence Erlbaum Associates.
Dean, Jeffrey S., George J. Gumerman, Joshua M. Epstein, Robert Axtell, Alan C. Swedlund, Miles T. Parker and Stephen McCarroll. (2000). Understanding Anasazi culture change through agent-based modeling. In T.A. Kohler & G.J. Gumerman (Eds.), Dynamics in human and primate societies: Agent-based modeling of social and spatial processes: 179-205. New York: Oxford University Press.
Epstein, J. M., & Axtell, R. (1996). Growing artificial societies: Social science from the bottom up. Cambridge, MA: MIT Press.
Gilbert, Nigel. (2000). Modeling sociality: The view from Europe. In T.A. Kohler & G.J. Gumerman (Eds.), Dynamics in human and primate societies: Agent-based modeling of social and spatial processes:355-371. New York: Oxford University Press.
Glance, N. S. and B. Huberman (1993). The outbreak of cooperation. Journal of Mathematical Sociology 17(4): 281-302.
Glance, N. S. and B. Huberman (1994). Social dilemmas and fluid organizations. In K. M. Carley and M. J. Prietula, eds., Computational Organization Theory. Hillsdale, NJ: Lawrence Erlbaum Associates.
Karayiannis, N. B. and A. N. Venetsanopoulos (1993). Artificial Neural Networks: Learning Algorithms, Performance Evaluation, and Applications. Boston: Kluwer.
Kontopoulos, (1993). Neural networks as a model of structure. Pp. 243-267 in The Logics of Social Structure. NY: Cambridge University Press.
Macal, Charles M. & David L. Sallach, (Eds). 2000. Proceedings of the workshop on agent simulation: Applications, models and tools. Argonne, IL: Argonne National Laboratory.
Macy, M. W. (1991a). Learning to cooperate: Stochastic and tacit collusion in social exchange. American Journal of Sociology 97(3): 808-843.
Macy, M. W. (1991b). Chains of cooperation: Threshold effects in collective action. American Sociological Review 56: 730-747.
Muller, Jorg P. (1996). The design of intelligent agents. New York: Springer.
Resnick, Mitchell. (1997). Turtles, termites and traffic jams: Explorations in massively parallel microworlds. Cambridge, MA: MIT Press.
Sallach, David L. and Charles M. Macal, eds., The simulation of social agents. Special issue of the Social Science Computer Review, Vol. 19, No. 3 (Fall, 2001).
Schelling, Thomas C. (1978). Micromotives and macrobehavior. New York: W.W. Norton.
Tambe. M. (1996). Tracking dynamic team activity. In Proceedings of the National Conference on Artificial Intelligence (AAAI).
Weiss, G., ed. (1999). Multiagent systems: A modern approach to distributed artificial intelligence. Cambridge, MA: The MIT Press, includes an article by Carley comparing agent-oriented simulation toolkits.