What is ABM?
An agent-based model (ABM) is a computational model for simulating the behavior of interacting agents. ABMs allow us implement a theory of the individual (with the characteristics below) and then experiment, with computer simulation, on the interactions of individuals. The patterns of interaction we observe are often emergent.
Characteristics of ABMs
Discrete. An agent is a self-contained individual with identifiable boundaries.
Situated. An agent exists in and interacts with an environment that generally includes other agents and may include other (nonagent) resources, dangers, and so forth.
Embodied. An agent may be embodied (robotic) or a purely software-simulated entity; the latter is more common.
Active. An agent not only is affected by the environment but also is assumed to have a behavioral repertoire that it can use proactively.
Limited information. An agent is usually assumed not to be omniscient but to be able to gather information only from its own local environment—for example, agents can see only their neighboring agents (not all agents) and only their behaviors (not their internal states, goals, etc.).
Autonomous goals. An agent has its own internal goals and is self-directed in choosing behaviors to pursue those goals, rather than being simply a pawn under the command of some centralized authority.
Bounded rationality. Agents ordinarily are assumed to gather information and generate behaviors by the use of relatively simple rules, rather than being capable of extensive computations such as maximizing expected utility.
Adaptation. Some models assume that agents use fixed, prespecified rules to generate their behavior; others use agents that can learn or adapt, changing their rules based on experience. (Smith and Conrey, 2007).
Explanation. Roughly speaking, there are two types of explanation: (1) law-like and/or statistical and (2) mechanistic. Examples of the first are Einstein’s explanation of the relationship between energy and mass (E = mc2) or correlation or regression associations among variables. This is in contrast to mechanistic and generative explanations that aim to explain behavior or psychological phenomena in terms of the interactions of the components of the mechanism.
Let’s consider an example: Attractive people tend to pair with others of similar attractiveness with a correlation, r, in the .5 to .6 range (e.g., Critelli, J. W., & Waid, L. R. (1980). Physical attractiveness, romantic love, and equity restoration in dating relationships. Journal of Personality Assessment, 44: 624-629.). This suggests that people prefer others with the same or similar degree of attractiveness. This is a statistical explanation.
The problem is that people typically prefer attractive people and not simple people of the same degree of attractiveness (Curran, J. P., & Lippold, S. (1975). The effects of physical attraction and attitude similarity on attraction in dating dyads. Journal of Personality, 43: 528-539.)
Kalick and Hamilton (1986, The matching hypothesis reexamined. Journal of Personality and Social Psychology, 51, 673-682) an ABM to resolve this paradox.
Agents were males and females that could use either of two simple rules:
1) Similarity matching rule: Choose another with similar attractivity.
2) Attractiveness-seeking rule: Choose another that is the most attractive (i.e., a 10 is chosen with probability 1.0, a 9 with probability .9, etc.).
Simulation experiment procedure: 1000 agents are created, 500 female, 500 male. Each agent is assigned an attractiveness of 10 to 1 randomly. At each time step, a male and female were randomly paired. If both offered to date based on their decision rules, they were removed from the population. The simulation continued until all males and females were paired.
Results: In those simulations in which rule 1) was used, the correlation between attractiveness in pairs was too high, in the .8 and .9 range. For rule 2), however, the correlation was as predicted even though agents did not seek similar attractiveness. Why?
Discussion: This is an example of mechanistic explanation (one has to understand how the process works to explain the results) and is emergent (i.e., not the “sum” of individual preferences in this case).
Another example: How would you explain the Stanford Prison Experiment?
Emergence. Any phenomenon that is not merely the sum of the properties of its parts. An emergent phenomenon, however, is explained by the interaction of its parts with each other and the environment. An important implication of emergence is that although emergent properties are explained by the interaction of parts, just knowing the properties of the parts does not explain or predict what will emerge. Emergent phenomena have mechanistic and generative explanations and, I will claim, cannot be explained by law-like or statistical explanations.
strong>Emergence is Ubiquitous:
Fall Web Worms (video)
Slime Mold (Dictyostelium Discoideum)
ABM provides just the framework we need to explain and understand emergent behavior.
Another example of emergent behavior from simple rules.
As we will see later, emergent patterns of behavior can even result from random rules of behavior.
Analysis using ABMs: Analysis using ABMs is different that mathematical or statistical analysis, but it is similar to experimental analysis. In an experiment with individuals, we set up a situation in which the individuals interact and manipulation one or more variables. We then make observations of what happens. In ABM, we can do the same thing except instead of running a few experiments, we can run thousands or even millions and systematically change all of the parameters. We can then determine what parameters and what rules produce the same or similar results as the actual phenomena we observe. Once we know the rules and the parameters of the situation in which individuals are interacting, we can the explain the behavior in terms of the mechanism (i.e., the interactions among individuals) and how this generates the behavior we observe.