In lieu of an abstract, here is a brief excerpt of the content:

chapter 3 New Tools The problem of showing how large-scale social patterns emerge from individual decisions is central to social science. The first complete instance was Adam Smith’s description of the way individual-level efficiency-seeking behavior leads to large-scale efficiency in the division of labor. But Smith’s analysis was facilitated by the fact that economic outcomes are denominated with money. Money has been established and enforced to objectivize economic preferences, and this objectivization in turn permits this kind of emergence. The problem of building a comparable analysis in realms in which there is no such cultural device has been far more difficult to conceptualize. But now we can do it. Strong growth in the general area known as human complex systems has produced several new families of computer models that can reproduce large-scale emergent phenomena. These are primarily agent-based models or multiagent models. The main units of such programs are agents, representing individuals or groups. What they do in computer simulations is make decisions and take action. They can be made to follow decision rules of almost any kind so long as they amount to something like “if x then do y, else do z” or “under condition x do y, else do z.” The program then recalculates the position of the agent that made it and others affected before going on to the next agent and in this way can cycle through any number of agents any number of times, accumulating the interactive effects of all the decisions by whatever measures seem relevant. With agent-based models providing a very general method for the computational side of the problem of getting from individual action to aggregate patterns, what remains is the conceptual side. Where do the parameters for the models come from? The parameters are of two main kinds: the constraints and resources that the agents are assigned and the rules for the decisions that the agents make with them. The constraints, in general, are assigned in one of three ways. They are purely hypothetical, invented by the analyst on the basis of generalized knowledge or a theory; they are estimates assigned to the problem on the basis of actual available data; or the agents can represent an actual population, and the constraints  human organizations and social theory can be the actual constraints they face and the actual resources they have. Similarly , the decision rules may be hypothetical (“what if”). They may be derived from one or more theoretical models such as game theory, microeconomics, or a simple idea such as minimizing travel distance. Or they may be the decision rules that the people in a community actually report as what they follow. Thus far, no social theory has explained such parameters systematically. This is what will be done here. Plainly put, I take resources and constraints as resources reported and confirmed ethnographically, and given this, I show that the decision parameters come from their cultural idea systems by way of the organizations formed on the basis of those idea systems. Ideas closely associated in cultural idea systems are closely associated as parameters in individual decision making. They are instantiated by being used to define the mutual adjustments of behavior in the context of the organizational purpose (or purposes). Even without such theory, however, agent-based models have provided dramatic reaffirmation of the general point that we do not have to have conscious collective control to have coherent collective action, precisely as Smith showed for the division of labor. A reliable feature of computers is that while they can “think” in the sense of follow rules, they have absolutely no imagination. They cannot make inferential leaps. The rules they follow must be mutually consistent and have no gaps. Otherwise, the computer either stops or produces garbage. This kind of consistency and coherence is one aspect of rationality. Another involves comparison. Rational people make comparisons between alternatives and choose the one that best meets some criterion. Finally, rationality also generally assumes some relation of means to ends. All of these are readily simulated, and it is usually easy enough to see if the program as written actually represents the action being modeled. There is no necessity for a black-box effect in which one is not sure that what the program is doing actually corresponds to what one is interested in. Properly designed computer simulations based on field observations allow absolutely clear and firm chains of evidence between those observations and analytic conclusions...

Share