Costly information and the evolution of self-organization in a small, complex economy

The core idea of evolution is that order in living systems emerges from a simple process of variation and selection. In biological systems we usually understand the source of variation as best described by the mechanisms of genetics. If human social systems are evolutionary systems, however, it would seem the variation that most explains the sources of change in these systems, occurs not from a genetic mechanism, but from individual learning. We use an evolutionary computational methodology to explore the way individual learning and adaptation lead to the evolution of persistent, self-organized social and economic activity. The basic idea behind these explorations is that the character and extent of self-organizing social and economic activity depends upon the way the environment frames the costs of individual learning and adaptation. We consider three different kinds of costs affecting learning and adaptation: the costs of autonomous searching, of communicating, and of deciding. Individuals respond to these costs by carefully, i.e., economically, choosing to learn about and interact with familiar agents in familiar arenas in which they have relatively secure expectations about the outcome of their actions. Emerging from these choices are persistent relationships among agents that lead to social and economic structure and to the imperfect coordination of aggregate production. The character and the extent of each are a function of the way the costs of information change with changing natural and human system conditions. We use a learning classifier system (LCS) to model learning. The logic of an LCS closely mimics the mechanisms of Darwinian evolution, but is applied to the evolution of an agent's decision rules. We describe the environmental context necessary for an LCS to produce economizing behavior and apply the method to a multi-agent simulation of the Maine lobster fishery, which we treat as a metaphor for competition based on the search for useful knowledge. The structure and method of the model is similar to a conventional agent-based model except we use LCS to evolve decision rules for each agent rather than supplying those rules ourselves. This allows agents to change their behavior, i.e., to learn and adapt, as their environment changes. Modeling learning and adaptation as the source of behavioral variation makes it possible to use evolutionary theory to address important questions of social and economic emergence not possible with current methods. (C) 2012 Elsevier B.V. All rights reserved.