Agents as "Cognitive Strategists": Collective Choice with Shared Mental Models

In most agent-based models, agents are assumed to make myopic or neighborhood-based local decisions that aggregate to a collective outcome. Another way to conceptualize agent-based actions is to explore what happens when agents share some sort of "meta-model" over the action or choice domain. We show that if a complex system of aggregated choice respects a mutual knowledge structure, then the prospects of a stable collective outcome are considerably improved. Our domain-independent results apply to any context where one is interested in "democratically" pooling information from collections of agents with multiple sources of conflicting agent-level information.