|
Abstract:
|
"We introduce a framework for modeling how individuals change the games they are placed in, a process we term institutional evolution, in contrast with 'within-game' behavioral evolution. Starting at random locations in an abstract game space, agents trace trajectories through the space by repeatedly choosing between 'neighboring' games, until they converge on 'attractor' games that they prefer to all others. The properties of attractor games depend on the specific features that agents use to define their preferences between games. We characterize the attractors of institutional evolutionary over three types of game theoretic agent: the absolute fitness maximizing agent of economic game theory, the relative fitness maximizing agent of evolutionary game theory, and the relative group fitness maximizing agent of multi-level/group selection theory. Computing institutional change trajectories over the space of two-player ranked-outcome games, we find that the institutional evolutionary process leads to very different attractors depending on the agent. While 'win-win' games account for 25% of all games in the space, this proportion is 50%, 0%, or 100% in the attractors, depending on whether agents are economic, evolutionary, or serving stable sub-groups. The first result is especially interesting: although economic agents are indifferent to the fairness of the games they choose between, the games they prefer tend incidentally to be two times more fair than baseline, as a side-effect of how preferred features co-occur. We thus present institutional evolution as a mechanism for encouraging the spontaneous emergence of cooperation among inherently selfish agents. We then investigate the sensitivity of these findings to behavioral contexts, and to games of more than two players. This work provides a flexible, testable formalism for capturing institutional evolutionary process, and for modeling the interdependencies of institutional and behavioral (between- and within- game) evolutionary processes."
|