A dynamic over games drives selfish agents to win-win outcomes

TL;DR

This paper introduces a framework for modeling how agents evolve their interaction contexts, leading to the spontaneous emergence of fairness and cooperation among selfish agents through institutional evolution.

Contribution

It presents a novel formalism for modeling institutional evolution, contrasting it with behavioral evolution, and demonstrates its role in fostering cooperation among selfish agents.

Findings

Self-interested agents over-represent fairness by 100% in attractor games.

Fairness co-occurs with self-serving features, influencing agent preferences.

Institutional evolution encourages cooperation among selfish agents.

Abstract

Understanding the evolution of human social systems requires flexible formalisms for the emergence of institutions. Although game theory is normally used to model interactions individually, larger spaces of games can be helpful for modeling how interactions change. We introduce a framework for modeling "institutional evolution," how individuals change the games they are placed in. We contrast this with the more familiar within-game "behavioral evolution". Starting from an initial game, agents trace trajectories through game space by repeatedly navigating to more preferable games until they converge on attractor games that are preferred to all others. Agents choose between games on the basis of their "institutional preferences," which define between-game comparisons in terms of game-level features such as stability, fairness, and efficiency. Computing institutional change trajectories over the two-player space, we find that the attractors of self-interested economic agents over-represent fairness by 100% relative to baseline, even though those agents are indifferent to fairness. This seems to occur because fairness, as a game feature, co-occurs with the self-serving features these agents do prefer. We thus present institutional evolution as a mechanism for encouraging the spontaneous emergence of cooperation among inherently selfish agents. We then extend these findings beyond two players, and to two other types of evolutionary agent: the relative fitness maximizing agent of evolutionary game theory (who maximizes inequality), and the relative group fitness maximizing agent of multi-level/group selection theory (who minimizes inequality). This work provides a flexible, testable formalism for modeling the interdependencies of behavioral and institutional evolutionary processes.