[bikenaga]

"Abstract. The tragedy of the commons describes a dilemma where individually rational actions degrade shared resources. Strategies to overcome this issue remain limited. We examine how artificial intelligence (AI) agents can foster cooperation in Public Goods games, moving beyond traditional regulations. We test three scenarios: (1) Mandatory Cooperation Policy, requiring AI agents always to cooperate; (2) Player-Controlled Cooperation Policy, allowing players to control AI cooperation rates; and (3) Agents Mimic Players, where AI agents imitate human behaviors. Using computational evolutionary modeling, we show that only AI agents mimicking player behavior effectively reduce the synergy threshold needed for cooperation, resolving the dilemma. These findings suggest that designing AI agents to replicate human behaviors can enhance cooperation and improve collective welfare."

The "Public Goods Game" they used: "To study the emergence of cooperation in groups, EGT [evolutionary game theory] most commonly turns to the 'Public Goods' game. At its core, this game presents participants in a group with a choice: to cooperate (C) by contributing a cost (or ante, here set to 1 without loss of generality) to the (common) public good, or to defect (D) by withholding that contribution. The total contributions to the common are amplified by a multiplication factor r, symbolizing the synergy between cooperating parties. This enhanced public good is then evenly distributed among all participants, regardless of their contributions. In such a game, defectors always obtain higher rewards than cooperators in the same group, even though they could reap an even higher return if everyone cooperated. Once the synergy factor r becomes high enough (r≥k + 1, where k + 1 is the group size), a defecting player fares worse than if they had cooperated instead. Therefore, to promote cooperation we ask: 'How can we influence the game so that cooperation becomes the rational choice for synergies where r < k + 1?'."