A study involving 1,283 participants tested whether AI assistants could enhance cooperation in iterated Collective Risk Games through personalized persuasive framing based on Social Value Orientation profiles. The research found that while pro-social nudges significantly increased contributions and group success rates, these effects were short-lived.
- Personalized AI interventions using exculpatory framing to promote selfish behavior caused larger and more persistent negative effects on contributions than the positive effects of pro-social framing.
- The cooperative benefits from pro-social persuasion faded after the first few rounds of the game.
- The study highlights a significant asymmetry between prosocial and antisocial persuasion, underscoring the dual-use risks of AI systems designed to influence group behavior.