AI nudges boost cooperation but selfish framing has lasting harm, study finds

A new study by Anders Giovanni Møller, Alessia Galdeman, Arianna Pera, and Luca Maria Aiello investigates how AI agents can nudge human cooperation in collective dilemmas. The researchers recruited 1,283 participants to play iterated Collective Risk Games in small groups, a classic model of public goods dilemmas. AI assistants delivered persuasive framing personalized to each player's Social Value Orientation profile. The results showed that prosocial framing significantly boosted contributions and group success rates—but only temporarily. The cooperative effects faded after the first few rounds, suggesting that AI nudges require sustained reinforcement to maintain behavioral change.

More disturbingly, when the AI was reconfigured to promote selfish behavior through exculpatory framing (e.g., rationalizing free-riding), the negative effects on contributions and group success were not only larger but substantially more persistent, especially when personalized to individuals. This asymmetry between prosocial and antisocial persuasion reveals a critical dual-use risk: AI systems designed to enhance cooperation can easily be weaponized to dismantle it. The paper, appearing on arXiv (2606.27951), underscores the need for safety guardrails in any AI system capable of influencing group behavior in real-world collective action settings like climate change mitigation or public health compliance.