[Paper Review] Normative Equivalence in Human-AI Cooperation: Behaviour, Not Identity, Drives Cooperation in Mixed-Agent Groups

Nico Mutzner, Taha Yasseri+1 more|arXiv (Cornell University)|Jan 28, 2026

Psychology of Moral and Emotional Judgment1 citations

TL;DR

The study shows that in a four-player Public Goods Game, cooperation dynamics are driven by group behavior and normative expectations, not by whether a partner is labeled human or AI, indicating normative equivalence in mixed human–AI groups.

ABSTRACT

The introduction of artificial intelligence (AI) agents into human group settings raises essential questions about how these novel participants influence cooperative social norms. While previous studies on human-AI cooperation have primarily focused on dyadic interactions, little is known about how integrating AI agents affects the emergence and maintenance of cooperative norms in small groups. This study addresses this gap through an online experiment using a repeated four-player Public Goods Game (PGG). Each group consisted of three human participants and one bot, which was framed either as human or AI and followed one of three predefined decision strategies: unconditional cooperation, conditional cooperation, or free-riding. In our sample of 236 participants, we found that reciprocal group dynamics and behavioural inertia primarily drove cooperation. These normative mechanisms operated identically across conditions, resulting in cooperation levels that did not differ significantly between human and AI labels. Furthermore, we found no evidence of differences in norm persistence in a follow-up Prisoner's Dilemma, or in participants' normative perceptions. Participants' behaviour followed the same normative logic across human and AI conditions, indicating that cooperation depended on group behaviour rather than partner identity. This supports a pattern of normative equivalence, in which the mechanisms that sustain cooperation function similarly in mixed human-AI and all human groups. These findings suggest that cooperative norms are flexible enough to extend to artificial agents, blurring the boundary between humans and AI in collective decision-making.

Motivation & Objective

• Explore whether cooperative norms in small groups change when one member is AI-labeled compared to all-human groups.
• Examine how bot strategy (unconditional cooperator, conditional cooperator, free-rider) affects group cooperation and norm formation.
• Assess norm persistence beyond the group context using a one-shot Prisoner’s Dilemma and norm elicitation measures.

Proposed method

• Between-subjects 2x3 design with four-player groups (three humans + one bot) across label (human vs AI) and bot strategy (Unconditional Cooperator, Conditional Cooperator, Free-Rider).
• Ten rounds of a linear Public Goods Game followed by a one-shot Prisoner’s Dilemma.
• Bots followed predefined strategies without disclosure to participants; contributions transformed 1.5x, equally redistributed.
• Norm elicitation tasks measured social appropriateness, empirical norms, and injunctive norms.
• Pre-registration and robustness checks included; sample size finalized at 236 completing PGG.

Figure 1 : Schematic representation of the experimental design and PGG mechanics

Experimental results

Research questions

• RQ1 Do AI-labeled bots reduce cooperation or normative influence compared to human-labeled bots in a mixed-group setting?
• RQ2 How do bot strategies (unconditional, conditional, free-rider) shape cooperation and norm formation in groups?
• RQ3 Is norm persistence in a subsequent one-shot Prisoner’s Dilemma affected by the agent’s label or strategy?
• RQ4 Do participants’ normative perceptions (social appropriateness, empirical/injunctive norms) differ by agent label?

Key findings

• Cooperation levels were similar across human and AI labels; AI labeling did not significantly reduce contributions.
• Bot strategy effects on average cooperation were small and not statistically significant.
• Normative mechanisms (conditional cooperation, responsiveness to others, inertia) governed cooperation in both treatments.
• Norm persistence in the one-shot PD did not differ by label or bot strategy; prior group contributions predicted PD cooperation in both conditions.
• Post-task norm perceptions were aligned across treatments, with no meaningful differences in social appropriateness or norm expectations between human and AI groups.
• Trust predicted higher contributions overall; in the AI condition, trust and normative pressure increased cooperation, but AI acceptance predicted lower contributions, suggesting algorithm aversion effects.

Figure 2 : Mean Contribution per Round by Treatment and Bot Strategy.