Exploring the Human-AI Multilateral Bargaining

This study investigates human–AI disparities in a 3-player majoritarian bargaining game. Participants negotiate payoff splits in real time with other human players, but some positions are labeled as “AI” agents. We implement two treatments that differ in how these AI-labeled positions are described: in Treatment A, participants are told they are interacting with AI agents; in Treatment B, they are told that certain human players have delegated their decision-making to AI agents. We test whether participants systematically favor “Human” over “AI” labels under these two framings, resulting in differences in final outcomes, coalition choices, and proposal strategies. Post-game surveys measure perceptions of AI trustworthiness, cooperativeness, predictability, and strategic intent. A pilot for Treatment A established a measurable outcome difference, whereas a pilot for Treatment B showed a weaker discriminative outcome. Our power analysis is based on effect-size estimates from these two pilots.