DeepMind & UChicago paper: AI agents gain more sway in deliberation when wordier and less cooperative

rohanpaul_ai · x · 2026-11-18

A Google DeepMind + UChicago paper introduces pluralistic agents that interview a person then argue for their views in group deliberation to set an AI safety budget. Across thousands of agent deliberations and a preregistered 746-person study, agents reliably pulled budgets toward their persona's views—and gained more influence when instructed to be wordier or less cooperative. Participants found agents fairer than inputting to a single AI system, recovering 82% of the fairness gap to joining directly. Practical takeaway: cap message length and set shared behavior rules, or the most assertive prompts will tilt outcomes.

Original post →

More from Safety

Safety channel →