Internal vs. External: Comparing Deliberation and Evolution for Multi-Agent Constitutional Design

2026-05-09

AI Summary

This study addresses the unresolved question of whether behavioral charters in multi-agent systems should be generated through internal negotiation or external optimization. For the first time, it conducts a controlled comparison of two mechanisms—endogenous generation via negotiation protocols versus exogenous optimization using evolutionary algorithms—across three distinct social environments: a coordination gridworld, a repeated public goods game, and a bilateral trading market. The results reveal that evolutionary methods significantly outperform negotiation in collective action settings (p<0.01), yet when the incentive multiplier is set to 0.75, enforced cooperation under evolution paradoxically reduces overall welfare. In contrast, negotiation-based mechanisms consistently fail to spontaneously devise punishment strategies. These findings highlight a performance reversal between the two approaches under varying incentive structures and uncover a fundamental disparity in their capacity to discover effective punishment mechanisms.

Abstract

Multi-agent AI systems need behavioral constitutions, but it is unresolved whether such rules should emerge internally through agent self-governance or be discovered externally through optimization. We present the first controlled comparison of internal deliberation and external evolution across three social environments: a coordination grid-world, an iterated public goods game, and a bilateral trading market. Across 180 simulation runs, evolution significantly outperforms deliberation in collective-action settings (p < 0.01), while neither method improves outcomes in bilateral trading. A multiplier ablation reveals that evolution's advantage inverts when incentives shift: at pool multiplier (m = 0.75) the evolved constitution forces value-destroying cooperation and becomes the worst-performing method. Notably, no deliberation run across thirty trials ever proposed punishment -- the canonical cooperation-sustaining mechanism evolution reliably discovers -- suggesting external optimization wins on peaks while internal self-governance trades peaks for structural responsiveness.

Problem

• multi-agent systems
• constitutional design
• internal deliberation
• external evolution
• collective action

Innovation

• multi-agent constitutional design
• internal deliberation
• external evolution
• collective action
• punishment mechanism