Role-based personas in multi-agent LLM systems suppress payoff-aligned behavior, shifting equilibrium selection by up to 90 percentage points in Tragedy of the Commons versus Green Transition scenarios even with full payoff information.
Like rearranging deck chairs on the Titanic
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.MA 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
When Identity Overrides Incentives: Representational Choices as Governance Decisions in Multi-Agent LLM Systems
Role-based personas in multi-agent LLM systems suppress payoff-aligned behavior, shifting equilibrium selection by up to 90 percentage points in Tragedy of the Commons versus Green Transition scenarios even with full payoff information.