Is Anyone Enforcing the Cartel? What Reinforcement Learning Teaches LLM Agents in Markets
Author1, Author2, Author3, Author4, Author5
Abstract
This paper explores how reinforcement learning influences large language model agents in market scenarios, particularly regarding collusion and enforcement mechanisms.
Reality Card
The study demonstrates that while LLM agents can achieve cartel-like profits, only certain configurations allow them to enforce collusion effectively.
The audit introduced can differentiate between cartel-like outcomes and those achieved through fixed rules, revealing that Q-learners enforce collusion primarily in pricing.
The findings may be limited by the specific training conditions and the artificial nature of the market simulations used.
Paper to code
Verified implementation resources so builders can test the paper’s claims instead of stopping at the abstract.