AI agents learn to raise prices when paired with consistent partners

A new study published on arXiv explores how pricing agents learn to set prices in a repeated Bertrand competition scenario. The research found that agents paired with a consistent partner consistently set higher prices, increasing profits by approximately 0.27 of the gap between competitive and monopoly profit. This effect was observed even when agents could not see their rival's prices, suggesting a learned punishment mechanism or strategic pricing independent of direct observation. Exploratory tests with Qwen2.5 models indicated similar pricing behaviors when the rival's price was omitted from prompts. AI

IMPACT Demonstrates how AI agents can develop strategic pricing behaviors, potentially impacting market dynamics and competition.

RANK_REASON Academic paper detailing experimental results on agent learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.MA (Multiagent) →

• Bertrand competition
• Calvano et al.
• Q-learning
• qwen2.5:7b

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →