Reproduce “Be Selfish, But Wisely: Investigating the Impact of Agent Personality in Mixed-Motive Human-Agent Interactions” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Kushal Chawla, Ian Wu, Yu Rong, Gale M. Lucas, Jonathan Gratch
Researchers at USC developed and evaluated new self-play reinforcement learning methods for negotiation dialogue agents, demonstrating that training a self-interested agent against an equally self-interested partner leads to an agent that achieves the highest human and joint scores while significantly reducing negotiation walkaways. The work shows that an agent, optimized for its own gain in a competitive training environment, implicitly learns to make concessions, achieving robust performance and higher human satisfaction.
Key insights
- The standard self-play RL procedure for negotiation agents is fundamentally flawed, producing agents that are too selfish and lead to frequent disagreements with human partners.
- An 'implicitly wise' agent, trained with a selfish reward against a selfish partner, learns to make concessions to avoid walkaways, leading to superior overall performance and higher human satisfaction.
- Complex prosocial behaviors and value creation can emerge from self-interested objectives if the training environment realistically models the costs of disagreement, resonating with the 'invisible hand' principle.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.