Reproduce “Model-free conventions in multi-agent reinforcement learning with heterogeneous preferences” 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

Raphael Köster, Kevin R. McKee, Richard Everett, Laura Weidinger, William S. Isaac, Edward Hughes, Edgar A. Duéñez-Guzmán, Thore Graepel and 2 more

The authors test whether model-free agents form conventions through interaction in a simulated berry world. Agents with different color preferences learn from experience while harvesting and replanting. In one long-run run, an 11-agent group shifts berries toward red, and agents from other groups later replant red. Free riding limits coordination; the result is simulated, not evidence about human conventions.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.