Reproduce “The Convention Gap: Beyond Literal Content in Human-AI Cooperation” 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
Makoto Fukushima, Hua-Dong Xiong, Ehsan Moradi Pari
The authors introduce a Hanabi diagnostic that subtracts observed failure rates from exact literal-risk predictions based on hints and visible cards. The gap is strongly positive in human-human play (+26.2 percentage points), while it is near zero in simulated AI-AI play (−0.64 percentage points). Human-AI gaps differ by partner, but the measure does not establish why players acted.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.