Reproduce “Disentangling Abstraction from Statistical Pattern Matching in Human and Machine Learning” 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

Sreejan Kumar, Ishita Dasgupta, Nathaniel D. Daw, Jonathan D. Cohen, Thomas L. Griffiths

The authors test whether success on rule-generated tile boards reflects learning their rules or exploiting associated statistics. They create task metamers by sampling boards from learned tile-prediction networks, preserving selected statistics without applying the original rules, then compare human and agent performance. People do better on rule-generated boards while tested agents favor metamers, a behavioral contrast within this task family.

Summary by alphaXiv.

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