Reproduce “Learning to Complement Humans” 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

Bryan Wilder, Eric Horvitz, Ece Kamar

The authors train a classifier and query policy together, using logged human responses to simulate consultations. They test it on Galaxy Zoo and breast-cancer-metastasis diagnosis. In Galaxy Zoo, the model reduces team loss—classification error plus query cost—by an average of 38.9% versus its fixed counterpart across query costs from 0 to 0.2. CAMELYON16 has 127 images.

Summary by alphaXiv.

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