Reproduce “When Everyone Can Build a Model: Mechanistic Inference in the Age of AI” 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
Marvin van Aalst, Adèle H. Ribeiro, Anna Matuszyńska
The authors explain how to assess what grey-box models support. They distinguish predictive fit from causal identification, use causal assumptions to determine which effects are identifiable, and compare equally well-fitting models to expose uncertain predictions. In a simulated example, a collider-based fit has R² 0.92 yet predicts the wrong intervention direction; this is not a clinical finding.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.