Reproduce “auto-psych: Automating the science of mind using agent-driven theory discovery and experimentation” 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
Ben Prystawski, Kushin Mukherjee, Daniel Wurgaft, Linas Nasvytis, Michael Y. Li, Noah D. Goodman, Michael C. Frank
Stanford University researchers developed `auto-psych`, a framework that fully automates the scientific discovery process in computational cognitive science, enabling AI agents to generate theories, design experiments, collect human behavioral data online, and refine theories based on empirical evidence. The system's agent-discovered models provided a superior fit to human data compared to existing models in a subjective randomness task.
Key insights
- The nested loop architecture, combining both "lab-in-the-loop" experimentation and "Box's loop" (model critique), is critical for the system's successful model recovery and discovery.
- The
- theories as code
- paradigm, coupled with scalable programmatic human data collection via crowdsourcing, makes computational cognitive science particularly suited for full automation.
- AI agents can move beyond analyzing existing datasets to autonomously designing novel experiments and acquiring new empirical evidence from human participants to test and refine their theories.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.