Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster A in Poster Session A: Tuesday, August 4, 9:30 – 11:15 am, Kimmel Center, Shorin & Rosenthal Rooms

Automated Adversarial Collaboration for Advancing Theory Building in the Cognitive Sciences

Suyog Chandramouli¹, George Kachergis², Akshay K Jagadish¹; ¹Princeton University, ²Stanford University

Presenter: Suyog Chandramouli

Cognitive science often evaluates theories through narrow paradigms and local model comparisons, limiting the integration of theories across a broad range of tasks and models. We introduce an automated adversarial collaboration framework for adjudicating among competing theories without requiring manual a priori specification of candidate models or experiments; the system combines LLM-based theory agents, program synthesis, and information-theoretic experimental design into a closed loop. In a simulation study spanning three classic categorization theories, we found that this framework can successfully recover ground-truth theories across various noise settings. Together, these findings provide the first proof of concept for an in-silico closed-loop theory adjudication system for cognitive science.

Topic Area: Methods, Tools, Theory & Neural Coding