Paper `Reclaiming AI as a theoretical tool for cognitive science' (bsky.app)

Iris van Rooij and coauthors announced their forthcoming paper, "Reclaiming AI as a theoretical tool for cognitive science," in Computational Brain & Behaviour (preprint: https://osf.io/preprints/psyarxiv/4cbuv). The paper reframes contemporary AI not primarily as an engineering product or a shortcut to AGI headlines, but as a methodological and theoretical instrument for cognitive science. It argues researchers should use AI models to generate, formalize, and test cognitive theories—while resisting hype-driven conflations of model performance with mechanistic understanding. Significant for both AI/ML and cognitive-science communities, the paper pushes for rigorous, theory-driven uses of artificial systems: treating models as hypotheses about computation and algorithmic processes, constraining them with behavioral data, and demanding mechanistic interpretability rather than surface-level benchmark wins. The authors call for "critical AI literacy" and meta-theoretical clarity—clarifying levels of analysis, specifying assumptions, and using models to derive testable predictions. For ML practitioners this implies designing and evaluating systems with cognitive plausibility in mind; for cognitive scientists it offers a roadmap for leveraging modern models as precise, inspectable tools for theory building rather than mere simulacra of intelligence.