{"pk":49129,"title":"Cognitively Inspired Interpretability in Large Neural Networks","subtitle":null,"abstract":"Large Language Models (LLMs) and Vision Language Models (VLMs) have become a dominant force in artificial intelligence and have already made a major impact on the cognitive sciences, but debate persists concerning the extent to which they possess emergent cognitive capacities. Investigation of these systems at the level of behavioral outputs has led to conflicting findings, and the question of how these outputs are generated (at a mechanistic or algorithmic level) remains open. Yet, the abilities they do exhibit behaviorally offer an unprecedented opportunity to answer longstanding questions about how neural networks could, even in principle, achieve abilities that are thought to require structured representations—such as syntactic, combinatorial, and variable-binding operations. In this symposium, we highlight a recent body of work that addresses this gap in understanding by investigating the internal mechanisms that support cognitive processing in LLMs and other large-scale neural networks. The symposium brings together researchers with backgrounds in both computer science and psychology, exploring ways in which mechanistic interpretability research and cognitive science can mutually inform one another.","language":"eng","license":{"name":"","short_name":"","text":null,"url":""},"keywords":[],"section":"Symposia","is_remote":true,"remote_url":"https://escholarship.org/uc/item/4mc1z6qd","frozenauthors":[{"first_name":"Anna","middle_name":"","last_name":"Leshinskaya","name_suffix":"","institution":"University of California, Irvine","department":""},{"first_name":"Taylor","middle_name":"","last_name":"Webb","name_suffix":"","institution":"Microsoft Research","department":""},{"first_name":"Ellie","middle_name":"","last_name":"Pavlick","name_suffix":"","institution":"Brown University","department":""},{"first_name":"Jiahai","middle_name":"","last_name":"Feng","name_suffix":"","institution":"University of California, Berkeley","department":""},{"first_name":"Gustaw","middle_name":"","last_name":"Opielka","name_suffix":"","institution":"University of Amsterdam","department":""},{"first_name":"Claire","middle_name":"","last_name":"Stevenson","name_suffix":"","institution":"University of Amsterdam","department":""},{"first_name":"Idan","middle_name":"A","last_name":"Blank","name_suffix":"","institution":"University of California, Los Angeles","department":""}],"date_submitted":null,"date_accepted":null,"date_published":"2025-01-01T18:00:00Z","render_galley":null,"galleys":[{"label":"PDF","type":"pdf","path":"https://journalpub.escholarship.org/cognitivesciencesociety/article/49129/galley/37090/download/"},{"label":"PDF","type":"pdf","path":"https://journalpub.escholarship.org/cognitivesciencesociety/article/49129/galley/38635/download/"}]}