Proceedings of the Annual Meeting of the Cognitive Science Society

Unlike previous digital technologies, state-of-the-art large language models (LLMs) exhibit capabilities that span the full cognitive hierarchy, from low-level linguistic processing, to high-level functions such as social reasoning and metacognitive monitoring (Coda-Forno et al., 2024; Didolkar et al., 2024; Street et al., 2024). For these reasons, humans are increasingly collaborating with LLMs for complex cognitive tasks via conversational chat interfaces, and LLMs are increasingly studied as cognitive systems in their own right by cognitive scientists, psychologists and philosophers of mind (Hagendorff et al., 2023). Yet, the dynamics of human-LLM collaborations remain understudied.