In humans, self-directed speech (whether inner or private) is theorized to play numerous metacognitive functions, from facilitating planning and shifts in attention to grounding the meaning of abstract concepts and serving as a (or, potentially, the only) medium of our conscious control over propositional thought. It is thus unsurprising that many of the recent performance breakthroughs in artificial intelligence (AI)-and large language models (LLMs), in particular-were achieved by implementing forms of self-directed and private self-talk. Early discoveries showed that coaching these systems to “show their work” via “chain-of-thought” prompting could provide dramatic improvements on mathematics and reasoning problems. These early successes inspired more ambitious uses of self-directed talk in LLM-based agents, innovations that were in many cases directly inspired by work on human inner speech.
This year’s Chambers Conference is part of a two-conference series jointly organized with Cameron Buckner at the University of Florida. The two conferences are designed to bring research on the metacognitive functions of language in conversation with research on AI and LLMs.
The conference will be held April 24-25 in 229 Andrews Hall (Bailey Library). It is free and open to all.