Scientists from multiple institutions have convened to examine how large language models could advance understanding of human cognition and brain processing, with potential applications in treating psychiatric and neurological disorders.
Benjamin Hayden, professor of neurosurgery at Baylor College of Medicine and adjunct professor of electrical and computer engineering and linguistics at Rice University, and Suzanne Kemmer, associate professor of linguistics and cognitive sciences at Rice, organised a symposium bringing together researchers from Rice, Baylor College of Medicine, the University of Texas, the Georgia Institute of Technology, the University of Montreal and other institutions.
The event explored connections between language processing in the brain and interpretability in AI systems, treating large language models as model organisms for studying how the brain implements language.
“We are very optimistic about their utility as scientific tools. In particular, we think that they have the power to answer basic questions about how language works and about how the brain implements language,” says Hayden.
The researchers noted that examining the internal processes of large language models reveals similarities to brain activity. This parallel offers neuroscientists an accessible system for studying language and conceptual processing compared to direct human brain research.
Hayden identified insufficient brain data as a primary challenge, noting that hundreds of hours of speech recordings with simultaneous monitoring of thousands of neurons would significantly advance understanding of how the brain processes language.
The convergence of brain research and AI could improve diagnosis and prognosis of psychiatric and neurological disorders. Large language models could analyse language patterns to provide rapid assessments comparable to blood pressure tests, potentially warning patients when intervention is needed.
Hayden expressed optimism about neuromodulatory treatments for psychiatric diseases, noting that large language models could play a critical role in measuring patient responses to treatment and optimising therapeutic interventions.
Kemmer noted that recent technological developments in brain research, combined with large language models, represent a significant advancement. Researchers can now observe language processing through monitoring brain activations at fine-grained levels during language use, whilst comparing results with language modelling technology.
The potential applications extend to dementia, addiction, PTSD, epilepsy, Parkinson’s and other neurological conditions, with researchers identifying a clearer roadmap for addressing these challenges than existed five years ago.
