Nature study finds hippocampal neurons perform predictive language processing, including part-of-speech differentiation and next-word prediction, under general anesthesia.
Key Takeaways
Baylor researchers used Neuropixels probes in the hippocampus during epilepsy surgery to record single-neuron activity under general anesthesia.
Hippocampal neurons distinguished novel tones over time, indicating neural plasticity without consciousness, and also differentiated nouns, verbs, and adjectives from played stories.
Predictive coding, anticipating upcoming words in sentences, was observed unconsciously, a behavior previously assumed to require conscious attention.
Authors draw a direct parallel to LLM next-token prediction, suggesting shared computational principles between biological and artificial language systems.
Limits: results specific to one anesthesia type, one brain region; applicability to sleep, coma, or other states is unknown.
