Under full anesthesia, the brain kept predicting the next word of a story

Seven people under general anesthesia lay unconscious on an operating table while researchers played short stories into the room, and neurons deep in their brains kept tracking the words. A team from Baylor College of Medicine recorded hundreds of individual hippocampal neurons during epilepsy surgery and found that the cells encoded nouns, verbs and adjectives and could be used to predict which word was coming next. The patients had no memory of the stories afterward.

Tech news worth reading: Get Techpresso, the daily email with the most interesting tech news and insights, read by 400,000+ professionals from Google, Apple and OpenAI.

Get Techpresso →

We may earn a commission if you sign up.

The result appeared in the journal Nature in a paper called “Plasticity and language in the anaesthetized human hippocampus.” It cuts against a common intuition that anesthesia switches off the brain’s handling of the outside world, though the authors are careful about how far the finding goes.

Neuropixels recordings in the hippocampus

The participants were patients with severe epilepsy who were scheduled for an anterior temporal lobectomy, a surgery that removes part of the temporal lobe. According to Scientific American’s account of the study, all seven were anesthetized intravenously with propofol as the main drug, and surgeons temporarily placed Neuropixels probes in the hippocampus before the tissue was removed.

Neuropixels probes are dense silicon electrodes that can record from hundreds of single neurons at once, which is what made the experiment possible. That resolution let the researchers watch individual neurons respond to sound rather than inferring activity from a blurred average, and it is why the claim about word categories can be made at the level of single cells. The hippocampus, a structure best known for its role in memory, was a workable target because the tissue was slated for removal as part of the epilepsy treatment, which is what allowed probes to be placed there during surgery. It also means the study asks a narrow question about one region, not about the whole brain.

Sameer Sheth, a Baylor neurosurgeon and professor, summed up the finding in a statement carried by ScienceDaily: “Even when patients are fully anesthetized, their brains continue to analyze the world around them.” Co-author Benjamin Hayden said the kind of predictive coding the team saw is something usually associated with being awake and attentive.

Oddball tones, then stories

The first experiment used repeating tones with occasional unexpected frequencies. Science News reported that more than 70 percent of the monitored neurons responded to sounds and distinguished the rare oddball tones from the standard ones. The discrimination improved over a 10-minute session, and Sheth told Scientific American it emerged over time and was not detectable in the first few minutes. The authors read that improvement as a sign of plasticity, meaning the neurons were learning the pattern even while the patients were unconscious.

The second experiment used language. Some patients heard the Moth Radio Hour podcast, and others heard educational videos and storytelling for 10 to 20 minutes. Neural activity carried information about nouns, verbs and adjectives, and firing patterns tracked meaning, with the word “cat” represented closer to “dog” than to “pen.” Crucially, the activity could be used to predict the upcoming word in real time.

That prediction is the reason the work resonates beyond neuroscience. Large language models are built on the same basic task of guessing the next word, and the authors link the hippocampal result to predictive coding, the theory that brains constantly forecast incoming input. They do not claim the patients understood the stories, and the patients reported no conscious memory of the sounds or the narratives when they woke, so the finding concerns neural machinery rather than experience.

Limits of the study’s evidence

The Technology Networks summary of the Baylor release lists the same limits: one type of anesthesia and one brain region. The sample is small, and all seven patients were anesthetized with one regimen. Sheth was explicit with Scientific American that the findings were obtained under a particular anesthetic regimen and that the team does not yet know whether sleep, coma or other drugs would produce the same results. Only the hippocampus was recorded, so processing elsewhere in the brain remains unmeasured.

Outside researchers read the result as an addition to a shifting picture rather than a break with it. Athena Akrami, a neuroscientist at University College London who was not involved, said the field was already moving toward a more nuanced picture, and that the computations looked nearly identical to those in awake brains. Katlowitz, the lead author, said the processing can function entirely independently of consciousness.

There is a practical question underneath. Janna Helfrich, a Yale anesthesiologist who was not part of the study, noted that the results fit earlier reports of above-chance word recognition under anesthesia, and asked how much of the auditory environment patients process during anesthesia. Katlowitz also raised the possibility that these signals could one day inform speech prosthetics for patients with stroke or injury. The paper itself holds the full methods and data.

This article was produced with the assistance of AI and reviewed by Morning Overview editors prior to publication.

Tech news worth reading: Get Techpresso →

We may earn a commission if you sign up.