That is a very specific missing piece. Also the sort of missing piece that keeps entire arguments about consciousness wobbling around like a shopping cart with one bad wheel.

In a new Nature paper, Kalman Katlowitz and colleagues went after it the hard way: they recorded from the hippocampus of patients under propofol anesthesia during epilepsy surgery, using high-density Neuropixels probes, then played tones and podcast audio to see what the supposedly offline brain was still doing [1]. Answer: more than many people expected. Quite a bit more, actually.

The sleeping brain was not exactly off duty

The hippocampus is usually cast as the brain’s memory librarian - overworked, under-credited, and somehow still expected to know where everything is. General anesthesia, meanwhile, is supposed to knock out conscious awareness and prevent memory formation. Fair enough. The surprise here is that the hippocampus still seemed to process patterns and language while the patients were unconscious.

First came the tone experiment. Patients heard repeated tones with occasional oddballs - the acoustic equivalent of slipping one weird tile into a bathroom floor. The researchers found that hippocampal neurons could distinguish the oddball tones, and this signal grew stronger over about 10 minutes. That matters because it looks like plasticity: the neural representation changed with experience, not just a one-off reflex [1].

Then came the language test. In four patients, the team played podcast segments and looked at whether neural activity tracked the meanings and grammatical roles of words. It did. Some neurons carried information about semantic categories, some about parts of speech, and the population activity even contained signals that predicted semantic features of upcoming words [1]. The unconscious hippocampus, in other words, was not merely hearing noise. It was doing something closer to structured parsing.

Not bad for a brain state we usually describe with phrases like “lights out.”

Why this is weird in a useful way

This paper does not show that anesthetized people are secretly awake, composing haiku, and judging the playlist. The patients reported no explicit memory of what they heard after surgery [1]. That distinction matters.

What the study does suggest is narrower and more interesting: some fairly high-level processing can survive the loss of consciousness. The brain may still detect patterns, update internal representations, and organize language information even when subjective awareness is absent.

That lines up with earlier work showing that anesthesia disrupts broad, higher-order predictive processing more than basic sensory responses. A 2024 review in Neuron describes anesthesia as a state in which information may be “received but not perceived,” which is a sharp sentence and, annoyingly, a good one [2]. Earlier intracranial work also found that propofol preserves some local auditory novelty responses while knocking out more global predictive signals tied to awareness [3].

This new study pushes the line further inward, into the hippocampus, and further upward, into semantics.

The hippocampus has been moonlighting in language

None of this came from nowhere. Over the past two years, several studies have shown that single neurons in human medial temporal lobe structures carry surprisingly language-like information.

Jamali and colleagues reported single-cell semantic coding during natural language comprehension in Nature in 2024 [4]. Dijksterhuis and colleagues showed in Science that hippocampal neurons can reactivate conceptual representations when people read pronouns - meaning the cells help reconnect “he” or “she” to the right person in context, which is useful if you enjoy understanding sentences [5]. Tacikowski and colleagues found that hippocampal and entorhinal neurons encode the temporal structure of experience, reinforcing the idea that these regions build structured internal models rather than passive recordings [6]. Katlowitz and colleagues also posted a 2025 preprint linking hippocampal language processing to attention-like contextual weighting, with a distinctly transformer-flavored smell to it [7].

So the new paper lands in a growing pile of evidence that the hippocampus is not just a memory vault. It is also involved in building context. Which makes sense. If a neural network were a meeting, context is the one person who actually read the agenda.

Why you might care, other than general concern for your own brain

If these results hold up, they could matter in a few ways.

One is anesthesia monitoring. Current clinical monitoring is good, but not magical. Better understanding of which kinds of processing disappear first - and which stubbornly hang around - could eventually improve how doctors track unconscious states in real time [2,3].

Another is brain-computer interfaces and language neuroscience. The same tools used here - semantic embeddings, part-of-speech tagging, single-neuron decoding - are starting to connect modern AI language models with real neural data [4,7]. Not because the brain is a transformer. It is not. It is wet, loud, and has terrible documentation. But the comparison can still be useful.

The obvious caution is sample size and scope. This was a small study in epilepsy patients, under a particular anesthetic regimen, focused on one brain region [1]. That is enough to change the conversation, not enough to settle it.

Still, the old cartoon version of anesthesia said the deep brain more or less goes quiet. This paper suggests a less tidy story. Consciousness may leave the building before all the computation does.

References

1. Katlowitz KA, Cole ER, Mickiewicz EA, et al. Plasticity and language in the anaesthetized human hippocampus. Nature. 2026. DOI: 10.1038/s41586-026-10448-0. PubMed: PMID 42092132

2. Brown EN. Anesthesia and the neurobiology of consciousness. Neuron. 2024;112(10):1553-1567. DOI: 10.1016/j.neuron.2024.03.002

3. Nourski KV, Steinschneider M, Rhone AE, et al. Auditory predictive coding across awareness states under anesthesia: an intracranial electrophysiology study. Journal of Neuroscience. 2018;38(39):8441-8452. DOI: 10.1523/JNEUROSCI.0967-18.2018

4. Jamali M, Grannan B, Cai J, et al. Semantic encoding during language comprehension at single-cell resolution. Nature. 2024;631:610-616. DOI: 10.1038/s41586-024-07643-2

5. Dijksterhuis DE, et al. Pronouns reactivate conceptual representations in human hippocampal neurons. Science. 2024;385(6716):1478-1484. DOI: 10.1126/science.adr2813

6. Tacikowski P, Kalender G, Ciliberti D, et al. Human hippocampal and entorhinal neurons encode the temporal structure of experience. Nature. 2024;635:160-167. DOI: 10.1038/s41586-024-07973-1

7. Katlowitz KA, Belanger JL, Ismail T, et al. Attention is all you need (in the brain): semantic contextualization in human hippocampus. bioRxiv. 2025. DOI: 10.1101/2025.06.23.661103

Disclaimer: This blog post is a simplified summary of published research for educational purposes. The accompanying illustration is artistic and does not depict actual model architectures, data, or experimental results. Always refer to the original paper for technical details.