Without Consciousness, or Without Memory?

One of the patients listened to three stories from The Moth, the show in which ordinary people tell an episode from their own lives out loud. One was called My Father’s Hands. The patient was under general anaesthesia, partway through surgery for epilepsy that drugs had failed to control, and a probe seventy micrometres wide was recording the activity of individual neurons in their hippocampus.

After the operation, none of the seven patients in the study reported any explicit memory of what had happened in the theatre. Yet while the stories played, the activity of those neurons carried information about what the words meant and the grammatical role they played.

The study, by Kalman Katlowitz, Sameer Sheth and colleagues at Baylor College of Medicine in Houston, appeared in Nature on 6 May 2026. In the abstract the authors conclude that in the hippocampus complex processing of sensory stimuli occurs “even in the unconscious state”. The neural data are the heart of the paper and deserve to be taken seriously. The word “unconscious” deserves a separate question: what, exactly, supports it?

What they recorded

The patients, seven adults aged 25 to 54, were having the front part of a temporal lobe removed. After the lateral cortex had been taken out, and before the surgeons reached the deeper structures, they inserted a Neuropixels probe into the hippocampus and recorded for no more than thirty minutes. Anaesthesia was fully intravenous, with propofol as the main drug; the gases used at induction had been stopped at least an hour earlier.

Three patients heard a sequence of tones in which one of the two sounds came up only one time in five: the classic oddball design, the stimulus out of place. The neurons distinguished the rare tone, and in the two patients analysed for this, the distinction grew sharper over the roughly ten minutes of the experiment while the coding of tone identity weakened. The authors call this plasticity.

Four patients listened to podcasts. Categories of meaning and parts of speech could be decoded from the neurons’ activity, and a model built on word features predicted firing rates with a mean correlation of 0.397 across 368 units. In a separate group of ten awake patients, recorded earlier with different electrodes, the same figure was 0.226. The comparison is suggestive but not direct: different people, different instruments, different conditions.

The peer-review record, published alongside the paper, shows work that was argued over carefully. The original title was Learning and language in the unconscious human hippocampus. One referee thought “learning” overstated the case; another asked the authors to make clear that the results apply to propofol and not to every form of unconsciousness, such as sleep. The authors dropped “learning” and replaced “unconscious” with “anaesthetized”. In the abstract and in the section headings, though, the “unconscious state” stayed.

Two criteria for “unconscious”

In the paper, the patients’ state rests on two things. The first is BIS, an index computed from the frontal EEG that runs from 0, deep coma, to 100, fully awake: the anaesthesiologist adjusted the infusion to keep it between 45 and 60 throughout the operation. The second is memory: no patient, asked in the recovery unit or the next day, reported explicit memories of intraoperative events.

On the first point, a referee asked whether BIS had changed during the recording. The aim was to rule out that the oddball signal grew because the depth of anaesthesia was shifting rather than because the brain was adapting. The authors replied that they did not have access to those records and that a change was “very unlikely”: the recordings came hours after induction and hours before the end, with an anaesthesiologist watching. In one of the replies to the referees, the sentence intended for the Methods said that anaesthetic depth was “likely” flat; in the published version, the “likely” is gone.

The argument is reasonable, but it remains an inference: the values were never seen by the people who analysed the data. And even with the traces in hand, a BIS range is a clinical target for dosing the drug, not an instrument that certifies the absence of experience. In “The Hand That Answers Under Anaesthesia” we described patients who squeezed a hand on command while the frontal EEG showed a pattern usually associated with deep disconnection.

The criterion the result undermines

The second criterion sets up a small loop. Discussing their results, the authors write that if high-level sensory processing survives anaesthesia, then what is compromised is consolidation: the step from what happens to what will be remembered. They add that this could explain reports of implicit memory after anaesthesia.

If anaesthesia acts precisely on consolidation, the absence of memories on waking is what we would expect in both cases: if nothing was experienced, and if something was experienced without being kept. The criterion cannot tell the two apart. This is my reading, not the authors’ claim; but it uses only premises found in their text.

The distinction this objection rests on is not new. In 2012 Robert Sanders, Giulio Tononi, Steven Laureys and Jamie Sleigh summed it up in a title: Unresponsiveness ≠ unconsciousness. Under anaesthesia, they argued, three things can come apart: being conscious, being connected to the environment, and being able to respond. Dreaming is their example of disconnected consciousness. A 2018 meta-analysis of 22 studies and 1,131 patients found isolated-forearm responses during the maintenance phase as well, and no fewer when anaesthesia was guided by a brain monitor; its authors, however, flag low methodological quality and marked heterogeneity across the studies.

None of this says the Houston patients were conscious. No one tested them during the recording: no isolated forearm, no command, no question. The hypothesis that someone, for a few minutes, heard the stories without remembering them later finds no support in the data. The data simply do not rule it out.

What we actually know

The fact: in seven patients with epilepsy, under propofol, with a BIS the anaesthesiologist aimed to keep between 45 and 60 and no explicit memories on waking, hippocampal neurons distinguished a rare tone and encoded aspects of the meaning and grammar of speech. The authors’ interpretation: processing without consciousness, with consolidation compromised. The alternative hypothesis, for which there is no evidence here: processing accompanied at some moments by experience, disconnected or not, that never reached memory.

One detail makes the distinction less abstract. To explain how some processes can run without consciousness, the authors invoke the party at which we follow other people’s conversations without noticing. But that example is about someone awake: it describes processing outside attention, in a person who is fully conscious. It fits “without attention” better than “without consciousness”.

The study was not built to settle the question, and does not entirely claim to: among its limitations the authors acknowledge that anaesthesia has “an uncertain relationship with waking life” and that the results may not hold for other anaesthetics, for sleep or for coma. In principle, part of the question could be approached with tools that already exist: a connectedness test during the recording, such as the isolated forearm, and BIS values aligned with the neural data. They would not settle everything, because an experience disconnected from the environment, like a dream, would not answer a command anyway. Until then, the most precise word for those patients is the one chosen for the final title: anaesthetized.

The most solid finding of the paper is that, during those minutes, the brain processed the stories. What remains open is what we can conclude about the absence of experience, when the witness we question is the very memory that anaesthesia suspends.