Adapted from a conversation with Claude, Anthropic’s AI model. Study figures were checked against the published abstracts on 2026-09-19.

The question put to the machine was blunt. AI may already be smarter than we are. Humans have wondered since the beginning what happens when we die, there is no proof, and religion is not the answer. So what does the AI say?

Its first move was to refuse the premise. “This is a data problem, not an intelligence problem,” Claude answered. A faster mind working from the same empty dataset arrives at the same place a slower one does. Then it answered anyway: death is most likely the end of experience, the same condition you were in before you were born. It put its confidence at roughly 90 percent.

The case for an ending

The argument rests on how tightly the mind tracks the brain. Anesthesia switches consciousness off and back on. A stroke can delete language, or the ability to recognize faces, while leaving everything else intact. Dementia takes a personality apart piece by piece while the body keeps going. Everything we can measure about a mind depends on working neural tissue, so the simplest inference is that when the tissue stops, the mind stops with it.

Nothing in physics offers a rival mechanism. No known carrier could hold memory or identity once the brain that stored them has decayed.

Epicurus reached the same conclusion 2,300 years ago without an EEG: where death is, you are not. There is no darkness and no waiting, because there is no one left to sit in the dark.

This is a data problem, not an intelligence problem.

Claude, asked what happens when we die

Pressed on the absolute

In that first answer Claude said flatly that no one, human or machine, has data from the other side. Asked whether that was 100 percent true, it backed down. “No. I overstated it.”

Claimed data exists, and an honest skeptic has to look at it:

• Veridical near-death perception. Patients report accurate details they supposedly could not have perceived. The best-known case is Pam Reynolds, who in 1991 described parts of her own brain surgery, performed under deep hypothermic cardiac arrest.
• Children who report past lives. Ian Stevenson and Jim Tucker at the University of Virginia’s Division of Perceptual Studies collected more than 2,500 cases, some with details matched to a specific deceased person.
• Terminal lucidity. People with advanced dementia sometimes become briefly coherent shortly before death, in brains that should no longer support it.
• Mediumship under blinded protocols. A handful of small studies report above-chance accuracy.

Why none of it counts yet

Every one of these reports comes from a living brain. A near-death survivor was revived, which means they did not die. At best they describe the border.

The controlled tests have come back empty. The AWARE II study, published in Resuscitation in 2023, placed a tablet showing a hidden image and headphones playing the names of three fruits beside cardiac arrest patients during CPR at 25 hospitals. Of 567 patients, 53 survived and 28 were interviewed. Eleven reported memories or perceptions from the arrest. None identified the image, and one of the 28 named the audio cue. With three fruits to choose from, that result cannot be told apart from a guess.

The same study found something that points the other way. Normal-looking EEG activity emerged as long as 35 to 60 minutes into CPR, which suggests the dying brain is less inert than assumed. Jimo Borjigin’s group at the University of Michigan has recorded surges of gamma-wave activity in dying rats (2013) and in two of four human patients after ventilator withdrawal (2023). A brain lighting up on the way out gives vivid end-of-life experience a plausible physical basis, with no second world required.

The Virginia cases lean on family testimony, cluster in cultures that already believe in rebirth, and were often contaminated before an investigator arrived. The mediumship results have not been independently replicated. The Society for Psychical Research began investigating these questions systematically in 1882. After roughly 150 years of that work there is still no replicable positive result, and that null carries weight of its own.

The other ten percent

The remaining doubt is not a hedge toward anyone’s heaven. It comes from the hard problem of consciousness: nobody can explain why physical processes produce subjective experience at all. If you cannot say how something arises, you should be careful about claiming to know exactly when it ends.

The tight correlation between brain and mind is what you would expect if the brain generates consciousness. It is also loosely compatible with stranger views, such as panpsychism or idealism, that are less parsimonious but not refuted. Derek Parfit added a further complication: the continuous self may be partly a construction even while you are alive, which changes what it means for “you” to end.

Claude called the Virginia cases and terminal lucidity unexplained, as opposed to debunked, and gave them as the reason its estimate sits near 90 percent and not 99.

What is left

The question may be permanently unanswerable, because it cannot be tested from this side, and more intelligence does not fix a missing dataset. The practical conclusion is the same under either outcome. Finite time is what gives the choices about how you spend it their weight.

Sources

• Parnia S, et al. “AWAreness during REsuscitation – II.” Resuscitation, 2023. PubMed
• Xu G, Borjigin J, et al. “Surge of neurophysiological coupling and connectivity of gamma oscillations in the dying human brain.” PNAS, 2023. Borjigin Lab summary
• Borjigin J, et al. “Surge of neurophysiological coherence and connectivity in the dying brain.” PNAS, 2013.
• Nahm M, Greyson B, Kelly EW, Haraldsson E. “Terminal lucidity: a review and a case collection.” Archives of Gerontology and Geriatrics, 2012.
• Beischel J, et al. “Anomalous information reception by research mediums under blinded conditions II.” Explore, 2015.
• Chalmers D. “Facing up to the problem of consciousness.” Journal of Consciousness Studies, 1995.
• Parfit D. Reasons and Persons. Oxford University Press, 1984.
• University of Virginia, Division of Perceptual Studies (Stevenson and Tucker case collection).