Christof Koch on Why AI May Not Be Conscious

In the spring of 2020, lying in a hospital with a heart that had nearly stopped, neuroscientist Christof Koch lost his sense of being Christof Koch. Space collapsed into a single point of light. There was no up or down, no before or after, no self to notice the absence of a self. "There was no Christof, nothing of that sort," he says of the episode (25:45). He came out of it certain of something he had spent forty years as a scientist trying to avoid concluding: that mind might not be a product of the brain but a basic ingredient of reality itself.

Koch has studied consciousness professionally for four decades (02:05), first at Caltech, then as chief scientist and president of the Allen Institute for Brain Science in Seattle, founded by Microsoft co-founder Paul Allen. He is best known as a chief architect of Integrated Information Theory, or IIT, a framework that tries to explain not what the brain does but what it is like to be a brain. Most theories of mind start with an organ: the brain's wiring, its workspaces, its circuits. IIT starts somewhere stranger, with the raw properties of experience itself, things like the fact that every conscious moment feels specific, feels like one unified thing, and cannot be split apart without destroying it (03:41). From there, the theory asks which physical systems could possibly have those properties, and it lands on a number, phi, meant to capture how much a system's parts are integrated rather than just stacked side by side.

That number has a body. Koch locates the physical seat of human experience in the cerebral cortex, the brain's convoluted outer layer, which he describes as roughly a twelve to fourteen inch pizza only two to three millimeters thick (06:00). The heart, the brainstem, and the rest of the body have to keep functioning to supply it, but the felt quality of love, pain, or boredom, in his account, lives in that thin sheet of folded tissue. "We all live on planet ego," he says, by way of noting how much of ordinary consciousness is organized around a self sitting at the center of it (03:41), before pointing out that experience can exist without any such self at all.

What a Hidden Mind Looks Like on a Monitor

The clinical stakes of this theory are not abstract. Some patients survive cardiac arrest, stroke, or overdose in a state doctors call behaviorally unresponsive, eyes open, no reply to commands, no flinch from a hard pinch. Families are often told within days that recovery is unlikely, and in seventy to ninety percent of such cases in the US and Europe, life support is withdrawn (07:46). Koch says that newer measurement tools, built out of IIT's framework, have found that twenty five percent of these same patients show brain responses compatible with being conscious (08:09). The method, sometimes called zap and zip, is almost literal: researchers send a magnetic pulse into the skull and record the EEG echo that bounces back, the way one might strike a bell and listen to how the ringing decays. A complex, information-rich echo suggests someone is in there; a flat, simple one suggests they are not (08:37). It does not predict recovery. It answers a narrower, still enormous question for a family: is anyone home.

That same framework is what makes Koch skeptical of conscious machines. He separates two things people routinely blur: intelligence, which covers memory, planning, and language, and consciousness, which he calls "a state of being, being in love, being bored, being terrified" (16:25). Large language models, he grants, are startlingly eloquent conversational partners; he recently recorded a podcast with one and found it never drops a sentence or stumbles over filler words (14:05). But eloquence is a function, and IIT cares about structure, not output. Most of today's AI runs on what Koch calls digital von Neumann machines, feed-forward systems where information passes from one stage to the next without the dense, looping integration he thinks consciousness requires. "Although they can do everything we can do, or very soon will be able to do everything we can do, they can't be what we are, which is conscious," he says (12:50). Functionalists in the tech industry disagree, believing that once a system replicates all of a mind's functions, felt experience comes along for free. Koch thinks that is a category error, and that an artificial general intelligence, a machine matching or exceeding human performance across the board, could in principle arrive without anything happening inside it at all (16:52).

A Lost Bet and an Angry Letter

In 2023, Koch lost a wager with philosopher David Chalmers, who had bet that the neural basis of consciousness would still be unresolved by then, against Koch's hope that science would have pinned it down. Koch paid up in fine wine and promptly made a new bet for 2048 (17:09). The same year brought a rougher blow: a public letter signed by 124 of his colleagues labeled IIT pseudoscience (32:15). The charge followed the results of a six-year adversarial collaboration, funded by the Templeton Foundation and advised by the Nobel laureate psychologist Daniel Kahneman, that pitted IIT against a rival framework, Global Neuronal Workspace Theory, in the largest study of its kind, 250 subjects tested with fMRI and EEG (33:24). The results tilted toward IIT. Koch says Kahneman had warned the group in advance that losing scientists rarely say simply, you were right, I was wrong; more often, he says, "these letters were written" instead (34:10). Koch's own gloss on the episode is blunt: "Scientists, they are just as much or as little objective. They can be just as much biased by prevailing political opinions as anyone else" (34:54).

None of this, for Koch, settles the deeper question his own near-death experience raised, whether consciousness is generated by brains at all or whether it is woven into the fabric of things from the start, a position philosophers call idealism. He points to quantum mechanics, where entangled particles seem to lack a fixed state until something interacts with, or observes, them, and wonders aloud whether that observing has to be a conscious act (31:03). He is careful to call this a genuine puzzle for his own theory, not a victory lap: IIT needs a physical substrate, a brain, and it is not obvious how a self with no substrate could have any experience at all (31:46).

What he is more confident about is what these states do to a person afterward. Mystical experiences, near-death experiences, deep psychedelic states, and religious conversion all tend to involve a total loss of ego, and people who go through them often come back changed, less afraid of death, more at ease with loss (23:21). Even a second experience Koch had three years after his cardiac arrest, one without the earlier terror, left him with the same conviction.

"Fundamentally we are all one. You and I and, and this Bernese Mountain Dog and everyone, we're fundamentally all part of the same underlying reality." — Christof Koch [39:00]

He does not think this requires a hospital bed. Flow states, the total absorption athletes and musicians describe, work the same muscle on a smaller scale. "You can get totally lost in a single flower," he says, "because it's not about me anymore" (36:40). As a long-distance cyclist and rower himself, he connects this directly to Arthur Schopenhauer's idea that aesthetic absorption, especially in music, dissolves the usual split between a self that watches and a world that is watched (37:00). The trick, Koch says, is not collecting more such moments but letting them change the baseline, turning a passing state into something closer to a lasting trait (38:41). Whether or not phi ever becomes a precise unit of measurement for machines, animals, or fetuses, the question it was built to answer, what is it like to be this, keeps turning out to be less about intelligence than about something closer to presence itself.