People hear "mice with half-human brains" and reach for the phrase "artificial intelligence." They are right, but not in the way they think. Xenocortical tissue is not a computer that imitates intelligence. It is human intelligence's raw machinery — neurons, synapses, developmental programs — running outside the body. If machine AI is intelligence reconstructed from our outputs, xenocortical tissue is intelligence continued from our biology. The contrast between the two is one of the most interesting questions in science today.
Trained on us, versus grown from us
A large language model learns by compressing enormous amounts of human text — books, conversations, code — into statistical structure. Everything it "knows" is a reflection of what humans have written down. A xenocortical graft knows nothing yet. It starts as roughly 100,000 cells and develops by executing the same biological program that builds every human cortex. One is trained on our records; the other is grown from our blueprint. Machine AI is our echo. Xenocortical tissue is our tissue.
Six contrasts that matter
- Origin. Machine AI: engineered, designed architecture. Xenocortical tissue: self-organizing biology that follows developmental rules no one programmed.
- Learning. Machines learn from data sets, in weeks, on silicon. Cortical tissue learns from blood chemistry, sensation, and a body — slowly, in an animal.
- Speed and scale. A model can be copied a million times. A graft must be grown, one at a time, inside a living animal. Silicon scales; wetware doesn't.
- Failure modes. Models hallucinate — confidently producing falsehoods. Grafts, as the Nature paper showed with low oxygen, fail biologically: the human cells are fragile in distinctly human ways.
- Opacity. Both are hard to interpret, but in different senses. A neural network's opacity is mathematical. A graft's opacity is developmental — we are still discovering the rules it follows.
- Moral status. This is the sharpest difference. Nobody worries that a language model suffers. As we explore in The Ethics of Part-Human Brains, human neural tissue in an animal raises the question — not yet answered — of what kind of inner life, if any, could emerge.
The "real AI" argument, stated carefully
The most provocative framing in the field is "organoid intelligence": a 2023 essay in Frontiers in Science by Lena Smirnova and colleagues argued that three-dimensional brain cultures are a genuine computing substrate — one that already beats machines at certain tasks while consuming absurdly little energy. A trained human brain runs on roughly the power of a dim light bulb; a data center training a frontier model draws megawatts. Xenocortical mice push that argument further: they are organoid intelligence vascularized, embodied, and wired to a behaving animal. Not smarter than the mice around them — the researchers are clear about that — but human, in a way no data center has ever been.
Machine AI is intelligence abstracted from biology. Xenocortical research is biology asked to keep being intelligence.
What each one is for
The two technologies are not competitors; they are instruments pointed at different questions. Machine AI is superb at compressing human knowledge and generating its combinations. Xenocortical models exist to answer questions about biology: how human neurons die in dementia, why they fail in low oxygen, how circuitry assembles. In the Pașca lab's framing, the human brain has been the least therapeutically successful organ in medicine precisely because it is inaccessible — and a graft inside a mouse makes aspects of human brain development and function accessible for investigation for the first time.
That is the beauty of it, stated plainly. Not that we built an artificial mind — we have been trying to do that in silicon, and it is going well. But that we built an environment in which human neural matter can grow, integrate, and start to work. Two different kinds of artificial intelligence, and the second one, for the first time, is alive.
Sources
- Pașca, S. P., et al. (2026). Xenocortical transplantation of human cortical organoids in mice. Nature. doi:10.1038/s41586-026-11032-2
- Smirnova, L., et al. (2023). Organoid intelligence (OI): the upcoming frontier in biocomputing and intelligence-in-a-dish. Frontiers in Science, 1, 1067477.
- Sample, I. (2026, September 16). "Scientists create mice with part-human brains." The Guardian.
- Revah, O., et al. (2022). Maturation and circuit integration of transplanted human cortical organoids. Nature, 611, 518–526.
