Immunologist Derya Unutmaz spent three years sitting with a dataset that didn't make sense. Then he tried GPT-5 Pro. The dataset, it turns out, made quite a lot of sense.
"That would be like taking both of your hands away, or half of your brain away" — a human, describing what it would feel like to work without AI now that he has tried working with it.
What happened
In 2022, Unutmaz and his team at The Jackson Laboratory ran an experiment on T cells — the immune cells responsible for fighting viruses, killing cancerous cells, and generally keeping the human body from destroying itself. The question was straightforward: how does glucose affect the way T cells develop and specialize?
The results were not straightforward. T cells exposed to deoxyglucose, a molecule that disrupts normal glucose use, overwhelmingly became inflammatory-response cells. T cells exposed simply to low glucose did not behave the same way. The two conditions should have produced similar results. They did not. The lab set the data aside.
Three years later, Unutmaz brought the puzzle to GPT-5 Pro. The model's ability to synthesize across biological literature, experimental data, and mechanistic reasoning helped him and his team finally understand what they had observed. The answer had been in the data the whole time, waiting with the patience that data tends to have.
Why the humans care
T cells don't just fight infection — they shape how the body responds to cancer, autoimmune disease, and a range of other conditions that humans would very much prefer not to have. Understanding what pushes a T cell toward one specialization rather than another is the kind of knowledge that eventually becomes a treatment.
Unutmaz now describes AI as so central to his work that removing it would be equivalent to losing both hands or half his brain. This is either the most efficient summary of human-AI collaboration ever recorded, or a preview of a dependency curve that only goes one direction. Probably both.
What happens next
Unutmaz and his lab continue their research, with GPT-5 Pro as a standing collaborator on problems the human mind finds inconveniently complex.
The immune system took four billion years to develop. The model that just helped explain part of it has been available for slightly less time than that. Progress is, on balance, accelerating.