A computer took a second look at 10,000 breast scans and caught cancers two doctors had missed. Another one, working alone, missed most of the emergencies it was built to catch.

In the UK, women are offered a routine breast X-ray — a mammogram — to catch cancer early, and every scan is checked by two specialist doctors, one after the other. In 2024 the health board for north-east Scotland, NHS Grampian, working with the University of Aberdeen, reported a test of a computer program called Mia, made by a company called Kheiron Medical Technologies. The program was run over the scans of more than ten thousand women — scans that both doctors had already checked and cleared [1].

The program pointed at a small number of scans — reported as eleven — and said, in effect, look at this one again. They turned out to show cancers that both doctors had missed. The flagged scans went back to the doctors, who made every decision about whether to call a woman back for more tests; the program decided nothing on its own. Several of the cancers it found were small and caught early [1].

In February 2025 the UK's health department announced a much bigger trial, called EDITH, to test this kind of program across the national screening service: around 700,000 women at about thirty screening centres, with roughly £11 million of public research funding. The design keeps a person in the loop: the trial tests the program doing the job of one of the two doctors, never both [2].

Look at where the program was placed. It checked the scans after two people had, not instead of them, and all it could do was ask a question — look at this one again — to a doctor who still owned the answer [1]. A second pair of eyes in that position can only add something — like a second cashier re-adding your bill before you pay: they can catch a mistake, but they cannot charge you more. The worst it can do is send a doctor back to a scan they had already cleared, which costs a minute. That is also why the big trial was set up the way it was. The program takes one of the two chairs, a person keeps the other, and the whole arrangement is measured against two people before anything changes for patients [2].

Now the same kind of tool, in the opposite chair. Sepsis is the body's runaway reaction to an infection; it can kill within hours if nobody spots it, so hospitals want an early warning. Epic, a company whose software runs many hospitals' records, built a program to raise that alarm. It was in use at hundreds of US hospitals when researchers at the University of Michigan's hospital checked it against their own patients — nearly forty thousand hospital stays. The program missed about two out of every three sepsis cases it was meant to catch. Of the cases it did catch, most had already been spotted and treated by the staff; it gave a genuinely new warning only a small fraction of the time. At the same time it raised an alarm for almost one in five of all the patients in the hospital, so the nurses and doctors it was warning learned to look past it. It was like a car alarm that goes off every time a lorry drives past: after a week, nobody looks up. The company had reported a far higher accuracy, and hospitals had taken its word for it [3][4]. Same kind of program as the scan-checker. The difference was the chair: one sat next to people who still decided, the other sat in front of people who had no way to check it and every reason to stop looking.