US researchers' face-photo AI rates its own staff as dying
US researchers built an AI to help doctors gauge how sick a patient is from one photo of their face, then tested it on their own staff midweek.
"...and that is when we fed it our own staff badge photos," said the lead researcher, who has not blinked since the results came back. "All of them. Taken on a Wednesday. It told us we were dying."
The AI was built to help doctors judge how ill a patient is from a single photograph of their face, as US researchers have reported. That part looks promising. The team has filed it under fine.
The trouble started with the control group. A control group is the people you compare everyone else against. These were fourteen healthy colleagues. The AI ranked all fourteen as the sickest faces it had ever seen.
The team checked for luck. A p-value tells you how likely a result is to be a fluke. Theirs was tiny. On Thursday the same fourteen faces scored better. By Friday afternoon the machine read them as marathon runners.
The researchers stress the sample is small. Fourteen people. One week. One coffee machine, broken since Tuesday. Until more work is done, they advise that no hospital, airline or national government make a serious decision on a Wednesday.
The medical breakthrough has been moved to an appendix. The lead researcher has asked to be photographed only on Fridays.
