An AI tries to read biological age and survival odds from a face

Doctors have long sized up how well a patient looks. Researchers at Mass General Brigham built a model to put a number on that impression. FAHR-Face learned from more than 40 million facial photographs. It was then fine-tuned for two jobs: estimating biological age and predicting survival.

The survival version was tuned on 34,363 photos of people with cancer. Among cancer patients, those it placed in the highest-risk quarter had more than three times the mortality of those in the lowest. The two models picked up different information, and using both together improved predictions.

So far, it has been tested mainly in cancer patients, so its performance in the general population is unknown. The authors present it as support for a doctor's judgment and plan broader testing.

A health estimate drawn from a selfie raises another question: who else might run your photo through a model like this, and what they would do with the answer.