Grim reaper’s ETA

Want to get an idea of when you’re going to die? Ask Google.

The omnivorous data-collecting company fed an artificial intelligence program 46 billion bits of data, including odd scraps such as notes written on patients’ charts, on more than 216,000 patients in hospital.

The program can be used to double-check doctors’ diagnoses. But its real purpose is to forecast the statistical probability of certain outcomes for any given patient, including how long that person would have to stay in the hospital, and how likely that person would be readmitted for further treatment.

In virtually all categories, the program was more accurate in its medical forecasts than conventional human methods.

In predicting death, for example, the program scored .95 accuracy out of a possible 1.00, compared to the .86 that traditional methods attain.


By predicting patient outcomes, Google’s program can help hospitals better plan the allocation and use of its beds and other resources. It also can serve as a check on quality of care, to see if doctors’ diagnoses and treatments solved patient illnesses on the first try.

More broadly, medical care is one of the largest targets for artificial intelligence. Other targets include speeding drug design, managing hospitals, and customizing care for individuals. Entrepreneurs and investors are lining up to cash in.

Skip to content