This AI Can Predict When Someone Dies, But We Do Not Know How It Does That

Geisinger, a healthcare provider based in Pennsylvania, has come up with an AI that can predict which patients may die in the next few years, as reported by New Scientist.

The artificial intelligence is trained with 1.77 million ECG logs of 400,000 patients so that it can identify patterns that can suggest cardiac conditions like atrial fibrillation and heart attacks.

According to the researchers, the results are impressive and scary at the same time. The AI yields better results than existing methods in telling which patients are going to dies in the next few years and who will live.

According to Brandon Fornwalt, the study’s lead researcher, the method based on voltage is more accurate than any other way making use of an ECG.

The AI can even point out cardiac problems on those who have been cleared by cardiologists. This suggests that the machine is able to see things human doctors cannot or do not take seriously as they think the symptoms are normal. Fornwalt said that the AI may show us the things we have had wrong for a few decades.

This is not the first nor the only attempt to use the power of artificial intelligence to predict death. In 2018, Google made a model with a similar purpose to predict the length of stay of a patient, discharge time, as well as when he or she will die.

AI has already been applied in diagnosing lung cancer and heart disease. Sometimes they can be more accurate than a human doctor.

This seems like a piece of good news but like any other model, the AI of Geisinger has a major catch, its researchers cannot explain how it is able to reach this level of accuracy. So many professionals are reluctant about making any kind of decisions based on the predictions of these algorithms.