AI learned to predict death

Scientists from Stanford University have developed an artificial intelligence system capable of predicting the probability of a patient’s death within a year to within 90%. Preliminary results were published in the archive of preprints arXiv.org.

The algorithm is based on deep training of neural networks, an approach that allows to operate with huge volumes of data. The training used data on the health and mortality of two million children and adults who were being treated in several large hospitals.

According to preliminary results, the algorithm is able to predict the probability of a patient’s death within a year with an accuracy of up to 90%.

According to the polls, most people would prefer to spend their last days not in the hospital, but at home, with relatives and friends, the researchers explain. And calculating the date of death will allow doctors to decide whether to keep the patient in the hospital or to let him go home.