The news is published anonymously, if the author has no desire to disclose his name. In order for personal data to appear in the column on authorship, you must specify your contact information in a special form below.
The development of artificial intelligence began to move beyond the limits of the possible. Doctors in the US have developed a neural network that predicts death is more likely than doctors. To unravel the future, you need to show the cardiogram of a human neural network and that will make a conclusion about his future situation. The principle of functioning of system of artificial intelligence is not yet clear.
Testing occurred in Pennsylvania, in a specialized medical facility Geisinger under the leadership of Brandon Fornwalt and his colleagues. The neural network was programmed in the study of 1.77 million results of an electrocardiogram (in total about 400 000 people) to predict death of patients within the next year.
After the neural network has performed an analysis of the electrical activity of the heart, she separated those patterns that have changed from heart disease, including heart attacks and atrial fibrillation. Under study were two of the neural network: from raw data of ECG and ECG data in combination with age and gender of patients.
To measure performance, the researchers used the AUC, through which you can understand how well the AI was able to distinguish patients who died during the year and those who survived. The neural network has consistently scored above 0,85. This indicator is used to assess risks doctors. They have it varies from 0.65 to 0.8 points. The data obtained confirm that the program accurately predicts death, unlike doctors, however, to calculate the calculation algorithm program and failed.
After researching Brandon Fornwalt said - “this discovery shows that the algorithm picks up things that people probably can't see, or at least ignore and think it's okay. Artificial intelligence can potentially teach us things that we may have misinterpreted for decades.”