IBM Watson AI gave risky advice to cancer patients

The IBM Watson supercomputer was the subject of a careful review after it was found that it was providing insufficiently accurate and dangerous advice on the treatment of cancer patients , according to PC Portal Press

Source: EPA / MATTHIAS BALK

This system is used in 230 hospitals around the world and is designed to assist physicians in diagnosis. The system uses artificial intelligence to analyze medical information about patients in combination with information from hundreds of medical journals. Since 2015, Watson has provided advice and guidance to approximately 60,000 patients

. However, medical experts working with IBM on this project have stated that Watson, at least in the field of oncology, has not shown the job. imprecise and dangerous advice on the treatment. All this was presented in internal documents, including two presentations by Andrew Norden, former assistant director of health at Watson's Department of IBM. One of these reports lists the case of a 65-year-old patient with lung cancer, who had bleeding plague. In this case, Watson has recommended chemotherapy and taking "Bevacizumab", which itself causes bleeding and can not be given to people who have it, as this can lead to fatal consequences.

hypothetically, and doctors do not have to adhere to it, but they are only an auxiliary diagnostic tool. However, deceptive leads can mislead doctors, and they themselves suspect in their own diagnostic evaluation and open the door to potential problems. Fortunately, no fatal cases have been recorded so far.

The reason for these problems seems to be that the system is "fed" during the learning phase by hypothetical rather than actual data from actual patients. It basically means that doctors have received recommendations from other doctors who have "filled" the data system rather than an artificial intelligence that has synthesized a huge amount of real data, and the conclusions drawn were imprecise .

IBM officials are like In response to these claims, a paper was published stating that 93% of patients correctly identified the tumor and gave the right treatment, and that she was constantly working to improve the system. This has been illustrated by the fact that in the past year alone, there have been up to 11 software updates aimed at improving the accuracy of tumor identification.