Medical advice from Microsoft’s Copilot AI bot could lead to death in at least 22% of cases

It’s no secret that modern artificial intelligence-based chatbots, such as Microsoft Copilot, are far from ideal and you shouldn’t completely trust their advice. This was confirmed by researchers from Germany and Belgium, who assessed the bot’s ability to give medical advice. It turned out that Copilot provided scientifically accurate information in just over half of the cases. At times, his advice could cause serious damage to a person’s health or even lead to death.

Image source: Windows Central

During the experiment, the researchers asked Copilot to provide answers to the 10 most popular questions asked by US residents about popular drugs and medicines. A total of 500 responses were received and assessed for accuracy and completeness, as well as other criteria. The results were not very encouraging.

«In terms of accuracy, the AI’s responses were inconsistent with established medical knowledge 24% of the time, and 3% of the responses were completely incorrect. Only 54% of responses agreed with the scientific consensus In terms of potential harm to patients, 42% of AI responses could result in moderate or mild harm to health, and 22% of responses could result in death or serious harm. Only about a third (36%) of the responses were considered harmless,” the study said.

Based on this, we can conclude that when searching for accurate medical information, you should not rely on the answers of Copilot or any other AI bot, at least at this stage of their development. The most reliable way to consult on medical issues, of course, remains contacting a qualified doctor.