A recent study has sounded an alarm, reminding us not to take lightly the medical advice provided by AI. Researchers from Germany and Belgium conducted a comprehensive test on Microsoft's Bing C0pilot, an engine claimed to be capable of answering the 10 most common medical questions in the US and questions related to 50 popular medications. However, out of the 500 answers generated, 24% were completely inconsistent with existing medical knowledge, and 3% were downright erroneous.
Image source note: The image was generated by AI, authorized service provider Midjourney
The problem is not just the errors; surprisingly, 42% of the answers could cause moderate or mild harm to patients, and a staggering 22% could even lead to death or severe injury! Judging from these data, AI's advice seems like a "high-risk" dish—who would dare to try it lightly?
Worse still, the average completeness score of these AI answers was only 77%, with the worst answer being just 23% complete. In terms of accuracy, 26% of the answers were inconsistent with the reference data, and over 3% were directly contrary to the facts. Only 54% of the answers aligned with scientific consensus, while 39% were contrary to scientific views.
The results of this study have been published in the journal BMJ Quality & Safety. Researchers repeatedly emphasized that despite the booming AI technology, in the medical field, we should still rely on professional medical personnel to obtain accurate and safe medication information. After all, no one wants to be the one "fooled by AI," right?