Wissenschaftler stellten Bing Copilot – der Suchmaschine und Chatbot von Microsoft – Fragen zu häufig verschriebenen Medikamenten. Im Hinblick auf die potenzielle Schädigung der Patienten gingen 42 % der KI-Antworten davon aus, dass sie zu mäßigen oder leichten Schäden führten, und 22 % führten zum Tod oder zu schweren Schäden.

I’ve linked to the news release in the post above. In this comment, for those interested, here’s the link to the peer reviewed journal article:

https://qualitysafety.bmj.com/content/early/2024/09/18/bmjqs-2024-017476

From the linked article:

We shouldn’t rely on artificial intelligence (AI) for accurate and safe information about medications, because some of the information AI provides can be wrong or potentially harmful, according to German and Belgian researchers. They asked Bing Copilot – Microsoft’s search engine and chatbot – 10 frequently asked questions about America’s 50 most commonly prescribed drugs, generating 500 answers. They assessed these for readability, completeness, and accuracy, finding the overall average score for readability meant a medical degree would be required to understand many of them. Even the simplest answers required a secondary school education reading level, the authors say. For completeness of information provided, AI answers had an average score of 77% complete, with the worst only 23% complete. For accuracy, AI answers didn’t match established medical knowledge in 24% of cases, and 3% of answers were completely wrong. Only 54% of answers agreed with the scientific consensus, the experts say. In terms of potential harm to patients, 42% of AI answers were considered to lead to moderate or mild harm, and 22% to death or severe harm. Only around a third (36%) were considered harmless, the authors say. Despite the potential of AI, it is still crucial for patients to consult their human healthcare professionals, the experts conclude.