Artificial intelligence in medicine: When it comes to medicines, chatbots can’t be trusted yet

by time news

2024-10-13 09:46:00

What was prescribed for me? How much of this can I swallow per day? What side effects are there? Those who take medications often prefer to consult the Internet rather than their doctor or pharmacist. But according to a new study, this can be dangerous.

AI-powered search engines and chatbots do not provide reliable information about drugs, a study says. The answers were repeatedly inaccurate, incomplete and often difficult to understand, the Erlangen researchers write in the specialized magazine “BMJ Quality & Safety”.

They therefore advise caution and require warnings from users. A central conclusion of the study is that the quality of chatbot responses is not yet sufficient for safe use by users.

Furthermore, first author Wahram Andrikyan from the Institute of Experimental and Clinical Pharmacology and Toxicology at the University of Erlangen explained: “In our opinion, a clear indication that the information provided by the chatbot cannot replace professional advice is therefore essential.”

Inaccuracies cannot be recognized by laypeople

The starting point of the study was the experience that patients wanted to find out about the medicines they were prescribed on the Internet. Therefore, in April 2023, the research team asked the AI-supported chatbot of the Microsoft Bing search engine ten common questions about the 50 most prescribed drugs in the United States, including questions about how to take them, side effects or contraindications.

Overall, the chatbot answered questions with a high degree of completeness and accuracy, Andrikyan said. But for some questions this was not the case. “This poses a risk to patients because, as lay medical professionals, they cannot evaluate the accuracy and completeness of the responses generated by the AI ​​themselves,” Andrikyan summarized.

Compared to last year’s study, there has been rapid progress in AI-supported search engines with integrated chatbot functions, the expert highlighted. However, improvements are not sufficient and risks to patient safety remain until further notice.

However, since chatbots are trained with different data sets, he believes it makes sense to also examine the security and quality of other technical systems.

dpa/goal

#Artificial #intelligence #medicine #medicines #chatbots #trusted

You may also like

Leave a Comment