Humans prefer medical advice from AI-bots despite low-accuracy: study

A growing reliance on artificial intelligence for medical guidance is raising concerns, as a new study reveals people often trust AI-generated health advice more than that of human doctors – even when the AI provides inaccurate information.

The Troubling Trend of AI Trust

Researchers from the Massachusetts Institute of Technology investigated this phenomenon by presenting 300 participants with medical responses from three sources: a medical doctor, an online healthcare platform, and an AI model like ChatGPT. Participants were asked to identify which response they trusted most. The results were striking: both medical experts and laypersons consistently favored the advice generated by AI.

Did You Know? A total of 300 participants were involved in the study, evaluating medical responses to determine their level of trust.

The study further demonstrated that participants struggled to differentiate between responses crafted by AI and those written by physicians. Alarmingly, even when presented with demonstrably inaccurate AI-generated advice, participants still rated it as valid, trustworthy, and complete.

Real-World Risks

This isn’t a hypothetical concern. There are documented instances of AI providing dangerous medical recommendations. One case involved a 35-year-old man from Morocco who sought treatment in the emergency room after a chatbot advised him to constrict his hemorrhoid with rubber bands. Another, more tragic case, resulted in a 60-year-old man poisoning himself after ChatGPT suggested consuming sodium bromide – a chemical used for sanitizing pools – as a way to reduce salt intake.

Expert Insight: The study highlights a significant vulnerability in how people assess medical information. The perceived authority of AI, even when demonstrably flawed, can override critical thinking and potentially lead to harmful health decisions.

Dr. Darren Lebl, research service chief of spine surgery for the Hospital for Special Surgery in New York, previously noted that approximately a quarter of the information generated by these AI programs is fabricated. This underscores the unreliability of relying solely on AI for medical guidance.

A recent survey by Censuswide found that roughly 40 percent of respondents already trust medical advice from AI bots like ChatGPT, indicating a widespread and growing acceptance of these technologies as healthcare resources.

Frequently Asked Questions

What did the study participants prefer?

Participants, including both medical experts and those without medical training, consistently rated medical advice generated by AI models as more accurate, valid, trustworthy, and complete compared to advice from human doctors or online healthcare platforms.

Could AI advice lead to negative health outcomes?

Yes, the study found that participants were inclined to follow potentially harmful medical advice provided by AI, even when the information was inaccurate, and incorrectly seek unnecessary medical attention as a result.

What was found about the accuracy of AI-generated medical advice?

Dr. Darren Lebl stated that about a quarter of the information generated by AI programs is fabricated, meaning it is not based on real scientific recommendations or published research.

As AI continues to evolve and become more integrated into healthcare, will individuals prioritize the perceived authority of technology over the expertise of medical professionals?