A recent study published in a medical journal highlights significant shortcomings in the health advice provided by AI chatbots, indicating that these systems frequently deliver inaccurate recommendations. Researchers investigated how well popular chatbots perform in comparison to standard Google searches when it comes to preliminary diagnoses and determining appropriate actions, such as whether to call for emergency assistance or manage care at home.
The study involved over 1,200 participants from the UK, all without medical training, who were presented with clinical scenarios featuring various symptoms and medical histories. Participants were tasked with using chatbots to identify the correct course of action. Alarmingly, the correct solution was identified in less than 50% of cases, with accurate diagnoses occurring only about 34% of the time.
The researchers emphasize that none of the tested AI chatbots are currently suitable for direct patient interaction. They argue that these systems should be designed to ask clarifying questions and gather more information, similar to the approach taken by healthcare professionals. While developers acknowledge the issue and claim that newer models are increasingly requesting additional information and making fewer mistakes in emergency situations, the study's authors caution that relying on AI as a substitute for human doctors remains hazardous.
In a related incident, a Russian woman contracted leptospirosis during her vacation in Bali after coming into contact with a rodent. Local doctors misdiagnosed her condition as dengue fever, leaving her untreated. In a turn of events, she uploaded her test results to ChatGPT, which suggested leptospirosis and recommended a course of antibiotics. Remarkably, her condition improved within four days following that treatment.
The findings of this study raise critical concerns for the healthcare technology market, emphasizing the need for improved accuracy in AI health advisory systems. Competitors in the field may need to reconsider their strategies to ensure user safety and build trust in AI-assisted medical advice.
Informational material. 18+.
