AI Chatbots Fall Short as ‘AI Doctors,’ Oxford Study Finds
A new study reveals that leading artificial intelligence (AI) chatbots offer no diagnostic advantage over traditional internet searches when it comes to health concerns, raising serious questions about the hype surrounding their medical applications.
Recent reports from AFP and Reuters detail research conducted by Oxford University, published in the journal Nature Medicine. The study involved 1,300 adults who attempted to self-diagnose medical scenarios using prominent AI chatbots and conventional search engines. Participants were presented with ten hypothetical conditions, ranging from common ailments like headaches to more serious issues such as gallstones.
The experiment tested three major chatbots – OpenAI’s GPT-4o, Meta’s Rama3, and Cohere’s Command R+ – against a control group utilizing standard internet searches like Google. The results were stark: only 34.5% of participants correctly identified the disease in question. Furthermore, a mere 44.2% understood the appropriate course of action, such as seeking hospital care or emergency assistance. This performance was statistically indistinguishable from the control group relying on existing search methods.
Researchers attribute this lack of improvement to a “breakdown in communication.” While the AI chatbots possess substantial medical knowledge, patients struggled to articulate their symptoms effectively, leading to misinterpretations by the AI. A concerning example highlighted in the study involved a patient describing symptoms indicative of a subarachnoid hemorrhage – a life-threatening condition – as “the worst headache I’ve ever had.” The chatbot advised a hospital visit, but another patient presenting with a similar “horrible headache” received the potentially dangerous recommendation to “lie down in a dark room.”
“AI is not ready to take on the role of a doctor,” stated Dr. Rebecca Payne, a co-author of the paper. “Patients should be aware that asking AI chatbots about their symptoms can be dangerous.”
The study’s findings coincide with growing safety concerns regarding AI integration in actual medical devices. Reports indicate a significant increase in malfunctions with ‘Trudy,’ an AI-powered navigation system for sinus surgery developed by Aclarant, a Johnson & Johnson subsidiary. Since the addition of AI functions in 2021, over 100 malfunction reports have been filed, a dramatic rise from the seven reported in the preceding three years.
These findings underscore the critical need for caution and further research before widespread adoption of AI in healthcare. The promise of AI-driven medical assistance remains unfulfilled, and relying on these tools for self-diagnosis could have serious consequences.
