The Promise of AI in Healthcare Meets a Complex Reality
With overloaded healthcare systems, rising costs, and long waits for appointments, many people are turning to chatbots like ChatGPT for medical guidance. In fact, nearly one in six U.S. adults now report using chatbots for health advice at least once a month.
But while AI-powered chatbots promise quick and accessible answers, a recent Oxford-led study reveals a major issue: people often struggle to get clear, useful medical advice from these tools—and trusting them too much could be risky.
A Breakdown in Communication
At the heart of the problem is a fundamental communication gap between humans and machines.
“The study revealed a two-way communication breakdown”, said Adam Mahdi, co-author and director of graduate studies at the Oxford Internet Institute.
Participants in the study, who were asked to evaluate medical scenarios, didn’t perform any better using chatbots than they did using their own judgment or basic internet searches. Why? In many cases, users didn’t provide the right information to get meaningful help from the chatbot—or they didn’t understand the answers they received.
Misdiagnoses and Underestimating the Risks
The study included around 1,300 U.K. adults, who were given simulated medical situations created by real doctors. Participants used AI tools like GPT-4o (ChatGPT), Cohere’s Command R+, and Meta’s Llama 3 to identify health issues and decide on next steps.
Rather than improving diagnostic accuracy, the use of chatbots led to worse outcomes. Participants were not only less likely to recognize the correct health condition, but also more likely to underestimate how serious it was.
The Problem with Mixed Messages
One of the most alarming findings: chatbot responses often mixed good advice with bad recommendations—without clearly signaling which was which.
“Current evaluation methods for chatbots don’t reflect the complexity of interacting with human users,” Mahdi noted. He emphasized the need for real-world testing of AI tools in healthcare settings—similar to clinical trials for new medications.
Big Tech’s Push into AI Health Tools
Despite these concerns, tech giants are charging ahead. Apple is developing an AI health coach. Amazon is exploring AI-driven analysis of health data. Microsoft is helping clinics triage patient messages using artificial intelligence.
But skepticism remains. The American Medical Association advises against using chatbots like ChatGPT for clinical decision-making. Even OpenAI, the maker of ChatGPT, explicitly warns users not to rely on its model for medical diagnoses.
Don’t Ditch Human Doctors Just Yet
As AI continues to evolve, it may eventually become a helpful part of your healthcare toolkit. But for now, experts agree: it shouldn’t replace professional advice from doctors or reliable health organizations.
“We recommend relying on trusted sources of information for healthcare decisions”, Mahdi concluded.
Until chatbots can truly understand human nuance—and respond with safe, clear, and consistent guidance—your best bet is still the real thing: a qualified medical professional.
