AI chatbots unreliable sources for stroke care information – Harvard T.H. Chan School of Public Health
By
Amy Roeder
Current generative AI chatbots are not reliable sources of information about stroke care, according to a new study led by researchers from National Taiwan University and Harvard T.H. Chan School of Public Health.
They tested three chatbots that can generate conversational text and answer questions—ChatGPT, Claude, and Gemini—on realistic stroke scenarios, using several common techniques aimed at prompting suitable responses. They found that none were able to achieve the minimum threshold for providing safe, high-quality patient advice.
The study was published July 29 in npj Digital Medicine.
The researchers noted that while different styles of prompts could sometimes generate responses that were more empathetic, actionable, or accurate, the chatbots’ overall performance was inconsistent. Responses about stroke treatment were particularly unreliable.
“Our results show that while generative AI is impressive for general health information, it remains unreliable when patients face high-risk medical situations like stroke,” said first author John Tayu Lee, associate professor at National Taiwan University and senior researcher at Harvard Chan School’s Health Systems Innovation Lab (HSIL).
Stroke is the second-leading cause of death and the third-leading cause of disability worldwide, and disproportionately affects lower socioeconomic groups.
“Existing evidence suggests generative AI has real potential to help close health gaps and ease the shortage of health care workers in underserved and rural areas, especially when specialist access is limited,” Lee said. “But for this potential to be realized, we must not only improve the technology but also educate patients on how to ask the right questions to get safer, more useful answers.”
Senior author Rifat Atun, professor of global health systems and HSIL director, said, “Generative AI holds huge potential for enhancing global health equity, as GenAI solutions can be disseminated readily for wide application at low cost. But these solutions must be deployed responsibly, with robust governance, rigorous clinical validation, and human oversight to ensure appropriateness and safety.”
Amy Roeder
Senior Editor, Harvard Chan Magazine; Senior Writer in the Office of Communications
Stay connected with Harvard Chan School
In addition to our degree programs, we offer highly targeted executive and continuing education, directed and taught by Harvard faculty.
Stay connected with newsletters on climate, nutrition, communication — and the latest from Harvard Chan School.
