ChatGPT invents breast cancer screening advice responses

Occurred: April 2023

Can you improve this page?
Share your insights with us

ChatGPT has been found to make up fake information when asked about breast cancer screening, prompting doctors to warn users not to use the chatbot for medical advice.

University of Maryland School of Medicine researchers asked ChatGPT to answer 25 questions related to advice on getting screened for breast cancer, with each question asked three separate times and the results analysed by radiologists trained in mammography.

According to the researchers, ChatGPT answered one in ten questions about breast cancer screening wrongly, some of which were ‘inaccurate or even fictitious'. They also discovered that correct answers generated by the bot were not as ‘comprehensive’ as those found through a simple Google search.

The findings highlighted concerns about ChatGPT's tendency to 'hallucinate' false information, and prompted medical experts to warn users to avoid using the system for medical advice.


Operator: Hana L. Haver, Emily B. Ambinder, Manisha Bahl, Eniola T. Oluyemi, Jean Jeudy, Paul H. Yi
Developer: OpenAI
Country: USA
Sector: Health
Purpose: Provide cancer screening advice
Technology: Chatbot; NLP/text analysis; Neural network; Deep learning; Machine learning; Reinforcement learning
Issue: Accuracy/reliability
Transparency: Governance