The Effect of Patient Race on ChatGPT’s Probability of Success When Answering Medical Test Questions
Document Type
Conference Proceeding
Abstract
Although AI systems have successfully answered medical exam questions, one concern is that their response preferences may be based on construct-irrelevant item features. In this study, ChatGPT was given medical questions rewritten to include patient race. Despite being medically irrelevant, race was found to affect ChatGPT’s success on these items.
Recommended Citation
Braum-Bharti, Nina; Baldwin, Miles; and baldwin, peter, "The Effect of Patient Race on ChatGPT’s Probability of Success When Answering Medical Test Questions" (2023). NERA Conference Proceedings 2023. 2.
https://digitalcommons.lib.uconn.edu/nera-2023/2