TUESDAY, April 9, 2024 (HealthDay News) -- A significant proportion of artificial intelligence (AI) responses to primary care-related questions are "accurate with missing information," according to a study recently published online in the American Journal of Preventive Medicine.
Joseph Kassab, M.D., from the Cleveland Clinic Foundation, and colleagues assessed the capacity of ChatGPT-4 and Google Bard to deliver accurate recommendations in preventive medicine and primary care. The analysis included 56 questions presented to ChatGPT-4 in June 2023 and Google Bard in October 2023, with responses independently reviewed by two physicians.
The researchers found that after reaching a consensus, 28.6 percent of ChatGPT-4-generated answers were deemed accurate, 28.6 percent inaccurate, and 42.8 percent accurate with missing information. For Bard-generated answers, 53.6 percent were deemed accurate, 17.8 percent inaccurate, and 28.6 percent accurate with missing information. Notable inaccuracies (80 percent) were detected in both models in response to questions related to the Centers for Disease Control and Prevention as well as immunization.
"A significant proportion of the AI models' responses were deemed 'accurate with missing information,' emphasizing the importance of viewing AI tools as complementary resources when seeking medical information," the authors write. "Future studies, repeating evaluations at different time points and incorporating a broader range of question formulations for the same topic, would be valuable to account for the evolving nature of the AI models' training data and algorithm updates."
Abstract/Full Text
