The Potential of AI to Create Personalized Exercise Plans

Abstract

Despite the evidence linking moderate intensity exercise with improved health outcomes and disease prevention, health care providers often fail to counsel and guide patients on exercise recommendations. Short visit times, personal exercise practices, and lack of knowledge are all cited as factors for omitting exercise counseling in health care appointments. In the context of these barriers, we investigated the use of GPT-3.5 in generating effective exercise recommendations. Thirty fictional patient cases were entered into GPT-3.5 and provided to two experts in lifestyle medicine. GPT-3.5 and experts were asked to create SMART and realistic goals for these patients, and to advise on when the patients could exercise, and what type of exercise they should complete. Three blinded reviewers graded the recommendations generated by GPT-3.5 and the experts in metrics of how (a) safe, (b) realistic, (c) personalized, (d) accessible (in line with the patient’s social determinants of health) the recommendations were, and (e) the quality of the recommendations overall. Differences between experts and GPT-3.5 were assessed using a Mann–Whitney U test. Differences between the three reviewer ratings were assessed using the intra-rater correlation coefficient. BothGPT-3.5 and experts in lifestyle medicine produced highly rated results. Our findings suggest that GPT-3.5 may be able to create safe and effective preliminary exercise recommendations.