Serious Games in a Graduate-Level Human Physiology Course

This paper describes a proof-of-principle design for integrating serious games into an online graduate-level Human Physiology course at Westfield State University, with the goal of teaching students both the benefits and limitations of generative AI in healthcare. The course alternates asynchronous preparation, including peer-reviewed case studies and video presentations, with synchronous sessions in which students complete an individual readiness test, then randomized team tests using scratch-off Immediate Feedback Assessment Technique forms that award partial credit. Teams then work through a tailored medical case presented as a slideshow, using ChatGPT to interpret it while navigating its errors.

The central design feature is the construction of cases intended to mislead ChatGPT. The instructor used the model itself to generate "distractor" signs and symptoms, omitting typical trigger terms, emphasizing unrelated systems, or mimicking other conditions. Reported results from the literature showing misdiagnosis rates of 51% to 72% motivate the approach, though the paper itself reports only the prompting procedure and the number of revisions required for each example condition, such as rhabdomyolysis, tetrodotoxin poisoning, and endometriosis. In every case the first iteration produced a correct diagnosis, and the signs had to be revised, in some instances up to three times, before a misdiagnosis occurred.

The paper's evaluation remains prospective. Planned measures include a participant survey on attitudes toward team-based learning, serious game elements, and AI integration, with the most successful elements to be trialled later in large undergraduate anatomy and physiology courses. The author also proposes the approach as a model for other disciplines and flags ethical considerations around AI in healthcare, though no outcome data on learning gains are yet reported.