Integrating AI-based exercises into teaching epidemiology: welcoming or bowing to the inevitable?

Introduction: Utilizing Artificial Intelligence (AI) tools in education has drawn tremendous attention recently. While certain disciplines (e.g., engineering) align more naturally, effective and ethical integration of AI in some fields, like veterinary medicine, requires a longer learning curve. We aim to share our experience of incorporating ChatGPT in teaching epidemiology and evidence-based veterinary medicine (EBVM) courses with practical examples and discuss the observed pros and cons.

Methods: We combined ChatGPT prompts with classic exercises in epidemiology (graduate-level) and EBVM (undergraduate-level) courses. ChatGPT scenarios were designed to address routine class exercises under various topics, including search strategies, causation web, measuring disease frequency and association, study design, sampling, and analyzing bias. As an example, students were asked to design a study with appropriate sampling strategy to estimate the prevalence of a disease in Hong Kong. They were then instructed to do the exercise using ChatGPT, compare their answers, and finally explore “pre-defined” prompts (prepared and refined by teachers) focused on limitations. This was repeated for most topics with necessary modifications.

Results: Some exercises (e.g., interpretation of association measures) were addressed quite accurately by AI in the first attempts. However, meticulous prompt engineering was required for topics like calculating frequency measures or sampling methods. Students believed adding AI components to routine learning activities helped them verify their thought processes, better understand interpretations, get familiar with the dangers of fully trusting ChatGPT, and prepare for the future.

Discussion: The integration process was time-consuming for teachers and demanded numerous trials and errors in some cases to guide AI and students properly. Lack of expert supervision often led to inaccurate or impractical results by AI. Feedback collection from students was on a narrative basis.

Conclusion: While ChatGPT offers many benefits and increases engagement in learning activities, expert supervision is still necessary to address its limitations.