Table 6.

Central themes in GAI research and ethical considerations in PHE.

Theme	Pointer	Details	Source
Central Concepts in AI Research	Keyword Centrality	AI, intelligent tutoring systems, machine learning, deep learning, higher education are central keywords	Talan et al[19]
	Popular Topics	Data science, learning analytics, computer-based learning, educational data mining
AI in Health Professions Education	Current Benefits	PHE currently benefits from AI advancements	Ken et al[20]
	Future Potential	AI set to offer more benefits to PHE in the future
Ethical Considerations in AI	Ethical Focus	Focus on ethics in the use of AI systems within PHE
	Ethical Issues	Data gathering, anonymity, privacy, consent, data ownership, security, bias, transparency, responsibility, autonomy, beneficence
	Ethical Guidance	Guide provides concepts, their importance, coping strategies, and suggests further reading
	Proactive Measures	Encouragement for PHE teachers and administrators to be proactive in ethical AI usage
Educational strategies	LLMs as teaching assistants	LLMs can provide summaries, presentations, translations, explanations, and guides in customizable formats, enhancing personalized education. They are suitable for creating interactive simulations, like patient history taking practices for medical students	Clusmann et al[21]
Critical thinking concerns	Impact on student skills	LLMs might negatively affect students’ critical thinking and creativity by providing easy answers, risking the externalization of factual knowledge and medical reasoning. Transparent regulation and differentiation between LLM-generated content and student work are necessary
Responsible use of LLMs	Education & interaction guidelines	Essential to establish guidelines for LLM use to prevent misuse. Students in medical education should receive an introduction to LLMs, learning about biases, limitations, and proper prompt engineering to avoid misinformation
AI performance on exams	AMBOSS users comparison	ChatGPT scored in the 30th percentile on Step 1 without Attending Tip and 66th with it; on Step 2, 20th and 48th percentiles, respectively. Accuracy decreased with question difficulty	Gilson et al[22]
	NBME vs AMBOSS performance	ChatGPT performed better on Step 1 than Step 2 and better on NBME questions than AMBOSS for both levels. Outperformed GPT-3 and InstructGPT on all data sets
	Logical use of information	ChatGPT provided logical explanations, using internal information for most responses and external information significantly more often in correct responses