Artificial intelligence (AI) is a growing phenomenon, and will soon facilitate wide-scale changes in many professions, including medical education. In order for medical educators to be properly prepared for AI, they will need to have at least a fundamental knowledge of AI in relation to learning and teaching, and the extent to which it will impact on medical education. This Guide begins by introducing the broad concepts of AI by using fairly well-known examples to illustrate AI’s implications within the context of education. It then considers the impact of AI on medicine and the implications of this impact for educators trying to educate future doctors. Drawing on these strands, it then identifies AI’s direct impact on the methodology and content of medical education, in an attempt to prepare medical educators for the changing demands and opportunities that are about to face them because of AI.

Health Professions Education (HPE) has benefitted from the advances in Artificial Intelligence (AI) and is set to benefit more in the future. Just as any technological advance opens discussions about ethics, so the implications of AI for HPE ethics need to be identified, anticipated, and accommodated so that HPE can utilise AI without compromising crucial ethical principles. Rather than focussing on AI technology, this Guide focuses on the ethical issues likely to face HPE teachers and administrators as they encounter and use AI systems in their teaching environment. While many of the ethical principles may be familiar to readers in other contexts, they will be viewed in light of AI, and some unfamiliar issues will be introduced. They include data gathering, anonymity, privacy, consent, data ownership, security, bias, transparency, responsibility, autonomy, and beneficence. In the Guide, each topic explains the concept and its importance and gives some indication of how to cope with its complexities. Ideas are drawn from personal experience and the relevant literature. In most topics, further reading is suggested so that readers may further explore the concepts at their leisure. The aim is for HPE teachers and decision-makers at all levels to be alert to these issues and to take proactive action to be prepared to deal with the ethical problems and opportunities that AI usage presents to HPE.

Background: Edgar Dale's Pyramid of Learning and percentages of retained learning are cited in educational literature in a range of disciplines. The sources of the Pyramid, however, are misleading. Aims: To examine the evidence supporting the Pyramid and the extent to which it is cited in medical education literature. Methods: A review of literature (1946–2012) based on a search utilising Academic Search Complete, CINAHL, Medline and Google Scholar conducted from September to November 2012. Results: A total of 43 peer-reviewed medical education journal articles and conference papers were found. While some researchers had been misled by their sources, other authors’ interpretations of the citations did not align with the content of those citations, had no such citations, had circular references, or consulted questionable sources. There was no agreement on the percentages of learning retention, in spite of many researchers’ citing primary texts. Discussion and conclusion: The inappropriate citing of the Pyramid and its associated percentages in medical education literature is widespread and continuous. This citing undermines much of the published work, and impacts on research-based medical education literature. While the area of learning/teaching strategies and amount of retention from each is an area for future research, any reference to the Pyramid should be avoided.

Mobile technologies (including handheld and wearable devices) have the potential to enhance learning activities from basic medical undergraduate education through residency and beyond. In order to ...

Background: Large language models such as GPT-4 (Generative Pre-trained Transformer 4) are being increasingly used in medicine and medical education. However, these models are prone to “hallucinations” (ie, outputs that seem convincing while being factually incorrect). It is currently unknown how these errors by large language models relate to the different cognitive levels defined in Bloom’s taxonomy. Objective: This study aims to explore how GPT-4 performs in terms of Bloom’s taxonomy using psychosomatic medicine exam questions. Methods: We used a large data set of psychosomatic medicine multiple-choice questions (N=307) with real-world results derived from medical school exams. GPT-4 answered the multiple-choice questions using 2 distinct prompt versions: detailed and short. The answers were analyzed using a quantitative approach and a qualitative approach. Focusing on incorrectly answered questions, we categorized reasoning errors according to the hierarchical framework of Bloom’s taxonomy. Results: GPT-4’s performance in answering exam questions yielded a high success rate: 93% (284/307) for the detailed prompt and 91% (278/307) for the short prompt. Questions answered correctly by GPT-4 had a statistically significant higher difficulty than questions answered incorrectly (P=.002 for the detailed prompt and P<.001 for the short prompt). Independent of the prompt, GPT-4’s lowest exam performance was 78.9% (15/19), thereby always surpassing the “pass” threshold. Our qualitative analysis of incorrect answers, based on Bloom’s taxonomy, showed that errors were primarily in the “remember” (29/68) and “understand” (23/68) cognitive levels; specific issues arose in recalling details, understanding conceptual relationships, and adhering to standardized guidelines. Conclusions: GPT-4 demonstrated a remarkable success rate when confronted with psychosomatic medicine multiple-choice exam questions, aligning with previous findings. When evaluated through Bloom’s taxonomy, our data revealed that GPT-4 occasionally ignored specific facts (remember), provided illogical reasoning (understand), or failed to apply concepts to a new situation (apply). These errors, which were confidently presented, could be attributed to inherent model biases and the tendency to generate outputs that maximize likelihood.