ChatGPT versus human in generating medical graduate exam questions – An international prospective study
Open accessAbstract
Introduction: This is a prospective study on the quality of multiple-choice questions (MCQs) generated by the language model ChatGPT for the use in medical graduate examination. Methods: 50 MCQs were generated by ChatGPT with reference to two standard undergraduate medical textbooks (Harrison’s, and Bailey & Love’s). Another 50 MCQs were drafted by two university professoriate staffs using the same medical textbooks. All 100 MCQ were individually numbered, randomized and sent to five independent international assessors for MCQ quality assessment using a standardized assessment score on five assessment domains; namely, appropriateness of the question, clarity and specificity, relevance, discriminative power of alternatives, and suitability for medical graduate examination. Results: The total time required for ChatGPT to create the 50 questions was 20 minutes 25 seconds while it took two human examiners a total of 211 minutes 33 seconds for drafting the 50 questions. When a comparison of the mean score was made between the questions constructed by AI with those drafted by human, only in the relevance domain that the AI was inferior to human (AI: 7.56 +/- 0.94 vs human: 7.88 +/- 0.52; p = 0.04). There was no significant difference in question quality between questions drafted by AI versus human, in the total assessment score as well as in other domains. Questions generated by AI yielded a wider range of scores while those created by human were consistent and within a narrower range. Conclusion: ChatGPT has the potential to generate comparable-quality MCQs for medical graduate examination within a significantly shorter time.
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Integrity signals
Registry
Cites
Statistics
Claims the paper makes
The claims and statistics of this paper haven’t been checked yet.
Run the check — free, no account needed. It reads the open-access text and checks the sources it cites.
Claims other papers make about this one
Each is checked against this paper’s own text: does it say what the citing paper says it does?
Using course context and following best practices for prompting an LLM are prominent features of the pipeline.
“Using this particular context as well as following the best practices for prompting an LLM [6, 31] to generate an MCQ are the prominent features of the presented pipeline.”
From A Comparative Study of AI-Generated (GPT-4) and Human-crafted MCQs in Programming Education (2024) · cites this paper as [6]
The Methods section (page unavailable) describes giving ChatGPT instructions, a predefined exam context, and textbook reference text to generate MCQs. This supports using context and structured prompts, but the paper does not identify these instructions as LLM-prompting best practices or assess the citing authors’ pipeline. (LLM, conf=0.78)
Experimental — accuracy not yet measured
Self-check
Percentages, sums and reported differences re-computed from the paper’s own numbers.
The self-check needs the paper’s full text.
Cites
Artificial Intelligence in Education: A Review
Chen, L., Chen, P., & Lin, Z. (2020) · IEEE Access · doi:10.1109/access.2020.2988510 · Cited by 4,133 · cited 1× · see citing works
“In recent years, AI has been used to analyze student learning abilities and history, allowing teachers to create the best learning program for all students (Chen et al., 2020).”
BackgroundHaenlein, M., & Kaplan, A. (2019) · California Management Review · doi:10.1177/0008125619864925 · Cited by 2,655 · see citing works
On Faithfulness and Factuality in Abstractive Summarization
Maynez, J., Narayan, S., Bohnet, B., & McDonald, R. (2020) · Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics · doi:10.18653/v1/2020.acl-main.173 · Cited by 1,185 · see citing works
A Review of Multiple-Choice Item-Writing Guidelines for Classroom Assessment
Haladyna, T. M., Downing, S. M., & Rodriguez, M. C. (2002) · Applied Measurement in Education · doi:10.1207/s15324818ame1503_5 · Cited by 973 · see citing works
Flagged first, then by citations
Research works printed but not linked: 20
- Alkaissi, H., & McFarlane, S. I. (2023). Artificial Hallucinations in ChatGPT: Implications in Scientific Writing. Cureus, 15(2), e35179. doi:10.7759/cureus.35179 · DOI · Paper page
- Dion, V., St-Onge, C., Bartman, I., Touchie, C., & Pugh, D. (2022). Written-Based Progress Testing: A Scoping Review. Academic Medicine, 97(5), 747-757. doi:10.1097/acm.0000000000004507 · DOI
- Epstein, R. M. (2007). Assessment in Medical Education. New England Journal of Medicine, 356(4), 387-396. doi:10.1056/NEJMra054784 · DOI · Paper page
- Gierl, M. J., Bulut, O., Guo, Q., & Zhang, X. (2017). Developing, Analyzing, and Using Distractors for Multiple-Choice Tests in Education: A Comprehensive Review. Review of Educational Research, 87(6), 1082-1116. doi:10.3102/0034654317726529 · DOI
- Haladyna, T. M. (2004). Developing and validating multiple-choice test items: Routledge.
- Howard, F. M., Dolezal, J., Kochanny, S., Schulte, J., Chen, H., Heij, L., . . . Pearson, A. T. (2021). The impact of site-specific digital histology signatures on deep learning model accuracy and bias. Nat Commun, 12(1), 4423. doi:10.1038/s41467-021-24698-1 · DOI · Paper page
- Iñarrairaegui, M., Fernández-Ros, N., Lucena, F., Landecho, M. F., García, N., Quiroga, J., & Herrero, J. I. (2022). Evaluation of the quality of multiple-choice questions according to the students' academic level. BMC Med Educ, 22(1), 779. doi:10.1186/s12909-022- 03844-3 · DOI
- Kelly, F. J. (1916). The Kansas Silent Reading Tests. Journal of Educational Psychology, 7, 63- 80. doi:10.1037/h0073542 · DOI
- Kilgour, J. M., & Tayyaba, S. (2016). An investigation into the optimal number of distractors in single-best answer exams. Adv Health Sci Educ Theory Pract, 21(3), 571-585. doi:10.1007/s10459-015-9652-7 · DOI
- Kung, T. H., Cheatham, M., Medenilla, A., Sillos, C., De Leon, L., Elepaño, C., . . . Tseng, V. (2023). Performance of ChatGPT on USMLE: Potential for AI-assisted medical education using large language models. PLOS Digital Health, 2(2), e0000198. doi:10.1371/journal.pdig.0000198 · DOI · Paper page
- Loscalzo, J., Fauci, A. S., Kasper, D. L., Hauser, S., Longo, D., & Jameson, J. L. (2022). Harrison's Principles of Internal Medicine, Twenty-First Edition (Vol.1 & Vol.2): McGraw Hill LLC.
- Lowe, D. (1991). Set a multiple choice question (MCQ) examination. BMJ: British Medical Journal, 302(6779), 780.
- Mitchell, T. M. (2007). Machine learning (Vol. 1): McGraw-hill New York.
- Nassar, A. K., Reid, S., Kahnamoui, K., Tuma, F., Waheed, A., & McConnell, M. (2020). Burnout among Academic Clinicians as It Correlates with Workload and Demographic Variables. Behavioral Sciences, 10(6), 94. Retrieved from https://www.mdpi.com/2076-328X/10/6/94
- O'Connor, S., & ChatGpt. (2023). Open artificial intelligence platforms in nursing education: Tools for academic progress or abuse? Nurse Education in Practice, 66, 103537. doi:https://doi.org/10.1016/j.nepr.2022.103537 · DOI · Paper page
- Vegada, B., Shukla, A., Khilnani, A., Charan, J., & Desai, C. (2016). Comparison between three option, four option and five option multiple choice question tests for quality parameters: A randomized study. Indian J Pharmacol, 48(5), 571-575. doi:10.4103/0253-7613.190757 · DOI
- Williams, N. S., O'Connell, P. R., & McCaskie, A. W. (2018). Bailey & Love's Short Practice of Surgery: Taylor & Francis Group.
- Winston, P. H. (1992). Artificial intelligence: Addison-Wesley Longman Publishing Co., Inc.
- Wong, B. M., Levinson, W., & Shojania, K. G. (2012). Quality improvement in medical education: current state and future directions. Med Educ, 46(1), 107-119. doi:10.1111/j.1365-2923.2011.04154.x · DOI
- Yeoh, K. G. (2019). The future of medical education. Singapore Med J, 60(1), 3-8. doi:10.11622/smedj.2019003 · DOI
Cited by
Sorted by citations
A Comparative Study of AI-Generated (GPT-4) and Human-crafted MCQs in Programming Education
Doughty, J., Wan, Z., Bompelli, A., Qayum, J., Wang, T., Zhang, J., Zheng, Y., Doyle, A., Sridhar, P., Agarwal, A., et al. (2024) · Proceedings of the 26th Australasian Computing Education Conference · doi:10.1145/3636243.3636256 · Cited by 132 · see citing works
- Authors
- Cheung, B. H. H., Lau, G. K. K., Wong, G. T. C., Lee, E. Y. P., Kulkarni, D., Seow, C. S., Wong, R., & Co, M. T. H.
- Affiliations
- University of Hong Kong (HK)
- University of Edinburgh (GB)
- Ng Teng Fong General Hospital (SG)
- Ollscoil na Gaillimhe – University of Galway (IE)
- Corresponding author
- Co, Michael Tiong Hong
- ORCID
- Venue
- medRxiv
- Publisher
- openRxiv
- Published
- 2023-05-16
- OpenAlex
- W4376640309
- Cited by
- 10
- References
- 4
- Citations by year
- 2026: 3 · 2025: 2 · 2024: 4 · 2023: 1
- Type
- posted content
- Language
- en
- Open access
- green
- doi.org
- Licence
- cc-by-nd
- Locations
- medRxiv (acceptedVersion, cc-by-nd)
- Topics
- Health Sciences › Medicine › Health Informatics › Artificial Intelligence in Healthcare and Education
- Health Sciences › Medicine › Public Health, Environmental and Occupational Health › Innovations in Medical Education
- Physical Sciences › Computer Science › Information Systems › Artificial Intelligence in Education
- Keywords
- ChatGPT, large language models, multiple-choice questions, human-AI comparison, medical education, question relevance
About these checks
Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.
Statistics and arithmetic are recomputed, not judged.
Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.
What counts as a claim ›How we measure confidence ›How we check this ›
Are you an author, or do you see a problem with this page? Request a review
