Chatbots’ Role in Generating Single Best Answer Questions for Undergraduate Medical Student Assessment: Comparative Analysis
Bibliographic record
Abstract
Background: Programmatic assessment supports flexible learning and individual progression but challenges educators to develop frequent assessments reflecting different competencies. The continuous creation of large volumes of assessment items, in a consistent format and comparatively restricted time, is laborious. The application of technological innovations, including artificial intelligence (AI), has been tried to address this challenge. A major concern raised is the validity of the information produced by AI tools, and if not properly verified, it can produce inaccurate and therefore inappropriate assessments. Objective: This study was designed to examine the content validity and consistency of different AI chatbots in creating single best answer (SBA) questions, a refined format of multiple choice questions better suited to assess higher levels of knowledge, for undergraduate medical students. Methods: This study followed 3 steps. First, 3 researchers used a unified prompt script to generate 10 SBA questions across 4 chatbot platforms. Second, assessors evaluated the chatbot outputs for consistency by identifying similarities and differences between users and across chatbots. With 3 assessors and 10 learning objectives, the maximum possible score for any individual chatbot was 30. Third, 7 assessors internally moderated the questions using a rating scale developed by the research team to evaluate scientific accuracy and educational quality. Results: In response to the prompts, all chatbots generated 10 questions each, except Bing, which failed to respond to 1 prompt. ChatGPT-4 exhibited the highest variation in question generation but did not fully satisfy the "cover test." Gemini performed well across most evaluation criteria, except for item balance, and relied heavily on the vignette for answers but showed a preference for one answer option. Bing scored low in most evaluation areas but generated appropriately structured lead-in questions. SBA questions from GPT-3.5, Gemini, and ChatGPT-4 had similar Item Content Validity Index and Scale Level Content Validity Index values, while the Krippendorff alpha coefficient was low (0.016). Bing performed poorly in content clarity, overall validity, and item construction accuracy. A 2-way ANOVA without replication revealed statistically significant differences among chatbots and domains (P<.05). However, the Tukey-Kramer HSD (honestly significant difference) post hoc test showed no significant pairwise differences between individual chatbots, as all comparisons had P values >.05 and overlapping CIs. Conclusions: AI chatbots can aid the production of questions aligned with learning objectives, and individual chatbots have their own strengths and weaknesses. Nevertheless, all require expert evaluation to ensure their suitability for use. Using AI to generate SBA prompts us to reconsider Bloom's taxonomy of the cognitive domain, which traditionally positions creation as the highest level of cognition.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.080 | 0.275 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.006 | 0.003 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.002 | 0.003 |
| Open science | 0.002 | 0.004 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".