Facilitator-Mediated Versus AI-Mediated Problem-Based Learning in Pharmacology Education: A Quasi-Experimental Study Among Second Year Medical Students

Synthesis: Shetty and Rao (2026) compared problem-based learning led by a human facilitator against the same sessions run with generative AI in a pharmacology course for second-year medical students. Across 136 completers, both arms improved significantly on pre/post MCQs and the between-group difference was not significant, so AI-mediated PBL matched facilitator-led learning gains. The divergences appeared in process: facilitator-led groups scored significantly higher on a group assessment rubric and on a critical-thinking rubric, while AI-led students reported better access to materials and rated one item on engagement higher. Faculty reported strong concerns about plagiarism and over-reliance on AI. The authors conclude that a hybrid model — AI support under facilitator oversight — is the most promising instructional design for medical education.

Key Findings

1. Among 150 second-year medical students, 136 completed the four-week pharmacology PBL study (68 facilitator-led, 68 AI-led), and 14 students who missed one or more sessions were treated as dropouts.
2. Both arms improved significantly on pre/post MCQ tests (facilitator p < 0.0001, AI p = 0.000025), but the between-group difference was not significant (p = 0.474), so AI-mediated PBL matched facilitator-led academic gains.
3. Facilitator-led PBL scored significantly higher on the overall group assessment rubric (26.51 vs 24.74, p = 0.035), including collaboration and teamwork (p = 0.021), response to feedback (p = 0.003), and problem solving (p = 0.041).
4. Critical thinking was significantly better in facilitator-led PBL (14.75 vs 13.09, p < 0.001) across explanation, information gathering and interpretation, systematic investigation, hypothesis forming, and arriving at a conclusion.
5. Overall student satisfaction was high in both groups and did not differ significantly (49.46 vs 50.50, p = 0.243), though AI-led students rated access to learning materials higher (p = 0.046).
6. Among 57 faculty responses, 96.5% cited plagiarism and 92.9% over-reliance as concerns, yet only 15.8% had addressed student AI use in their course policies or instructions.

How the study was run

The trial ran with 150 second-year medical students attending pharmacology practical sessions, divided into a facilitator-led batch and an AI-led batch of 75 each, then split into seven small groups of 10–12 students. Over four weeks, case-based clinical triggers on pharmacokinetics and pharmacodynamic concepts were released weekly for team collaboration, discussion, and critical thinking. A ten-item MCQ test was given at the start and end of the study; a five-item critical-thinking rubric adapted from the AAC&U VALUE framework and an eight-item group assessment rubric measured process, scored by eight subject experts. Students also completed a 12-item satisfaction survey, and faculty a three-part survey on AI use. Data were analyzed in SPSS 27 with paired t-tests, independent t-tests, and the Mann–Whitney U test. Results were reported for the 136 students (68 per batch) who completed the study.

Academic gains were comparable across both arms

The clearest finding is equivalence on academic performance. Facilitator-led PBL showed positive pre/post MCQ scores in 63.2% of students and AI-led PBL in 58.8%; negative scores were 25% in both. The paired t-tests were significant within each arm (facilitator p < 0.0001; AI p = 0.000025), with mean gains of 1.00 ± 2.78 and 1.03 ± 2.52 respectively. Crucially, the independent t-test between the two arms returned a non-significant difference (p = 0.474). The authors read this alongside earlier work by Hayashi et al. that found no written-examination difference between tutored and tutorless PBL groups, arguing that both modalities are effective vehicles for learning. For instructors weighing whether to substitute AI for a facilitator, this is a reassuring result on knowledge acquisition — but it says nothing about the processes the second rubric set out to measure.

Facilitation preserved group dynamics and critical thinking

Where facilitator-led PBL pulled ahead was in the quality of collaboration and reasoning. The overall group assessment rubric favored facilitator-led teams (26.51 ± 4.63 vs 24.74 ± 5.11, p = 0.035), driven by collaboration and teamwork (3.35 vs 3.06, p = 0.021), response to feedback (3.29 vs 2.91, p = 0.003), and problem solving (3.24 vs 2.97, p = 0.041), though effect sizes were small to medium. The critical-thinking rubric showed a wider gap: 14.75 ± 1.41 for facilitator-led versus 13.09 ± 1.70 for AI-led (p < 0.001), spanning explanation, information collection and interpretation, systematic investigation, hypothesis formation, and conclusion. Meanwhile AI-led students rated access to materials, discussion, and productivity higher in the satisfaction survey. Faculty echoed this tension: they felt student AI use improved discussion outputs (3.63 ± 0.94) while possibly undermining the goal of PBL, and rarely felt confident spotting AI use in submissions (0% very confident). The study's own preferred resolution is a hybrid of facilitator guidance and AI scaffolding.

What this means for practice

• Instructors. Do not assume AI mediation protects critical thinking: facilitator-led groups scored 14.75 against 13.09 on the critical-thinking rubric (p < 0.001), so keep a human in the loop when reasoning quality is the target.
• Instructors. Expect collaboration and feedback to be the first casualties of going AI-only: the group rubric favored facilitator-led teams on teamwork (p = 0.021) and response to feedback (p = 0.003), so retain structured peer interaction alongside any AI support.
• Medical educators. Adopt a hybrid model rather than a substitution: academic gains were equivalent between arms, so AI can carry resource access and discussion while a facilitator carries scaffolding and norm-setting.
• Medical educators. Set AI-use policy before the next PBL block: only 15.8% of the 57 surveyed faculty had addressed AI use in course policies, while 96.5% worried about plagiarism and 92.9% about over-reliance.
• Researchers. Treat process outcomes as the discriminating measure: since the MCQ between-group test was null (p = 0.474), future comparisons should foreground rubrics, reasoning, and self-directed learning.

Limitations

• This is a Research Square preprint and the paper reports no peer review, so the findings have not been through journal evaluation.
• The quasi-experimental design had no random allocation, so selection bias and baseline differences between the two 75-student batches cannot be ruled out.
• The study is small and short (136 completers over four weeks at one institution), and the authors state that generalizing to other medical schools is not appropriate.
• Differences in students' AI familiarity, digital literacy, and prior PBL experience may have affected both performance and satisfaction, and satisfaction responses are subjective.
• Faculty responses were limited to 57 surveys, which the authors note constrains conclusions about faculty awareness of AI use.

Citation

Shetty, P., & Rao, M. (2026). Facilitator-Mediated Versus AI-Mediated Problem-Based Learning in Pharmacology Education: A Quasi-Experimental Study Among Second Year Medical Students. Research Square preprint.