Advancing Problem-Based Learning in Biomedical Engineering in the Era of Generative AI

Synthesis: This article presents a modularized, implementation-ready problem-based learning (PBL) framework tailored to biomedical AI, positioning generative AI (GenAI) as a guarded knowledge-summarization and coding-support module rather than an answer engine. A three-year case study (2021–2023) across Georgia Tech and Emory University engaged 248 students in interdisciplinary teams on real biomedical AI problems, with GenAI governed by disclosure, source-anchoring, verification, and version-logging policies. The framework reframes GenAI as a scaffold that accelerates baseline learning and frees students for higher-order critical thinking and innovation. A portable replication package (syllabi, milestones, rubrics, team procedures, and AI-usage templates) accompanies the design for adoption under variable resources.

Key Findings

Modularized PBL+AI framework: The design comprises four interdependent modules—problem formation via authentic biomedical problem briefs with curated deidentified datasets, AI-supported knowledge inquiry, problem-solving with robustness and reproducibility evaluation, and presentation through written reports, oral demos, and peer review.
Grade distribution shift: Post-integration cohorts (2021–2023) showed higher A-rates (66.4% vs. 39.1% control, Δ = +27.3 points, p = 0.042) and fewer low grades (6.1% vs. 22.4%), a trend that persisted after excluding the COVID-19-affected year (A-rate 67.4% vs. 39.1%, p = 0.037).
Coding fluency as an enabler: Students with more coding experience performed better (exploratory OLS β ≈ +1.88 points per year, p = 0.022), suggesting foundational coding readiness enables productive GenAI use for ideation and code assistance.
High research productivity: Implementation-focused PBL yielded 16 student-authored peer-reviewed publications addressing biomedical AI problems (e.g., wearable COVID-19 detection, graph-based single-cell RNA severity classification, synthetic augmentation for transplant-rejection imaging).
Strong teamwork outcomes: Peer evaluations remained high across courses (BHI 89.8–94.6%, MIP 88.8–90.6%, biostatistics ~92–94%), indicating effective collaboration under AI-augmented PBL.
Guardrailed AI use: Approved institution-reviewed tools only (DeepSeek banned), no PII/PHI submitted, mandatory disclosure and validation logs, source-anchored literature claims, and code provenance comments preserved both transparency and PBL's learner-directed intent.

Implications for AI in Education

The study demonstrates that GenAI is most valuable in PBL when positioned as a baseline knowledge assembler and coding scaffold rather than a source of direct answers, strengthening rather than bypassing the pedagogical intent of PBL. Its guardrail architecture—disclosure logs, source anchoring, verification, and version logging—offers a transferable model for responsible adoption in technical and specialty courses. Five scalability opportunities (guided problem exploration, knowledge acquisition, prototype support, automated formative assessment, and personalized learning pathways) map directly onto the faculty-resource and curricular-update bottlenecks that historically limited PBL. However, the findings also flag equity concerns (GenAI bias toward nonnative speakers and diverse backgrounds), hallucination risk in accuracy-critical biomedical domains (mitigated via retrieval-augmented generation and critical-evaluation training), and the need to balance AI assistance with periods of independent problem-solving to preserve critical thinking.

Connected Concepts

Connected Articles

Citation

Nnamdi, M.C., Tamo, J.B., Marteau, B., Shi, W., & Wang, M.D. (2026). Advancing problem-based learning in biomedical engineering in the era of generative AI. IEEE Transactions on Education, 69(2).