AI in Medical Training

How Can AI Be Used in Medical Education? Seven Uses and Their Limits

Seven ways Malaysian medical schools and hospitals can use AI in teaching, simulation and assessment, with the review step each needs and the main risks.

  • By
  • Updated
  • 5 min read
  • Facts checked
Short answer

AI can draft clinical cases, build simulation scenarios, prepare question-bank drafts, give students feedback, summarise literature, support lesson planning and run virtual patients for history-taking practice. Each use needs a named expert to review the output before students see it, and none should involve identifiable patient data.

Part of the guide: AI in Healthcare in Malaysia: Courses and Rules for Doctors, Nurses and Hospitals (2026)

On this page
  1. What are the main ways to use AI in medical teaching?
  2. Can AI draft clinical cases and simulation scenarios?
  3. Can AI write exam questions?
  4. What can virtual patients do for students?
  5. What are the risks for a medical school?
  6. How do you start a pilot in a faculty or training programme?
  7. What does a four-week pilot look like in practice?
  8. How do you keep assessment fair?
  9. What does the MMC guideline mean for teachers?
  10. What should you do next?
  11. Sources
  12. Frequently asked questions

AI is useful in medical education because most of the work is language: cases, scenarios, feedback, summaries. It is also the safest place to begin, since no live patient decision depends on the output. The condition is that a person with subject expertise reviews everything before a student sees it.

Checked on 22 October 2026. Uses below describe what is practical today, not a validated curriculum.

What are the main ways to use AI in medical teaching?

Seven uses are practical now. The table pairs each with the check it needs.

Use What AI does Check that must follow
Case drafting Produces a first draft of a case with history, findings and teaching points Clinician corrects facts, local protocols and drug names
Simulation scenarios Drafts scenario scripts, branching events and debrief questions Faculty runs it once and edits before students use it
Question-bank drafts Writes candidate items and distractors Expert review against the exam blueprint
Feedback on written work Comments on structure and clarity of case write-ups Educator spot-checks and decides the grade
Literature summaries Summarises papers for journal clubs Every claim traced back to the source paper
Lesson planning Outlines sessions, handouts and slides Educator rewrites for the cohort
Virtual patients Plays a patient so students can practise history-taking Faculty reviews the persona and the transcript

Can AI draft clinical cases and simulation scenarios?

Yes, and this is the most time-saving use. A model can produce a case outline in seconds. It often gets details wrong in ways that look right: an implausible dose, a protocol from another country, a lab value that does not match the diagnosis. That is why case drafting works best as "AI writes, clinician corrects". For Malaysian teaching, the correction step is where local content comes in: Ministry of Health clinical practice guidelines, local drug availability and the way referrals actually work. Our approach to simulation is in Simulation-based training with AI.

Can AI write exam questions?

It can draft them. It cannot validate them. A drafted item still needs to match the learning outcome, have one defensible best answer, avoid cueing and sit at the right difficulty. The examiner's work does not shrink; it moves from writing to reviewing. A review step designed in advance is what makes the draft useful. Our course on assessment is built around this expert review loop, drawing on the lead physician's years on a national postgraduate examination board.

What can virtual patients do for students?

They let a student practise taking a history as often as they like, at any hour. A 2026 systematic review of large-language-model virtual patients for history-taking (Li and Lutfi, JMIR Medical Informatics) found that most systems simulated internal medicine and mental health conditions, with few covering multimorbidity or rare conditions. We found no head-to-head evidence in our reading that they match human standardised patients, so use them as extra practice and not as an assessment of competence.

What are the risks for a medical school?

  • Hallucination. Models state invented facts with confidence. Reference lists and guideline citations need checking against the source.
  • Bias. In a 2023 study, four commercial chatbots all produced examples of race-based medicine when asked probing questions (Omiye et al., npj Digital Medicine).
  • Over-reliance. MMC guideline clauses 3(10) and 3(11) ask practitioners to guard against skills degradation. A student who always asks a chatbot for a differential may not learn to build one.
  • Patient data. Clause 3(6)(b) forbids sharing patient data on an AI system without consent. Use constructed cases or properly de-identified material.
  • Academic integrity. Students already use the tools. A survey of 443 undergraduates at a Malaysian healthcare institute found MBBS students significantly more likely to use ChatGPT for academic purposes than dental and foundation students (one institution, so indicative only). Set the rule openly.

How do you start a pilot in a faculty or training programme?

  1. Choose one course and one use from the table, such as case drafting for a single block.
  2. Name the reviewer for each AI-assisted item and set a time budget for review.
  3. Write the data rule: constructed or consented, de-identified cases only.
  4. Tell students what AI was used for.
  5. Measure something simple: faculty hours saved, student feedback, number of errors caught in review.
  6. Decide after one block whether to continue, change or stop.

More on the build order is in Integrating AI into clinical training. The student side is covered in AI for medical students in Malaysia.

What does a four-week pilot look like in practice?

Week Activity Output
1 Choose one block (for example, a clinical reasoning block) and one use (case drafting). Agree the reviewer One-page scope
2 Educator drafts three cases with AI; a clinician reviews each and logs every correction Corrected cases plus an error log
3 Run the cases with one student group; collect feedback Student comments
4 Compare time spent, errors found and student response; decide next step Short decision note

The error log is the most useful product. If reviewers find few errors, either the tool is good for this task or the review is shallow, and a second reviewer on a sample tells you which. If they find many, that is the number to put in front of colleagues who want to roll it out faster.

How do you keep assessment fair?

Assessment is where AI help meets the highest stakes. A workable approach separates three roles:

  • Drafting: AI proposes candidate items from a blueprint.
  • Review: an expert checks every item for accuracy, one best answer, cueing and fit with learning outcomes.
  • Standard setting and grading: humans only.

Keep a record of which items were AI-drafted and which corrections were made. That record answers questions from external examiners and students later. Do not let a tool grade high-stakes work unsupervised.

What does the MMC guideline mean for teachers?

The guideline binds registered practitioners, and many educators are practising doctors. For them clause 3(4) on training, clause 3(6) on patient data and clauses 3(10) and 3(11) on over-reliance apply to teaching as well as clinical work. Clause 3(15) asks for disclosure of conflicts of interest, which matters if a faculty member recommends a commercial tool they have a stake in.

What should you do next?

Our AI in medical and clinical education course is a one- or two-day programme for faculties, hospitals and training programmes, led by two emergency physicians. It ends with a pilot plan for your department. The overview of the whole topic is in the AI in healthcare guide for Malaysia. To talk it through, book a free 30-minute call. Two practical guides: AI virtual patients for OSCE practice and AI for writing MCQs and OSCE stations.

Sources

(checked November 2026)

General information, not medical or legal advice.

Frequently asked questions

Can AI write exam questions for medical students?

It can draft items quickly, but every item needs expert review for accuracy, clarity and fit with the blueprint before use. Treat AI output as a first draft from an unvetted author, never as a validated question.

Is it safe to use real patient cases for teaching with AI?

Only with consent and de-identification that meets your institution's rules. The MMC guideline says patient data must not be shared on any AI system without the patient's consent. Constructed or fully synthetic cases avoid the issue.

Do students need to be told when AI was used to prepare teaching material?

Good practice is to say so, and some institutions require it. Agree a faculty-wide rule rather than leaving each lecturer to decide.

Will AI make students worse clinicians?

It can, if students use it to skip the thinking. [MMC guideline](https://mmc.gov.my/wp-content/uploads/2025/09/MMC-Guideline-on-The-Ethical-Use-of-Artificial-Intelligence-AI-.pdf) clauses 3(10) and 3(11) warn about over-reliance and skills degradation. Supervisors should decide where trainees work with AI and where they work without it.