How Can AI Help Doctors? What Works, What Does Not and What the MMC Expects
Where AI saves doctors time in Malaysia, from letters to literature to teaching, where it fails, and the MMC duties on consent, data and validation that apply.
AI helps doctors most with language work: drafting referral letters and summaries, summarising papers, preparing teaching and structuring notes the doctor then corrects. It helps least where a judgement about one patient is needed. Under the MMC guideline the doctor stays accountable, must validate every recommendation and must not share patient data without consent.
Part of the guide: AI in Healthcare in Malaysia: Courses and Rules for Doctors, Nurses and Hospitals (2026)

On this page
- Where does AI save doctors time?
- Where does AI let doctors down?
- What does the MMC guideline expect when you use AI?
- What is a sensible way to start?
- What does a checked AI draft look like?
- How does this differ by specialty?
- How can you tell whether it is saving you time?
- What should you do next?
- Sources
- Frequently asked questions
AI is a drafting and summarising tool for doctors. It speeds up the part of the day that is writing and reading, and it needs checking every time. It does not examine a patient, and the Malaysian Medical Council puts accountability for the outcome on the doctor.
Checked on 22 October 2026 against the MMC guideline, Version 1/2025.
Where does AI save doctors time?
| Task | What AI does | Patient data in the tool? | Checking effort |
|---|---|---|---|
| Referral and discharge letter drafts | Turns your points into a structured letter | Avoid unless an approved institutional system | Read every line |
| Summarising a paper or guideline | Produces a short summary | None | Check against the paper |
| Journal club and teaching prep | Drafts slides, quizzes, handouts | None | Check facts and references |
| Patient education sheets | Rewrites a template in plain language and Bahasa Malaysia | None | Check clinical content and tone |
| Meeting and clinic notes | Transcribes and structures with an approved system | Yes, so approved system only | Correct before filing |
| Inbox and admin text | Drafts replies and policies | Avoid identifiable data | Read before sending |
The strongest published evidence is for documentation. Ambient AI scribes, which draft the note from the consultation, have been rolled out at scale in a large US health system (Tierney et al., NEJM Catalyst, 2024). We have not verified its figures here, and vendor or health-system claims of time saved are self-reported. They show potential, and they do not predict the result in a Malaysian ward with different systems and languages.

Where does AI let doctors down?
- It sounds sure when it is wrong. Invented references and plausible but wrong doses are the standard failures.
- It can carry bias. Four commercial chatbots all produced race-based medical misconceptions in one study (Omiye et al., npj Digital Medicine, 2023).
- Commercial prediction tools can underperform locally. A widely used sepsis model had a sensitivity of 33% in an external validation of 38,455 hospitalisations (Wong et al., JAMA Internal Medicine, 2021).
- Exam scores mislead. ChatGPT scored 52.4% to 75.0% on public USMLE sample questions (Kung et al., 2023), which says little about a real patient with an incomplete history.
What does the MMC guideline expect when you use AI?
Five duties matter in daily work:
- Training. Clause 3(4): seek appropriate training before use in clinical practice.
- Reasonable confidence in the tool. Clauses 3(2) and 3(3).
- Consent and disclosure. Clause 3(5)(b) of the MMC guideline before the consultation, and clause 3(12) when your view differs from the AI's.
- Confidentiality. Clause 3(6)(b): patient data must not be shared on any AI system without the patient's consent.
- Validation and accountability. Clauses 3(5)(c), 3(9) and 3(13).
The clause-by-clause reading is in The MMC guideline on AI, explained.

What is a sensible way to start?
- Pick one task from the table that involves no patient data.
- Use it for two weeks and keep a note of every error you catch.
- Write down your own checking routine (source check, dose check, read-aloud check).
- Only then consider a task with patient data, and only through a system your institution has approved.
If you teach, decide with your department where trainees may use AI and where they must work without it. See AI for medical students in Malaysia.
What does a checked AI draft look like?
The example below is constructed, with invented details, to show what checking means. Suppose you give a tool these points: "adult, three days of cough, no fever, chest clear, advise rest and fluids, review if worse". The draft returns a tidy letter. Read it with this list:
| Check | What to look for |
|---|---|
| Facts you gave | Is every fact present and unchanged ("no fever" must not become "fever")? |
| Facts you did not give | Has the tool added a finding, test or advice you never mentioned? Delete it |
| Numbers and names | Doses, dates, units, spellings |
| Tone and audience | Right for the reader, whether a patient or a colleague |
| Omissions | Is the safety-net advice still there? |
Added statements are the typical failure: the text sounds complete, so the extra line passes unnoticed. A routine of reading the output against your own points, line by line, catches most of it.
How does this differ by specialty?
The pattern holds across fields, and the examples shift. A surgeon may use AI for patient information leaflets and logbook summaries. A general practitioner may use it for plain-language explanations and referral letters. A paediatrician may need a leaflet at a parent's reading level in two languages. A hospital pharmacist or a public-health officer may use it for policy text. In each case the high-value tasks involve text with no patient data, or data inside an approved system. We give detailed examples for emergency departments in AI in emergency medicine.
How can you tell whether it is saving you time?
Track it for two weeks. Note the task, the minutes it took with AI including checking, and the minutes it took before. If checking takes longer than writing, the task is not a fit. The published scribe figures show savings at scale in one system, but your own log shows whether the result holds in your clinic.
What should you do next?
Our one-day AI course for doctors covers these uses on your own de-identified material and ends with a personal verification checklist. Read the overview in the AI in healthcare guide for Malaysia or book a free 30-minute call to plan a session for your department. Related questions: Can doctors use ChatGPT? and Can ChatGPT write clinical notes? For concrete cases, see Examples of AI in healthcare.
Doctors who want structured training can read what to look for in an AI course for doctors in Malaysia; nursing colleagues may start with Can AI replace nurses?.
Sources
(checked November 2026)
- Malaysian Medical Council, Guideline on the Ethical Use of AI in Medical Practice, Version 1/2025
- Tierney et al., Ambient Artificial Intelligence Scribes to Alleviate the Burden of Clinical Documentation, NEJM Catalyst 2024
- Wong et al., External validation of a widely implemented proprietary sepsis prediction model, JAMA Internal Medicine 2021 (PubMed)
- Omiye et al., Large language models propagate race-based medicine, npj Digital Medicine 2023 (PMC)
- Kung et al., Performance of ChatGPT on USMLE, PLOS Digital Health 2023
General information, not medical or legal advice.
Frequently asked questions
Can AI diagnose patients?
Some AI systems are built and validated for narrow diagnostic tasks and are regulated as medical devices. A general chat assistant is not validated for diagnosis. Under MMC clause 3(9), AI should complement, not replace, clinical judgement, and the doctor remains responsible for clinical decisions.
Will AI replace doctors in Malaysia?
Nothing in the MMC guideline or in the evidence we reviewed points to replacement. The guideline assumes a practitioner who evaluates every recommendation and can overrule it, and it requires human oversight for AI tools that affect patients.
What is the safest first use of AI for a doctor?
Tasks with no patient data and easy verification, such as drafting a teaching handout, summarising a paper you then read, or rewording a patient information sheet written from a template.
Do I have to tell patients if I use AI?
When AI is used in the patient's care, clause 3(5)(b) says to inform the patient before the consultation, where appropriate, about how it is used, the basis of its results, known risks and safeguards. Using AI privately for your own preparation involves no patient encounter.