Where AI is used in health care today
About 25 minutes · plain language
AI is already used in many parts of health care. Some uses carry little risk; others can directly harm a patient. In this lesson you will learn where AI is used today and how to judge how risky a particular use is.
- Name the main uses of AI in care, education, research and administration.
- Place a tool on the scale from low to high risk.
Key terms
- Clinical decision support (CDS)
- Software that gives clinicians information or suggestions to help with decisions about a patient.Example: An early-warning score that suggests escalation.
- Early-warning score
- A score that combines vital signs and other data to flag patients at risk of deterioration.Example: An AI version that also uses blood results and trends.
- Triage tool
- A tool that helps decide how urgently a patient needs care.Example: An emergency-department tool that sorts patients by urgency.
- Ambient documentation
- Tools that listen to a consultation and draft the clinical note.Example: A speech-based tool that writes a draft nursing admission note.
- Medical device software
- Software intended for diagnosis, prevention, monitoring or treatment, which regulators may classify as a medical device.Example: An app that detects skin cancer from photos.
- Risk level
- How likely and how serious harm would be if the tool is wrong.Example: Higher for a dosing suggestion than for a scheduling tool.
- Human review
- A person checks the AI output before it is used.Example: A nurse reads and edits an AI-drafted discharge summary.
- AI scribe (ambient scribe)
- A tool that listens to a consultation, with consent, and drafts the clinical note for a person to check and sign.Example: A tool that drafts a nursing admission note while the nurse talks with the patient.
- AI agent
- An AI system that can plan steps and take actions in other software, such as searching, filling forms or sending messages, not only answer a question.Example: A tool that books a follow-up appointment and drafts the referral letter by itself.
Where AI is used today
- Clinical care: image reading, early-warning and deterioration scores, triage, remote monitoring of patients at home.
- Documentation: speech-to-text, draft notes, discharge summaries and referral letters.
- Patients: symptom checkers, information chatbots, appointment reminders.
- Education: tutoring, instant feedback, simulation, question banks.
- Research: literature searching and screening, data analysis, writing support.
- Management: staffing, bed flow, supply chains.
Two questions decide most of the risk
First: how directly does the output affect a patient? A draft email that goes to a manager is low risk. A suggestion that changes a dose is high risk. Second: how likely is a person to notice a mistake? An error in a draft letter is easy to spot. A deterioration alert that never fires is invisible. The figure shows this as a ladder: the higher the rung, the more evidence, training and monitoring you need.
Regulation follows risk
Much of the AI that informs diagnosis or treatment is regulated as a medical device, so makers must show evidence before selling it. But not everything is covered: some decision-support software and general-purpose chatbots fall outside device rules. So “it is on the market” does not mean “it has been tested for our use”. The EU AI Act also sorts AI by risk and sets stricter duties for high-risk uses (Lesson 7).
Benefits are real too
Used well, AI can reduce documentation time, catch patterns humans miss, and give patients faster information. The goal of this course is not to avoid AI but to use it where it helps and check it where it can harm.
AI scribes: the fastest-growing tool
AI scribes are now the AI tool health workers meet most often. Early evaluations report less time spent on notes and less after-hours documentation, but the drafts can still contain errors: a symptom the patient never mentioned, a missed allergy, or a statement given to the wrong person. National guidance, such as NHS England’s, asks organisations to name a clinical safety lead, assess data protection, tell patients and ask for consent, and monitor errors after launch.
For nurses, three habits matter: tell the patient the tool is listening and let them say no; read the whole draft before you sign it, because you sign it as your own record; and report recurring errors.
AI agents: from answers to actions
Most tools so far answer a question or draft a text, and a person decides what to do with it. AI agents go further: they plan several steps and act in other systems, for example searching records, ordering a test or sending a message. This saves time but raises the stakes, because an error can turn into an action before anyone reads it. Before any agent touches patient care, ask: what exactly is it allowed to do, which actions need a person’s approval first, and is every action logged so it can be checked and undone?
Case: two tools on one ward
A medical ward starts two AI tools in the same month. The first drafts handover notes from the electronic record. The second is a deterioration score. The ward manager treats them differently. For the handover tool, nurses are asked to read and correct every draft. For the deterioration score, the ward asks for local evidence first, trains every nurse, agrees who can override it, and reviews missed cases each month.
Common mistakes
- Using the same level of trust for every tool.
- Assuming a tool is approved and safe because it is sold.
- Ignoring errors that nobody would notice.
Check your understanding
1. Which is higher risk: an AI that drafts staff rotas, or an AI that suggests insulin doses? Why?
The insulin-dose AI, because its output directly affects a patient and an error could cause serious harm.
2. Does “approved and on sale” mean a tool has been tested for your patients?
No. Some tools fall outside device rules, and approval does not show how a tool performs in your setting.
3. An AI scribe’s draft says the patient “denies chest pain”, but the patient never mentioned chest pain. What should the nurse do?
Correct the note before signing, because the signed note is the nurse’s own record, and report the error so the tool can be monitored.
Try it
- List the AI tools used in your ward, clinic or faculty and place each on the risk ladder.
Remember
- AI is used in care, documentation, education, research and management.
- Risk depends on how directly the output affects a patient and how easily errors are noticed.
- Higher risk needs more evidence, training and monitoring.
- AI scribes save time, but you sign the note: read all of it.
- AI agents act, not only answer: decide which actions need human approval.
- World Health Organization. Ethics and governance of artificial intelligence for health: WHO guidance. Geneva: WHO; 2021. www.who.int
- International Medical Device Regulators Forum. Good machine learning practice for medical device development: guiding principles (IMDRF/AIML WG/N88 FINAL:2025). IMDRF; 2025. imdrf.org
- Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act). Official Journal of the European Union. 2024; as amended in 2026. eur-lex.europa.eu
- Topol EJ. High-performance medicine: the convergence of human and artificial intelligence. Nat Med. 2019;25(1):44-56. pubmed.ncbi.nlm.nih.gov
- NHS England. Guidance on the use of AI-enabled ambient scribing products in health and care settings. London: NHS England; 2025 (updated 2026). www.england.nhs.uk
- Tierney AA, Gayre G, Hoberman B, et al. Ambient artificial intelligence scribes to alleviate the burden of clinical documentation. NEJM Catal Innov Care Deliv. 2024;5(3). doi.org
Lesson text is our own plain-language summary of these sources; it is not endorsed by their authors. Check the original before relying on any detail.