Healthcare is one of the areas where AI could do the most good — and where mistakes carry the highest stakes. AI is no longer experimental in medicine: hundreds of AI-enabled devices have been authorized by regulators, hospitals use AI to draft clinical notes, and AI-designed molecules are in clinical trials. Here is where it is actually used.

1. Medical imaging

Imaging is the most mature area. AI systems help detect and prioritize findings in X-rays, CT, MRI, mammograms and retinal photos — for example flagging possible strokes, lung nodules or diabetic eye disease. The US Food and Drug Administration maintains a public list of AI-enabled medical devices it has authorized, and radiology accounts for the large majority of them.

Typical uses are assistive: AI acts as a second reader or reorders worklists so urgent cases are seen first, while radiologists make the diagnosis.

2. Clinical documentation

Doctors spend a large share of their time on documentation. Ambient AI scribes listen (with consent) to patient visits and draft clinical notes, which clinicians review and sign. Health systems adopting them report reduced after-hours charting and better patient interaction, though careful checking for errors remains essential.

3. Clinical knowledge and decision support

Language models can answer medical questions at a high level. Google’s research on Med-PaLM, published in Nature in 2023, showed a model performing well on US medical licensing-style questions, while also documenting gaps compared with clinicians. In practice, AI decision support is used for summarizing records, suggesting differential diagnoses and checking guidelines — always with clinician oversight.

4. Triage and patient-facing tools

Symptom checkers, patient message drafting, appointment scheduling and multilingual communication tools help manage demand. These raise the bar for safety: patients may act on advice without a clinician involved.

5. Operations

Predicting bed occupancy, staffing needs, no-shows and supply use helps hospitals run more efficiently.

6. Research and drug discovery

From protein structure prediction to identifying trial candidates, AI accelerates biomedical research — see our article on AI in drug discovery.

The risks

  • Bias: models trained on data from some populations may perform worse for others.
  • Generalization: a model that works in one hospital can fail in another with different equipment or patients.
  • Automation bias: clinicians may over-trust AI outputs.
  • Hallucinations in language-model tools.
  • Privacy of highly sensitive health data.
  • Accountability: who is responsible when an AI-influenced decision goes wrong?

The World Health Organization has published guidance on the ethics and governance of AI for health, emphasizing human oversight, transparency, safety and equity.

Regulation

AI that diagnoses or guides treatment is usually regulated as a medical device. In the EU, such systems are also high-risk under the AI Act. Regulators are developing approaches for AI that updates over time.

What patients can expect to notice

Most healthcare AI is invisible to patients, but a few changes are increasingly common:

  • Your doctor looking at you instead of a keyboard. Where ambient scribes are used, you may be asked for consent at the start of a visit for the conversation to be recorded and summarized. You can usually decline.
  • Faster results for urgent scans. AI triage can move suspected emergencies, such as possible strokes or collapsed lungs, to the top of a radiologist’s list.
  • Drafted replies in patient portals. Some health systems use AI to draft responses to patient messages, which clinicians review and edit before sending.
  • Screening programs. In some countries, AI is being evaluated as one of the readers in breast cancer or diabetic eye screening, which can reduce waiting times.

It is reasonable to ask your provider whether AI is used in your care, what it does, and who checks its output. Good providers will be able to answer clearly.

What to watch next

  • Independent studies of AI scribes on accuracy, workload and patient outcomes.
  • Regulatory frameworks for continuously learning AI devices.
  • Large language models approved for specific clinical tasks.
  • Evidence from prospective trials rather than retrospective studies.

Key takeaways

  • Medical imaging is the most mature use, accounting for most authorized AI devices.
  • Ambient AI scribes draft clinical notes for clinicians to review.
  • Language models show strong medical knowledge but need oversight in practice.
  • Bias, generalization, automation bias and privacy are central risks.
  • Clinical AI is generally regulated as a medical device and as high-risk AI in the EU.

Frequently asked questions

Can AI diagnose diseases?

Some authorized AI devices detect specific conditions in images, usually supporting rather than replacing clinicians. General diagnosis by chatbots is not a substitute for medical care.

Is it safe to ask a chatbot about health symptoms?

It can help you understand information and prepare questions, but it can be wrong. Seek professional care for symptoms, and emergency services for urgent problems.

Will AI replace doctors?

Current evidence points to AI taking over specific tasks — documentation, triage, image screening — while clinicians remain responsible for diagnosis and treatment.