Will AI Replace Doctors? What Medical AI Already Does and What Stays Human

AI is not replacing doctors, but it already performs specific medical tasks. The FDA's list of AI-enabled medical devices held 1,614 authorizations in its September 2026 update, and 1,230 of them are radiology devices. Ambient AI scribes draft visit notes that physicians edit. The examination, the consent conversation, the clinical decision and the responsibility for it stay with the physician.

This is an article about work and the labour market, not medical advice. It covers which parts of a doctor's job software already handles, which parts it does not and why, what changes in the working day, and what it means for a clinic that also has to run as a business.

Which medical tasks does AI already do?

AI in medicine is a collection of narrow tools, each authorized or deployed for one job, rather than a general system that practises medicine. The clearest public record is the FDA's list of AI-enabled medical devices. It starts with two entries in 1995, passed 100 authorizations a year in 2020, and recorded 335 in 2025 alone. Most of those devices read images.

The FDA notes that its list is not comprehensive: devices are identified mainly by AI-related terms in their authorization summaries, and the agency says it will explore how to tag devices that use large language models. The count is a floor, not a census.

Why did AI arrive in radiology first?

Radiology leads because its input is already digital, standardised and backed by years of written reports. A CT study is a file, and past reports give a model many labelled examples of the finding it has to detect. That makes a narrow task, such as flagging a possible bleed on a head scan, both trainable and testable, and a device authorization needs a testable claim.

Even there, each tool covers its own intended use, not the whole specialty. Every row on the FDA list is a specific product from a specific company, with its own submission number and lead review panel. A device that flags one kind of finding says nothing about the rest of the image, and the radiologist still reads the study and signs the report.

What stays with the doctor, and why?

The parts of medicine that stay human are the ones that need hands in the room, a conversation built on trust, or a person who answers for the outcome. The reasons are structural, not only a matter of model quality:

• Physical examination and procedures. Palpating an abdomen, listening to a chest or performing a procedure needs a trained person at the bedside.
• Consent. WHO's 2021 guidance on ethics and governance of AI for health calls for valid informed consent, and explaining options and risks to a patient is a conversation between people.
• Human control. The same WHO guidance makes protecting human autonomy its first principle: people must remain in control of health-care systems and medical decisions.
• Error checking. WHO's 2024 guidance warns that large multi-modal models can produce false, inaccurate, biased or incomplete statements, and can encourage automation bias, where errors are overlooked or difficult choices are handed to the model.
• Accountability. A clinical decision is made and signed by a licensed professional; the software on the FDA list is authorized as a device, not licensed to practise.
• Scope of authorization. Each listed device was reviewed as one product for one purpose, which is a long way from a system that takes a history, examines, diagnoses and treats.

What changes in a doctor's working day?

The first thing that changes is the paperwork around care, not the care itself. The Permanente Medical Group's evaluation describes documentation in the electronic health record as a major driver of physician burnout, and its ambient AI scribe transcribes the visit through a smartphone microphone without retaining the audio. Within ten weeks of the October 2023 rollout, 3,442 of its physicians had used the tool, and they reported more personal patient interactions and less after-hours clerical work.

In practice the job shifts in a few concrete ways:

• Editing a drafted note instead of typing one from scratch.
• Treating an AI flag on a scan or ECG as a second look that has to be confirmed or overruled.
• Knowing what each tool is authorized for, and where its output stops being reliable.
• Spending more of the consultation facing the patient rather than the screen.

The skill that gains value is supervision: recognising when the draft or the flag is wrong. That is the same pattern the guide to AI and lawyers describes for legal work, and the one the overview of which jobs AI will replace finds across professions: tasks move before jobs do.

What does this mean for a clinic as a business?

For the owner of a clinic or a private practice, the fastest and lowest-risk automation gains are administrative, not clinical. They touch no diagnosis, involve no medical device, and usually free the front desk more than the doctor. These are the ones I see clinics start with:

• Online booking that writes straight into the clinic calendar - compared in the appointment scheduling software guide.
• Automatic reminders before each appointment by SMS, WhatsApp or email, covered in how to automate appointment reminders.
• Intake forms that fill in patient details before the visit instead of at the counter.
• Moving finished documents, such as referrals, summaries and invoices, into the right system without retyping them.

Patient information is sensitive data, so the first step in any of these projects is mapping which fields move between which systems, before choosing a tool. For what such a project typically costs, see the pricing page.

Sources

• U.S. Food and Drug Administration - Artificial Intelligence-Enabled Medical Devices (list, content current as of 09/04/2026)
• World Health Organization - AI ethics and governance guidance for large multi-modal models, January 2024
• Tierney et al. - Ambient Artificial Intelligence Scribes to Alleviate the Burden of Clinical Documentation, NEJM Catalyst, 2024
• UN News - WHO issues first global report on AI in health and six guiding principles, June 2021