Where AI is actually used in healthcare

AI in healthcare: an overview · Lesson 1 / 20

Three layers of application

When people say "AI in medicine", they picture a system that reads a scan and names a condition. That is the most visible form, not the most common one. Real deployments spread across three layers that differ sharply in risk, in payback speed and in who actually decides.

The layers

Administrative. Visit documentation, speech turned into text, routing of incoming requests, appointment booking, shift and equipment planning, answers to routine organisational questions. A mistake here costs time and patience rather than health.
Clinical support. Highlighting a region on an image, ordering a queue of studies by priority, reminding about a skipped protocol step. The system decides nothing; it changes the order in which a human pays attention.
Clinical decision. Tools whose output directly shapes management. This is where regulation, verification and accountability concentrate, and this is where deployments are rarest.

Why an administrator should care

A manager rarely chooses between adopting AI and not adopting it. The real choice is which layer to start in. Starting where mistakes are measured in health is expensive, slow and demands expertise most organisations lack. Starting in the administrative layer is cheaper, and staff notice the result within weeks. The map is therefore a way to set priorities, not a taxonomy.

Quick classification of any application:
1. What does the system output - text, a priority, a label, a recommendation?
2. Who receives it - front desk, clinician, patient?
3. What happens if the output is wrong?
4. Who catches the error, and at which step?
5. How does work continue when the system is unavailable?

Insight. The layer, not the technology, sets the verification burden. The same language model used for visit notes and for management hints represents two completely different projects with different costs of error.

Common mistake. Assuming that only diagnostic tools count as real AI. Organisations then ignore applications that deliver value today and spend years debating ones they are not ready for.

Pro tip. List the applications currently under discussion in your organisation and sort them into the three layers. Usually the debate is about layer three while the pain sits in layer one.

Scope of this course

This course explains how AI is applied in healthcare and how to evaluate such applications. It does not teach diagnosis, treatment or interpretation of investigations, and it does not replace medical education. Clinical decisions remain with the clinician. Everything here concerns organisation of work, evaluation of tools and handling of data, not patient management.

Cheat sheet

Three layers: administrative, clinical support, clinical decision.
The layer determines the cost of error and the depth of verification.
Most of today's quick value sits in the administrative layer.
This is an overview course: clinical decisions stay with the clinician.

1. What primarily determines how much verification an AI tool needs?

2. Which application belongs to the administrative layer?

3. What does this course explicitly not teach?

🔒 Answer the question correctly to move on to the next lesson.