← Back to Articles
The short answer

For a senior healthcare leader, AI literacy means being able to interrogate claims rather than build systems. Concretely: reading a validation study and identifying whether the population resembles yours, recognising when an aggregate accuracy figure is hiding subgroup variation, knowing what questions post-market monitoring should answer, and being able to tell when a vendor has changed the subject. It is a judgment skill applied to a new domain, not a technical qualification.

What it is not

It is not knowing how a transformer works. Executives are routinely given introductory technical content on the theory that understanding the mechanism produces good judgment about the application. It does not. A leader who can describe attention heads and cannot spot a validation study run on the wrong population is not equipped.

It is also not fluency with the tools. Being a capable user of a chat assistant tells you very little about whether a clinical AI system is safe to deploy.

A workable standard

We would define it as five capabilities:

  • Read a validation claim and locate its boundary. What population, what setting, what threshold, and what happens outside those conditions.
  • Recognise aggregation. Understand that a single accuracy figure can conceal materially different performance across subgroups, and know to ask for the breakdown.
  • Understand degradation. Know that model performance changes after deployment, why, and what monitoring would surface it.
  • Locate accountability. Know which regulatory role your organisation occupies for each system, and what that obliges.
  • Detect evasion. Notice when a question has been answered with an adjacent one. This is the skill that most reliably separates leaders who buy well from those who buy confidently.

Why building something is the shortcut

The fastest route to this capability, in our experience, is not a course about AI. It is building a working system yourself, badly, and watching it fail in a way you have to diagnose.

Someone who has assembled an agent, seen it produce a confident wrong answer, and had to work out why, has an intuition for model behaviour that no amount of explanation transfers. They stop treating outputs as authoritative. That single shift does more for procurement judgment than a term of technical content.

Our view

The AMA's 2026 survey found 92% of physicians wanting more AI education and 27% having received none. The gap is real. Our concern is with what most organisations will buy to close it.

An awareness session produces the sensation of having addressed the problem while leaving the capability unchanged, and it is worse than doing nothing because it discharges the obligation. If your AI literacy programme can be completed in an afternoon and produces no artefact, it has taught nobody to interrogate anything.

We would also target differently than most. The clinicians are asking. The people who more often lack this capability, and whose decisions carry further, are the executives approving procurement and the board accepting the risk. A clinician using a tool poorly affects their clinic. A leadership team buying the wrong platform commits the institution for years.

Sources

  1. American Medical Association, 2026 Physician Survey on Augmented Intelligence, 1,692 physicians surveyed 15 January to 2 February 2026. ascopost.com
  2. Ministry of Health Singapore, AIHGle 2.0, on developer, deployer and user responsibilities. go.gov.sg/aihgle

Building This In Your Team

Cohort programmes that produce this capability, not awareness of it.

Start a Conversation →