The AI Learning Hub Journal
◆ Start here

Read This First: Scope and Safety

What this course is, and what it is notan educational orientation — vocabulary and judgement, never clinical guidanceWHAT THIS COURSE DOES PROVIDEA shared vocabularyThe words these systems are described with, so you canfollow the discussion and ask a precise question.How to read the evidence criticallyWhat a study design can and cannot support, and where aclaim runs ahead of what was actually measured.Categories of capability and failureWhat this class of tool tends to do well, and the shapesits mistakes tend to take.Governance conceptsOversight, accountability, monitoring and withdrawal —the ideas, not any one institution’s policy.WHAT IT EXPLICITLY DOES NOTClinical guidanceNothing here tells you what to do for the patient infront of you.Thresholds, criteria or protocolsNo cut-offs, and no pathway you could lift from a slideinto practice.Product endorsementNo tool is recommended, compared, ranked or ruled outanywhere in this material.Regulatory or legal adviceObligations differ by place and change over time; thisis orientation, not compliance.THE STANDING RULE FOR EVERYTHING THAT FOLLOWSClinical judgement, institutional protocol and regulatory approval govern every real decisionNothing in this course displaces any of them, and nothing here substitutes for themHOW TO READ IT WHEN SOMETHING STARTS TO SOUND LIKE ADVICEIf it sounds like a recommendationIt is not one. Read it as howpeople reason about thesetools, not as what to do.If you need a threshold or a ruleThis is the wrong source. Thosecome from your protocol andyour own clinical judgement.If it would touch a patientIt goes through the institution’sown governance long before itgoes anywhere near care.
Educational orientation only — clinical judgement, local protocol and regulatory approval always govern

This Course Is Orientation, Not Clinical Guidance

Read this before anything else. This course is educational orientation for professionals who need to think clearly about AI in healthcare. It is not clinical guidance, not legal advice, and not a regulatory opinion. No output from an AI system — and nothing written in these lessons — substitutes for professional clinical judgement, your institution's protocols, or the approvals a product needs before it touches patient care. Nothing here should be used to make a decision about an individual patient. Where a lesson describes what a category of tool can do, that is a description of published capability patterns, not a recommendation to deploy it. A real deployment decision requires validation on your own local data plus formal institutional governance sign-off.

  • Educational orientation only — no lesson here constitutes clinical, legal, or regulatory advice
  • No AI output replaces clinical judgement, institutional protocol, or the relevant regulatory approval
  • Deployment is never a reading decision: it requires local validation on your own population and governance sign-off
  • This course names capability categories and evidence patterns — it never gives a threshold, dose, protocol, or triage rule you could act on

What You Will Deliberately Not Find Here

Some omissions are intentional. You will not find diagnostic cut-offs, dosing logic, treatment pathways, or triage rules — not because they are secret, but because a general course is the wrong place for anything a reader might apply to a patient without local validation behind it. You will also not find claims that a named commercial product is safe, effective, or approved for your setting. Regulatory status is jurisdiction-specific, version-specific, and intended-use-specific; it changes, and a course cannot track it for you. What you will find is the structure of the field: what the categories of tool are, what evidence exists for each, where that evidence is weak, and what questions to ask before anything reaches a patient.

  • No clinical thresholds, protocols, or decision rules — those belong to your institution and its evidence review
  • No claims about specific commercial products being approved or effective in your setting
  • What you get instead: categories, evidence patterns, failure modes, and the questions that separate a good deployment from a bad one

Saying "The Evidence Is Thin" Is the Point

Healthcare AI is discussed in two registers: vendor optimism and blanket scepticism. Neither is useful to someone who has to make a decision. The honest position varies sharply by application. For some narrow tasks there is substantial published evidence, including prospective work. For others there are mainly retrospective studies on curated datasets, which tell you far less than they appear to. For a few widely promoted applications, the evidence is genuinely thin or mixed, and some well-publicised systems have underperformed badly once studied independently. This course states which is which. Being able to say "we do not know yet" about a specific claim is the most valuable skill it can give you.

  • Evidence quality varies by application, not by technology — treat every claim separately
  • Retrospective performance on curated data is the weakest common form of evidence, and the most frequently cited
  • Independent evaluation has repeatedly deflated systems that looked strong in vendor or single-site studies
  • "We do not know yet" is a legitimate, defensible answer in a procurement meeting

Who This Is For

Three audiences, one shared vocabulary. Clinicians and clinical leads need to understand what a tool is doing to their workflow and what it can quietly get wrong. Health-tech builders need to understand why a model that performs well on a benchmark is nowhere near a deployable product, and what regulators and hospitals will ask of them. Administrators and procurement teams need to evaluate claims without being able to read the underlying statistics themselves. None of these roles requires you to be an AI expert. Several lessons include role-specific views so you can read the same material through the lens that matches your job.

  • Clinicians: what changes in your workflow, and what fails quietly
  • Builders: why benchmark performance is the start of the work, not the end
  • Administrators and advisers: how to interrogate a claim without doing the statistics yourself

Prefer slides, quizzes, and saved progress? Read this lesson in the library — free, no sign-up.