LO
Medicine
Trains you to take an undifferentiated complaint from a frightened person, decide what is most likely and what would be most dangerous to miss, and then carry the decision — including when the tools in the room disagree with you.
Pick one imaging or triage tool your teaching hospital has switched on, and find out three things about it: who signed it off, what it was evaluated against, and what happens when a clinician disagrees with it. Write the three answers down. If you cannot get one of them, that absence is the answer.
A single number for a whole major would hide the thing that matters: this degree trains several separate competencies, and they are not all moving in the same direction. Automation acts on tasks, so any assessment lives on the occupation pages below — not here.
Written for undergraduate and graduate-entry medical degrees leading to a licence to practise. The gate everywhere is registration and then a training post, and both are controlled by the state or a professional body rather than by employers — which is why this page's picture moves more slowly than the technology does. Nursing, pharmacy, physiotherapy and laboratory science are separate degrees with separate pages or separate gates, and their exposure is not the same. Health systems differ enormously in who may prescribe, who may report an image and who carries the liability, and that is the biggest thing this page cannot generalise across.
What this degree actually trains#
Not the course list — the competencies underneath it, and whether each one is worth more or less than it was.
Getting a usable account out of a person
HoldingTurning what someone chose to tell you — in the order they chose, with the parts they are ashamed of left out — into something you can reason from. Noticing the pause, the thing mentioned on the way out, the symptom described as normal.
Symptom checkers and intake forms now collect a structured version of this before the consultation, and ambient tools transcribe it during. Neither does the eliciting. What arrives at the door is still an account shaped by what the person was willing to say to a person, and the training for that has not become easier to replace — it has become the part of the visit the tools hand back to you.
Hands on a body
HoldingEliciting signs — palpating, percussing, listening — and knowing which of them changes what you do next.
Point-of-care ultrasound moved part of this from the hand to a probe and a screen, which is a change in method rather than in who does it. No deployment anywhere replaces the examination itself, and in systems where imaging is expensive or distant it is still the thing that decides whether a scan is ordered at all.
Ranking what it could be
HoldingGenerating the possibilities, ordering them by likelihood and by the cost of missing each one, and deciding what to do before you know.
Language models score very well on written diagnostic vignettes, and that is the most-cited evidence in this whole area. A vignette is pre-packaged: someone already decided which facts were relevant and wrote them down. A real patient arrives undifferentiated, and the site holds no record of a deployment where a machine produces the differential in practice. Capability is not deployment, and treating the vignette scores as though they were is the single most common mistake made about this degree.
Reading the scan, the slide, the trace
Worth less on its ownPattern recognition on images and signals — and the volume of normal studies you must read to stay good at spotting the abnormal one.
This is the one part of the degree where deployment is real and documented: hundreds of imaging algorithms are cleared for clinical use, and they are bought, installed and used on live worklists. It is not that the skill stopped mattering — someone still signs. It is that the routine volume that used to build and pay for the skill is where the machine was put, so the skill must now be paid for by something other than throughput.
Being the person who carries it
Worth more than beforeSigning. Being answerable — to the patient, to a coroner, to a regulator — for an action taken on another person's body, including an action a tool recommended.
Every regulator that has written rules here has written a person into them. The EU's high-risk regime, Korea's framework act requiring human management and supervision of high-impact systems, and the several jurisdictions that reserve the signature to a licence holder all point the same way. The more capable the tools became, the more explicitly the law named someone who has to be able to say no.
Explaining what the machine contributed
Worth more than beforeTelling a patient what was decided, what a tool contributed to it, and what you did with that — in a way that leaves them able to disagree.
This is work the tools created rather than work they took. Consent frameworks are beginning to require it, and nobody's training covers it yet: a graduate is taught to explain a diagnosis, not to explain a probability produced by a system whose errors are distributed differently from a human's. It is scarce because it is new.
Notes, letters and coding
Worth less on its ownWriting the encounter down so that the next clinician, the insurer and the auditor can all use it. An hour or more of most clinical days.
Ambient documentation is the fastest-adopted clinical application anywhere, for the same reason it is in nursing: it attacks the task clinicians most resent and does not touch the decision. The time it returns is real. Whether that time stays with patients or is absorbed into a shorter appointment slot is a management decision, not a technical one, and it is being made now.
Where it can lead#
Several directions, never one. Each says what your training reuses, what graduates typically lack, the real entry conditions, and one thing you can test this term.
General and primary care
task-level analysis →- What transfers
- The undifferentiated consultation is what this degree is actually built for: the history, the examination, and the decision about what would be dangerous to miss, all under time pressure and without the tests in hand.
- What graduates typically lack
- Graduates commonly arrive able to reason about a presented case but not to run a list — deciding how long each person gets, when to bring someone back rather than investigate now, and how to hold risk overnight. The tools entering this setting are triage and documentation tools, and nobody is taught when to override one.
- Entry reality
- Registration plus a training post, and the training post is the real bottleneck — it is rationed by the state or a professional body, not by demand. Expect a fixed number of years before you are the one signing.
Sit in on ten consultations and write down, for each, the moment the clinician decided what this was. Then ask them which fact made the difference. Half the time they will name something that was never written in the notes — that is the part no intake form collects.
Imaging and the diagnostic departments
task-level analysis →- What transfers
- Pattern recognition trained on volume, plus the anatomy and pathology the rest of the degree is built on. This is the specialty where the degree's technical core is most directly the job.
- What graduates typically lack
- Graduates entering this field now inherit a worklist where part of the routine volume has already been triaged or pre-read by a device. The scarce skill shifts from reading fast to knowing when the device's output is wrong in a way that matters — and the training still builds the first by doing the volume the device now takes.
- Entry reality
- A competitive specialty training post after registration. This is also the field where the most public predictions have been made about the job disappearing; the site's occupation page holds both those predictions and what has actually been measured, on the same page, deliberately.
Find out which imaging algorithms your teaching hospital has actually bought and switched on — not which it has trialled. Ask a registrar what changes in their day because of them. If the answer is nothing, that is a finding about deployment, and it is worth writing down with the date.
The procedural specialties
- What transfers
- Anatomy, physiology and the decision about when not to operate — the last of which is the part of surgical training that takes longest and transfers least.
- What graduates typically lack
- Graduates underestimate how much of these specialties is hours in a room doing a physical thing, and how little of the published material about AI and medicine is about that. This site does not have a page for surgery, anaesthetics or interventional work, so there is no task-level analysis here to send you to — that is our gap, not a judgement that the direction is safe.
- Entry reality
- The longest training of any direction here, and the one where the place you train decides most about what you can do afterwards.
Scrub in twice and time it: how many minutes of the case were the operation, and how many were positioning, consent, checklists and waiting. The ratio is the honest picture of the job, and it is not the one in the recruitment material.
Leaving the room: being the clinician the tools answer to
- What transfers
- A licence, plus the one thing no vendor and no engineer has: you have carried the decision yourself, so you know which of a tool's errors are survivable and which are not.
- What graduates typically lack
- Graduates have no training in procurement, evaluation design or how to say no to an institution that has already bought something. Deliberately unlinked here: this site's page on the person accountable for AI adoption is written about a company that buys and applies these tools, and a public hospital is not a company — the governance, the liability and the person who can veto are all different. Sending you there would be giving you an analysis of a different kind of organisation.
- Entry reality
- In most systems this is not a post you apply for. It accretes onto someone who is already clinically credible and who put their hand up once. Clinical credibility comes first, and it takes the same years as any other direction here.
Ask your hospital for the evaluation it ran before switching on one clinical tool — any tool. Whether you get a document, a slide deck, or nothing at all tells you more about this direction than any careers talk will.
What to add outside the classroom#
This is about what graduates commonly lack in practice — not a claim that your school failed to teach it.
Learn to read a diagnostic accuracy study properly — sensitivity, specificity, and what happens to both when the prevalence in your clinic is not the prevalence in the paper. Almost every claim you will be sold rests on this, and almost nobody checks it.
Keep a log of the times a tool in your placement was wrong, with what it said and what was true. Two years of that is a thing almost no graduate has, and it is the evidence base for every argument you will later need to make.
Get comfortable with the part of the law that names you — who may sign, who is liable, what consent must contain where you will practise. It is short, it is public, and it is the thing that decides how much of this degree is still yours.
This term#
One or two actions, each producing something you can show someone. Not a reading list.
Pick one imaging or triage tool your teaching hospital has switched on, and find out three things about it: who signed it off, what it was evaluated against, and what happens when a clinician disagrees with it. Write the three answers down. If you cannot get one of them, that absence is the answer.
Take ten consultations you observed and mark, for each, whether the decisive fact came from the person talking, from the body, or from a test. Count the three. That ratio is your own evidence about which part of this degree the machines are actually near.
Common questions#
This site will not answer that with a score, because a single number for a whole degree would hide the thing that decides it: medicine trains several separate abilities and they are not moving together. Read the two occupation pages this degree reaches — general practice and radiology — and compare them task by task. They differ sharply, and which one you are drawn to matters far more to this question than the degree does. If the answer you want is a number, nobody who has it is measuring your situation.
Those tests are written vignettes: someone already decided which facts mattered and wrote them down in order. That is the hard half of the job, done in advance, by a human. The honest way to read those results is that they establish capability, and capability is not deployment. Where deployment genuinely exists in medicine — imaging devices cleared and switched on in hospitals — the site records it as such on the radiology page, and it is a much narrower claim than the headline.
The site refuses to answer that with a year or a probability, and the radiology page says why. What it does give you is checkable: which imaging tasks have documented deployment, what the regulators have written about who signs, and the famous prediction about this field alongside what has actually been measured since. Two things worth knowing before you decide: the prediction most people quote came with a second sentence about there being plenty of radiologists, which almost no retelling carries; and the routine reading volume is both where the machines went and how the specialty used to train and pay for itself.
The useful thing about this degree's length is that the gate is not an employer. Registration and training posts are controlled by the state or a professional body, and those move slowly and deliberately — which is why every regulator that has written rules about clinical AI has written a person into them rather than out. The thing to watch is not whether the job exists but which tasks inside it are still yours, and that is what the occupation pages track. Check them again in your final year; the records carry dates precisely so that you can.
One thing, and it is cheap: keep a record of the times a clinical tool was wrong in front of you — what it said, what was true, what happened next. Two years of that is evidence almost no graduate has, and it is what you will need the first time you have to argue against something an institution has already bought. It is also the only part of this that compounds.
Method#
Assessments live on tasks, not on majors. Follow any direction above to its occupation page to see which tasks are changing, how strong the evidence is, and what it does not yet show.