LO
Anaesthesiologist / anaesthetist — tasks, one by one
The unit of analysis is the task, not the job title. Each one below carries its direction, whether the judgement rests on evidence or on platform inference, the reasoning, and what it does not establish.
Every task on this page#
Pre-operative assessment and planning
Being augmented≈ Platform inferenceExamining the patient, reviewing history, medicines and risks, and deciding the anaesthetic plan.
Language models know the textbook but struggle with judgement. GPT-4o answered 83.69% of a national anaesthesiology exam correctly, but did worse on application and analysis questions, and its most common error was an unsupported medical claim. On sample US oral board exams, examiners found no significant difference in overall scores between ChatGPT and anaesthesiology fellows, yet identified the machine's answers in 23 of 24 modules.
Exam studies, not patients; an exam answer is not an assessment of a real patient, and neither study measures work in hospitals.
Delivering and adjusting anaesthesia
Being augmented✓ Evidence-backedGiving and continuously adjusting the drugs that keep a patient asleep, pain-free and still.
Automated dosing works in trials and is kept under a person in practice. A meta-analysis of 17 randomised trials with 1,898 patients found closed-loop systems increased time within the target depth of anaesthesia by 17.6% compared with clinicians adjusting by hand. The US regulator approved one automated sedation machine for healthy adults having colonoscopy or gastroscopy, on condition that a professional trained in anaesthesia is immediately available and one person is dedicated to watching the device and the airway.
Trials of research systems and one restricted approval; no record shows automated anaesthesia in routine use, or how much of the anaesthetist's attention it frees.
Watching the patient and responding to deterioration
Being augmented✓ Evidence-backedReading blood pressure, heart rhythm, oxygen and depth of anaesthesia, and acting before problems become harm.
Warning algorithms add a signal; the response stays with the anaesthetist. The US regulator classified a hypotension prediction algorithm as an adjunctive indicator that is not intended to independently direct therapy. An early single-centre trial found it reduced low blood pressure during surgery, but two 2026 trials found it no better than simpler rules: a blinded trial found a plain alarm at a mean arterial pressure of 72 mmHg non-inferior, and another found it not superior to treating at 73 mmHg or below.
Small trials of one algorithm; they measure blood pressure, not staffing, and do not show whether anaesthetists watch fewer patients.
Airway management and emergencies
Still human-led✓ Evidence-backedPlacing and securing breathing tubes, handling a difficult airway, and leading the response when something goes wrong.
No record found shows a machine doing this. Even the automated sedation approval kept managing the airway with a dedicated person, and the warning algorithm's clearance leaves treatment decisions to the clinician.
The absence of a record is not evidence that nothing is being developed; research on predicting difficult airways was found only in reviews and single studies.
Recovery and pain management
Still human-led≈ Platform inferenceWaking the patient safely, handing over to recovery staff, and planning pain relief after surgery.
No record found here. The closed-loop meta-analysis reports a slightly shorter time to removing the breathing tube, but not a change in who manages recovery.
This judgement rests on the absence of records and on the other tasks, not on evidence about recovery or pain management itself.