Physician · Diagnoses illness and manages health for years afterward through examination and evidence, not a single operation — medicine's generalist and long-term guide.
Darker cells mean a higher score for this topic on that metric.
LessMore
Last reviewed Sources & creditsMedia creditsMethodology
A physician diagnoses and manages illness through examination, testing and medication, typically following a patient over months or years. A surgeon treats disease or injury by operating, usually for a single well-defined episode of care. Many physicians — internists, family doctors, cardiologists — never operate; some conditions are handled entirely by a physician, others are referred to a surgeon once medical treatment is exhausted.
Most countries require roughly ten to eleven years after secondary school: four to six years of medical school followed by three or more years of residency in a specialty such as internal medicine, family medicine or pediatrics. The United Kingdom compresses medical school to five or six years but adds a two-year foundation programme before specialty training begins, so the total ends up similar.
In everyday English, "doctor" covers any licensed physician, including surgeons and specialists. "Physician" more precisely means a non-surgical doctor practicing internal medicine or a related specialty. A "general practitioner" or family doctor is a physician trained broadly rather than in one organ system, usually a patient's first point of contact and the one who decides which specialist, if any, a case needs next.
It varies enormously by country and specialty. Primary care physicians in the United States earn roughly $220,000 to $260,000 a year on average; NHS consultant physicians in England sit on a national scale from about £93,666 to £126,281; salaried internal medicine specialists in Germany typically earn €90,000 to €110,000. Physicians in lower-income countries often earn a small fraction of these figures despite comparable training.
Diagnostic pattern-matching is exactly the kind of task large language models are improving at fastest, and controlled studies already show some AI systems matching or beating physicians on written case tests. What survives is harder to automate: physical examination, years of trust with a returning patient, weighing a person's full context, and carrying legal responsibility no algorithm can hold. That narrows the job rather than replacing it.
A physician completes medical school and residency, typically ten or more years of training, and in most countries can independently diagnose, prescribe and manage the full range of adult or pediatric disease. A nurse practitioner completes a nursing degree plus a master's or doctoral program, usually four to six years total, and in many US states can practice independently within a narrower scope, often in primary care.
Diagnosis is largely pattern-matching against a mountain of prior cases and published literature, which is precisely the task large language models are built for. In controlled studies, some AI systems already produce more accurate differential diagnoses than practicing clinicians on text-based cases — a result that would have sounded absurd a decade ago and should be taken seriously rather than dismissed.
This page separates what that actually means from what it doesn't: which specific tasks are already well automated, which have proven far more resistant, and what changes for physicians whose diagnostic reasoning is no longer their unchallenged advantage.
Roughly a third of the day-to-day cognitive work — matching symptoms to conditions, checking drug interactions, drafting notes and after-visit summaries — is already being matched or approximated by software. What resists automation is everything wrapped around that reasoning: the physical exam, a frightened patient's trust, and legal responsibility that only a licensed human can carry.
Scored from the tasks, not the job title. Lower is safer.
Jobs AI cannot take →A physician's license and malpractice liability make them the accountable party when a diagnosis or treatment plan turns out to be wrong; no AI system can hold a medical license or be sued, so a human must sign off regardless of which tool produced the reasoning.
Chronic disease management depends on a relationship built over years — a patient who will admit they've stopped taking a medication, or mention a symptom they'd otherwise dismiss, to a physician they already trust, not a new interface every visit.
Palpating an abdomen, listening for a specific heart murmur, or noticing a subtle change in gait requires hands-on skill and situated judgment that remote software, and even most current robotics, cannot replicate at the bedside.
Real patients arrive with multiple overlapping conditions, incomplete histories and symptoms that don't match any textbook case cleanly; AI systems tested on hard, atypical published cases still miss the correct top diagnosis a large share of the time.
Telling a patient a diagnosis is serious, or persuading someone to accept a treatment they're afraid of, is a negotiation that depends on reading one specific person's fear and values in the room, not on generating the medically correct sentence.
On text-based case vignettes and simulated patient interviews, AI systems have already matched or outperformed the clinicians they were tested against, including a 2024 study in which an AI system outperformed early-career primary care physicians and nurse practitioners on simulated consultations judged by specialists.
Ambient AI scribes that listen to a consultation and draft the clinical note are already in use in many hospital systems, cutting the after-hours 'pajama time' physicians have long spent finishing charts at home.
Software already cross-checks a patient's full medication list against interactions and current treatment guidelines faster and more exhaustively than a physician recalling them from memory or a quick manual lookup.
Algorithms increasingly do the first pass on incoming lab results, imaging reports and remote monitoring data, flagging what's normal versus what needs a physician's attention rather than a human reviewing every value in sequence.
Increasingly a physician's diagnostic reasoning runs alongside a second, software-generated opinion rather than in isolation, shifting the core skill from generating the differential from scratch to critically evaluating and correcting one a model already drafted.
Ambient scribing tools are shifting physicians' relationship with the medical record from typing during or after a visit toward reviewing and correcting an AI-generated draft, freeing more of the actual appointment for looking at the patient instead of a screen.
As remote monitoring, telemedicine and retail clinics absorb more minor acute complaints, physicians' time concentrates increasingly on complex, multi-condition chronic patients who need judgment a simpler protocol can't safely handle.
The long shift from independent practice into hospital and corporate employment — nearly three-quarters of US physicians by 2022 — is changing autonomy and pay structure faster than any AI tool, reshaping how much control a physician has over their own daily schedule.
A physician-technologist hybrid role, overseeing how electronic health records, AI diagnostic tools and clinical decision-support software actually get built and deployed inside a hospital system, increasingly its own board-recognized subspecialty.
Reviews and audits AI diagnostic and triage tools before and after deployment, checking their outputs against real outcomes — a role that barely existed a decade ago and exists specifically because hospitals need a licensed human accountable for software's mistakes.
Practices largely or entirely through video and messaging platforms, a role that expanded enormously during the COVID-19 pandemic and has remained a permanent, separate career track rather than reverting entirely to in-person practice.
Works at the level of a health system or insurer rather than one patient at a time, using data to target prevention and chronic-disease programs at whole populations — a growing complement to, not a replacement for, one-to-one clinical care.
Three reversible lenses: augment the work, replace a slice, or open a niche. Teaching marks — not forecasts.
Keep the role; AI speeds drafts, triage, or research while judgement and accountability stay human.
A narrow task stack may compress first (templates, first drafts, routine scoring) while adjacent craft grows.
Oversight, integration, and domain QA roles can appear where AI output must be trusted in regulated settings.
AI's diagnostic strength is currently narrowest exactly where it matters most: on hard, atypical real cases, one widely cited 2023 study found GPT-4 identified the correct top diagnosis only 39% of the time on deliberately difficult published case challenges, even though it did better when allowed to list several possibilities. That gap between promising controlled results and messy real-world diagnosis is where a physician's judgment still earns its keep.
The bigger near-term threat to the profession isn't automation but supply: the World Health Organization projects a global shortfall of roughly ten million health workers by 2030, concentrated in poorer countries, while the United States alone may be short tens of thousands of physicians by the mid-2030s. Whatever AI ultimately automates, most health systems will spend the next decade with too few physicians rather than too many.
Closest neighbours on the six-score profile — not the same field only.
Designs and fabricates the transistors inside every computer, phone and weapon, using machines precise enough that only a few factories on Earth can run them.
AI-resistant 60 🧠Diagnoses and treats mental illness through medication and therapy, one of medicine's only specialties with legal authority to detain a patient in crisis.
AI-resistant 73 🧫Uses clinical, trial and health-system data to generate reliable evidence for safer care, research and operational decisions.
AI-resistant 68 🦾Designs the machines that sense, decide and act in the physical world, where the hard problem was never intelligence but the world itself.
AI-resistant 65 🔐Protects systems, data and people by finding, preventing and responding to digital attacks.
AI-resistant 63 🌬️Designs, builds and improves wind, solar, storage and grid systems that turn renewable resources into dependable electricity.
AI-resistant 72The physician who operates to cure, from Sushruta's ancient rhinoplasty to today's robotic operating rooms, still alone with the decision at the table.
AI-resistant 92 💉Healthcare's largest profession: from Scutari's lamp to the ICU monitor, the person who is still at the bedside when everyone else has gone home.
AI-resistant 91 💊The medicines expert behind every prescription — from Baghdad's first drug shops and the apothecary's mortar to morphine, artemisinin and the modern dispensary.
AI-resistant 58 🐾Treats patients from goldfish to horses that cannot describe a symptom, and carries the legal power to end their suffering by euthanasia.
AI-resistant 78 🧠Diagnoses and treats mental illness through medication and therapy, one of medicine's only specialties with legal authority to detain a patient in crisis.
AI-resistant 73 🧬Helps people understand inherited conditions, testing choices and results, combining genomic science with non-directive, compassionate counseling.
AI-resistant 72