What AI is doing in medicine right now
AI has made significant clinical advances in specific domains of medicine. In diagnostic imaging, AI systems match or exceed human radiologists on certain tasks: detecting early-stage lung cancer in CT scans, identifying diabetic retinopathy in fundus photographs, and flagging suspicious lesions in mammography. Google DeepMind's AlphaFold has solved protein structure prediction, a problem that took years of human scientific effort. AI drug discovery platforms are shortening the time from target identification to candidate compound selection. AI triage tools in emergency departments assess patient acuity and direct workloads more efficiently than rule-based systems.
In administrative medicine (clinical documentation, coding, pre-authorisation, scheduling), AI is automating functions that previously occupied significant physician time. Ambient clinical intelligence tools (AI that listens to the patient-physician consultation and automatically generates clinical notes) are reducing the documentation burden that contributes significantly to physician burnout.
What AI cannot do in medicine
AI cannot replace the clinical judgment, empathetic communication, and holistic care that define what a doctor does. Diagnosis is not just pattern recognition: it involves taking a history from a patient who may not know what is relevant to mention, integrating information from physical examination, contextualising findings against the patient's full social and psychological picture, and making decisions that account for the patient's preferences and values. These are irreducibly human activities. Beyond diagnosis, the therapeutic relationship (the trust that enables patients to adhere to treatment, disclose sensitive information, and engage with mental health or lifestyle interventions) requires a human physician who the patient believes understands them.
Medicine also involves accountability and moral agency. A physician can be held responsible for a clinical decision in a way that an AI system cannot. Patients and families interacting with healthcare during difficult moments need a human who is answerable for the care delivered. This accountability dimension means the physician role is protected even as AI assists in the technical components of clinical decision-making.
The future of medical careers
The doctor of 2030 will use AI tools routinely: AI-assisted diagnosis, AI-generated draft clinical notes, AI-based decision support for prescribing and referral, and AI monitoring of chronic disease. The time this frees up should be redirected toward the patient relationship, complex decision-making, and the aspects of care that AI cannot provide. The medical roles most at risk are those where the primary value is isolated pattern recognition or information processing, not the full clinical encounter. The roles most protected are those that integrate technical skill with human relationship: GP, psychiatrist, geriatrician, palliative care physician, paediatric complex care.