General Internal Medicine Physicians

29-1216.00
Median wage $256,560/yr67,150 employed (US)Rank #759 of 923 scored · top 82% by substitution

Diagnose and provide nonsurgical treatment for a wide range of diseases and injuries of internal organ systems. Provide care mainly for adults and adolescents, and are based primarily in an outpatient care setting.

Sub-scores

0–100 · band = confidence interval from rater disagreement

Substitution16
Exposure18
Augmentation67

Substitution — the headline: capability discounted by cost, barriers and adoption.

Exposure — technical capability alone, regardless of whether anyone deploys it.

Augmentation — how much AI assists without replacing. High here + moderate substitution = a changing job, not a disappearing one.

Tasks on the substitution scale

19 rated tasks, binned by substitution score.

Position among all scored occupations

Distribution of 923 occupation scores; the marker is this occupation.

Tasks with substitution ≥ 70

0%

Run 1.0.0-draft.1 · computed 2026-08-05 · rater panel: claude-sonnet-5, claude-haiku-4-5-20251001 · intervals span rater disagreement.

Why this score

The five weighted dimensions of the composite, averaged across this occupation's tasks (importance-weighted, panel mean). Exact weights and formulas: /api/v1/methodology.

Task automatabilityw 35%17

panel mean rating 1.7/5 → substitution pressure 17/100

Technical feasibility todayw 20%21

panel mean rating 1.9/5 → substitution pressure 21/100

Cost vs. human wagew 15%16

panel mean rating 1.7/5 → substitution pressure 16/100

Adoption barriersw 20%inverted — strong barriers lower the score7

panel mean rating 4.7/5 (barrier strength) → substitution pressure 7/100

Sector adoption velocityw 10%23

panel mean rating 1.9/5 → substitution pressure 23/100

Task breakdown (19 tasks)

Substitution pressure per task, weighted by O*NET importance in the composite. Expand a task for the full rater audit trail — every rating, every model, every rationale.

Prepare government or organizational reports on birth, death, and disease statistics, workforce evaluations, or the medical status of individuals.

43

CI 25–60 · exposure 45 · augmentation 75 · importance 3.1/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Healthcare remains relatively slow in adopting autonomous AI; most use is assistive. Government health agencies and medical organizations are cautious adopters of AI-generated reports due to accuracy and accountability concerns, though data-processing tools are gradually deployed.
Sector adoption velocityclaude-sonnet-52/5Healthcare administrative reporting lags behind sectors like finance or tech in AI adoption due to compliance, data privacy, and interoperability challenges, with pilots more common than production-scale deployment.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can meaningfully assist by automating data extraction, statistical summaries, and report formatting, allowing physicians to focus on interpretation and clinical judgment. Current NLP and analytics tools provide useful productivity gains in the information-gathering phase of report preparation.
Augmentation potentialclaude-sonnet-55/5AI can significantly speed up drafting, summarizing statistics, and structuring reports while the physician retains responsibility for accuracy and final approval, making this a strong augmentation use case.
Task automatabilityclaude-haiku-4-5-202510012/5Report generation itself can be partially automated (data aggregation, formatting, basic statistics), but synthesizing medical status assessments, interpreting complex epidemiological patterns, and ensuring clinical accuracy requires physician judgment. Current AI can draft templates but cannot reliably produce the full clinical interpretation and sign-off physicians must provide.
Task automatabilityclaude-sonnet-54/5Drafting statistical reports and summaries from structured data (birth/death/disease records, workforce evaluations) is largely a data aggregation and narrative-writing task that current LLMs and analytics tools handle well, though final data validation and sign-off require human review., saving significant time.
Adoption barriersclaude-haiku-4-5-202510014/5High barriers: government health statistics reports typically require physician or epidemiologist sign-off for accuracy and legal liability; regulatory bodies (CDC, state health departments) often mandate credentialed professional authorship. Medical status assessments carry liability weight that restricts full substitution.
Adoption barriersclaude-sonnet-53/5These reports often require physician certification, accuracy for legal/public health records, and organizational accountability, creating moderate barriers even though the drafting itself isn't inherently restricted to licensed professionals.
Cost vs. human wageclaude-haiku-4-5-202510012/5The task requires physician-level expertise and legal sign-off, so even with AI assistance, the human cost remains the dominant factor. AI can reduce time by handling clerical and data formatting work, but not enough to drop total cost below human-equivalent rates.
Cost vs. human wageclaude-sonnet-54/5AI-assisted drafting and data compilation is far cheaper than physician time spent on non-clinical reporting tasks, though some oversight and data verification costs remain.
Technical feasibility todayclaude-haiku-4-5-202510012/5While AI tools can extract data and format reports, no deployed system reliably handles the full task of preparing authoritative government/organizational health statistics with the clinical rigor and liability standards required. Clinical documentation AI exists but scope is narrower and error rates remain material.
Technical feasibility todayclaude-sonnet-53/5Products exist for automated report generation and medical documentation (e.g., EHR analytics, AI scribes, statistical reporting tools) but full end-to-end preparation of official government/organizational reports with accuracy guarantees is not yet a mature, widely deployed turnkey product in this exact form.

Collect, record, and maintain patient information, such as medical history, reports, or examination results.

41

CI 32–50 · exposure 42 · augmentation 88 · importance 4.2/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510013/5AI-assisted documentation and EHR enhancements are spreading in hospital and large group settings, but adoption remains fragmented; many practices use basic tools, and real end-to-end automation without physician review is rare in production.
Sector adoption velocityclaude-sonnet-53/5Healthcare has historically lagged in digitization but ambient documentation AI has seen fast recent uptake in health systems, placing it in a middling-to-accelerating adoption phase.
Augmentation potentialclaude-haiku-4-5-202510014/5AI transcription, template population, and automated extraction of key data from prior records meaningfully reduce documentation burden and improve completeness, allowing physicians to focus on analysis and decision-making while the system handles routine recording tasks.
Augmentation potentialclaude-sonnet-55/5AI scribes and EHR summarization tools substantially reduce documentation time and cognitive load while the physician remains the responsible party for accuracy, a clear high-augmentation use case.
Task automatabilityclaude-haiku-4-5-202510012/5While AI can assist with data entry and some structured recording via transcription or templates, full end-to-end collection and maintenance of medical history requires clinician judgment about what is relevant, patient interaction to elicit information, and integration of unstructured clinical findings—tasks that remain human-centric.
Task automatabilityclaude-sonnet-53/5AI ambient scribes and dictation tools can transcribe and structure much of the documentation, but clinicians still must verify accuracy, gather history via judgment-based questioning, and finalize records, so full end-to-end automation isn't yet met.
Adoption barriersclaude-haiku-4-5-202510014/5Medical documentation carries significant liability exposure; regulatory requirements (HIPAA, CMS coding standards) mandate human accountability; legal standards expect physician attestation of medical records, creating strong organizational and legal friction against full automation.
Adoption barriersclaude-sonnet-53/5Medical records require physician attestation and accuracy for legal and billing purposes, creating oversight requirements, though no law mandates a human personally type/record data.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI transcription and data capture tools reduce clerical burden, but the infrastructure, integration, compliance, and required physician oversight to ensure accuracy and liability coverage make the all-in cost roughly comparable to traditional human documentation.
Cost vs. human wageclaude-sonnet-53/5Scribe subscriptions cost hundreds of dollars per physician monthly, which is cheaper than a human scribe but still requires physician oversight time, making savings moderate rather than order-of-magnitude.
Technical feasibility todayclaude-haiku-4-5-202510013/5EHR systems with AI-powered transcription and structured data capture are widely deployed in clinical settings, but they require substantial human oversight for accuracy, completeness, and clinical appropriateness; error rates in automated coding and data integrity remain material.
Technical feasibility todayclaude-sonnet-53/5Ambient AI scribe products (e.g., Nuance DAX, Abridge, Nabla) are deployed in real clinics and reduce documentation burden, but they still require physician review and have error rates that limit full autonomy.

Advise patients and community members concerning diet, activity, hygiene, and disease prevention.

27

CI 25–29 · exposure 25 · augmentation 75 · importance 4.4/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5While health systems are piloting AI documentation and decision-support tools, production deployment of fully autonomous patient counseling remains rare; adoption is slow due to regulatory uncertainty, liability concerns, and physician resistance to delegating the counseling relationship.
Sector adoption velocityclaude-sonnet-52/5Healthcare overall lags in AI deployment for direct patient-facing clinical advice due to regulatory caution, though administrative and documentation uses are growing faster.
Augmentation potentialclaude-haiku-4-5-202510014/5AI can significantly assist physicians by drafting personalized patient education materials, summarizing evidence-based guidelines for diet and activity, and flagging preventive care gaps—substantially raising physician productivity in preparing counseling content while the physician maintains clinical judgment and the patient relationship.
Augmentation potentialclaude-sonnet-54/5AI tools can draft patient education materials, summarize guidelines, and prep talking points, meaningfully speeding up physician counseling while the physician retains responsibility for delivery and judgment.
Task automatabilityclaude-haiku-4-5-202510012/5While AI can generate generic health advice on diet, activity, and hygiene using text generation, this task requires tailoring recommendations to individual patient contexts (comorbidities, medications, socioeconomic constraints, preferences), ongoing rapport-building, and addressing patient concerns—capabilities current systems cannot reliably perform end-to-end at the quality threshold.
Task automatabilityclaude-sonnet-52/5AI chatbots can generate generic diet/hygiene/prevention advice, but personalized clinical counseling requires integrating patient history, exam findings, and nuanced risk factors that current systems cannot reliably do end-to-end without physician oversight.'},
Adoption barriersclaude-haiku-4-5-202510014/5Strong barriers exist: physicians are the licensed professionals responsible for medical advice in most jurisdictions; patients often prefer human counseling for sensitive health topics; liability and malpractice exposure create institutional friction against full automation of medical guidance; regulatory bodies (FDA, state boards) constrain autonomous clinical advice.
Adoption barriersclaude-sonnet-54/5Medical advice giving is subject to licensure, malpractice liability, and scope-of-practice rules that generally require a licensed physician or supervised professional to deliver clinical counseling.
Cost vs. human wageclaude-haiku-4-5-202510012/5Current AI advisory systems (chatbots, text-generation APIs) are cheap per inference, but integration into clinical workflows, clinician review/oversight, and the need for human validation to ensure safety and legal compliance add significant overhead, bringing total cost closer to or exceeding the value of delegating to lower-cost clinical staff.
Cost vs. human wageclaude-sonnet-53/5AI-generated generic advice is very cheap, but when integrated with EHR review, oversight, and liability management the effective cost approaches or exceeds a portion of physician time for this task.
Technical feasibility todayclaude-haiku-4-5-202510012/5Chatbots and clinical decision-support tools can draft educational content, but deployed systems do not reliably deliver personalized disease-prevention counseling in production medical settings; regulatory and liability frameworks restrict fully autonomous AI advice-giving in clinical contexts.
Technical feasibility todayclaude-sonnet-52/5Consumer health apps and chatbots offer general wellness advice, but no deployed product independently delivers physician-level preventive counseling at scale in clinical settings.

Refer patient to medical specialist or other practitioner when necessary.

23

CI 20–25 · exposure 25 · augmentation 75 · importance 4.3/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5While some EHR systems incorporate referral flags, autonomous AI-driven referrals remain rare in production; most healthcare organizations retain physician discretion due to liability and quality concerns.
Sector adoption velocityclaude-sonnet-52/5Healthcare overall has been slower than other professional sectors to adopt AI for clinical decision-making tasks, especially those with direct liability implications like referrals.
Augmentation potentialclaude-haiku-4-5-202510014/5AI can substantially augment physicians by highlighting guideline-concordant referral opportunities, summarizing relevant clinical literature, and flagging high-risk presentations—raising the quality and speed of referral decisions without replacing physician judgment.
Augmentation potentialclaude-sonnet-54/5AI tools can flag abnormal results, suggest relevant specialists, and draft referral documentation, meaningfully speeding up the physician's workflow while the physician retains decision authority.
Task automatabilityclaude-haiku-4-5-202510012/5AI can assist in identifying when a referral is clinically indicated based on guidelines and symptom patterns, but the decision requires nuanced clinical judgment, patient context, and knowledge of local specialist availability—factors that resist full automation today.
Task automatabilityclaude-sonnet-52/5Deciding when and to whom to refer requires clinical judgment, synthesis of patient history, and accountability that current AI cannot fully replicate end-to-end, though AI can help identify candidates for referral or draft referral letters.'
Adoption barriersclaude-haiku-4-5-202510014/5The physician bears legal and professional liability for referral decisions; medical boards and malpractice frameworks require a licensed physician to exercise clinical judgment and take responsibility for appropriateness of care routing.
Adoption barriersclaude-sonnet-55/5Referral is a licensed medical act tied to legal and professional accountability; only a physician or authorized practitioner can make and sign off on this determination.
Cost vs. human wageclaude-haiku-4-5-202510012/5Implementing a reliable AI referral system requires significant infrastructure, clinical validation, and ongoing human review, making it costlier than a physician's intuitive clinical judgment in most current deployments.
Cost vs. human wageclaude-sonnet-52/5Physician judgment and liability for referral decisions still require a licensed human, so AI only reduces administrative overhead rather than replacing the decision cost.
Technical feasibility todayclaude-haiku-4-5-202510012/5Decision-support tools exist that flag potential referral triggers, but no deployed system reliably makes autonomous referral decisions end-to-end; physician oversight remains mandatory and substantial across deployed products.
Technical feasibility todayclaude-sonnet-52/5Clinical decision support tools can flag potential referral needs, but no deployed product autonomously makes referral decisions at scale in production without physician oversight.

Make diagnoses when different illnesses occur together or in situations where the diagnosis may be obscure.

20

CI 20–20 · exposure 25 · augmentation 63 · importance 4.3/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Adoption of autonomous diagnostic AI in clinical practice remains limited despite decades of research; most deployments are narrow (e.g., imaging-specific) or supportive rather than substitutive, reflecting organizational conservatism, liability concerns, and persistent performance gaps in real-world settings.
Sector adoption velocityclaude-sonnet-52/5Healthcare is a historically slow-adopting, highly regulated sector; AI diagnostic aids are in pilot/early deployment stages rather than widespread production use for complex diagnosis.
Augmentation potentialclaude-haiku-4-5-202510013/5AI diagnostic tools meaningfully assist physicians by surfacing differential hypotheses, flagging rare conditions, and organizing clinical information, thereby improving diagnostic accuracy and reducing cognitive burden in complex cases while the physician retains interpretive and final decisional authority.
Augmentation potentialclaude-sonnet-54/5AI differential-diagnosis tools and literature search meaningfully help physicians consider obscure or overlapping conditions, improving diagnostic breadth while the physician retains final judgment.
Task automatabilityclaude-haiku-4-5-202510012/5AI systems can assist in differential diagnosis and pattern recognition from patient data, but making diagnoses in complex comorbid or obscure cases requires integrative clinical judgment, synthesis of ambiguous evidence, and real-time decision-making with patient interaction that current systems cannot reliably perform end-to-end without significant physician oversight.
Task automatabilityclaude-sonnet-52/5AI can generate differential diagnoses and flag rare conditions from clinical data, but complex multi-morbidity diagnosis requiring physical exam, patient history nuance, and accountability cannot be done end-to-end without a physician today.stroke
Adoption barriersclaude-haiku-4-5-202510015/5Strong legal and regulatory barriers exist: physicians must legally perform diagnosis and take responsibility for diagnostic decisions; medical boards, liability frameworks, and standard of care all mandate human physician sign-off on diagnoses, particularly in complex or obscure presentations.
Adoption barriersclaude-sonnet-55/5Medical diagnosis is legally restricted to licensed physicians who bear liability for diagnostic decisions, especially in complex/obscure cases where error costs are high.
Cost vs. human wageclaude-haiku-4-5-202510012/5Current diagnostic AI tools require substantial integration, validation, and clinical oversight; when accounting for the cost of ensuring safety and appropriateness in complex cases, the all-in cost approaches or exceeds the loaded cost of physician time.
Cost vs. human wageclaude-sonnet-52/5AI query costs are cheap, but integration, physician oversight, liability review, and validation of diagnoses for complex cases keep total cost comparable to or only modestly below physician time.
Technical feasibility todayclaude-haiku-4-5-202510012/5While diagnostic decision-support tools exist (e.g., Isabel, DXplain), they operate as aids rather than autonomous systems; no deployed product reliably makes independent diagnoses in complex multimorbid or obscure presentations at the quality standard required for clinical practice.
Technical feasibility todayclaude-sonnet-52/5Diagnostic decision-support tools (e.g., differential-diagnosis generators) are deployed in some settings but are used as adjuncts with high error rates on complex cases, not as reliable standalone diagnosticians.

Advise surgeon of a patient's risk status and recommend appropriate intervention to minimize risk.

20

CI 20–20 · exposure 25 · augmentation 75 · importance 4.1/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Healthcare remains a laggard sector in AI adoption relative to finance or software; while risk-assessment tools are piloted in some health systems, end-to-end replacement of specialist consultation is rare and slow due to liability, regulatory, and organizational friction.
Sector adoption velocityclaude-sonnet-52/5Healthcare overall is a slower-adopting sector for autonomous clinical decision-making due to regulatory, liability, and safety concerns, though AI-assisted risk tools are increasingly piloted in perioperative care.
Augmentation potentialclaude-haiku-4-5-202510014/5AI can meaningfully augment this task by rapidly synthesizing patient lab results, comorbidities, and published risk scores, helping internists prepare more thorough consultations and catch overlooked factors—substantially raising the quality and speed of the human advisory process while the physician remains accountable.
Augmentation potentialclaude-sonnet-54/5AI-powered risk calculators, EHR-integrated predictive models, and literature synthesis tools meaningfully speed up and improve the physician's ability to assess and communicate risk to surgeons.
Task automatabilityclaude-haiku-4-5-202510012/5While AI can analyze patient data and flag risk factors, the task requires synthesizing complex clinical judgment, patient-specific context, and real-time communication with another specialist—activities current AI systems cannot reliably perform end-to-end without substantial human oversight and final decision-making.
Task automatabilityclaude-sonnet-52/5AI can help synthesize risk factors and flag known risk scores, but the actual judgment, communication, and accountability for surgical risk advice requires physician synthesis of nuanced clinical context that current systems cannot fully replace end-to-end.rating reflects partial support only.
Adoption barriersclaude-haiku-4-5-202510015/5Strong legal and regulatory barriers exist: the physician of record bears liability for clinical recommendations, and medical licensing law requires a licensed physician to exercise independent judgment and sign off on patient care decisions and surgeon consultations.
Adoption barriersclaude-sonnet-55/5This is a licensed medical judgment task involving direct clinical recommendations to another physician; only a licensed physician can legally provide such consultative advice and bear liability.
Cost vs. human wageclaude-haiku-4-5-202510012/5The loaded cost of a general internist's time for this consulting task is significant, and current AI tools still require human clinicians to validate, interpret, and take responsibility for recommendations, making full cost displacement unrealistic today.
Cost vs. human wageclaude-sonnet-52/5While AI-assisted risk stratification tools are cheap to run, the human physician oversight, liability, and integration costs keep overall cost comparable to or only modestly below physician time costs.
Technical feasibility todayclaude-haiku-4-5-202510012/5Clinical decision support tools exist and can highlight risk factors, but no deployed AI system reliably performs the full task of advising surgeons with the nuance, accountability, and contextual reasoning this requires; products remain narrow and require human validation.
Technical feasibility todayclaude-sonnet-52/5Clinical decision support tools and risk calculators exist and are used, but no deployed product independently advises surgeons on patient-specific risk and intervention with physician-level reliability in production.

Analyze records, reports, test results, or examination information to diagnose medical condition of patient.

18

CI 16–20 · exposure 25 · augmentation 75 · importance 4.5/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Despite high digitization in healthcare, diagnostic automation adoption remains limited to narrow use cases (radiology screening, lab interpretation); full diagnostic replacement faces entrenched physician gatekeeping, regulatory resistance, liability concerns, and slow organizational change in clinical workflows.
Sector adoption velocityclaude-sonnet-52/5Healthcare is a historically slow-adopting sector for autonomous clinical decision-making due to regulation, liability, and integration complexity, though AI-assisted tools are gradually piloted.
Augmentation potentialclaude-haiku-4-5-202510014/5AI systems demonstrably assist physicians by rapidly summarizing records, flagging abnormal results, suggesting differential diagnoses, and synthesizing evidence; these tools meaningfully raise diagnostic efficiency and breadth of consideration while the physician retains interpretive authority and accountability.
Augmentation potentialclaude-sonnet-54/5AI significantly aids physicians by summarizing records, suggesting differentials, and flagging abnormal results, meaningfully speeding up the diagnostic reasoning process while the physician retains final judgment.
Task automatabilityclaude-haiku-4-5-202510012/5While AI systems can assist with pattern recognition in test results and evidence summarization, diagnosis requires integrating complex patient history, physical examination findings, and contextual judgment that current AI cannot reliably perform end-to-end. The liability and consequence of diagnostic error mean human physicians must remain the primary decision-maker, preventing the ≥50% time-saving threshold from being met.
Task automatabilityclaude-sonnet-52/5AI can assist with differential diagnosis and flag patterns in labs/imaging, but integrating full patient context, history nuance, and clinical judgment for final diagnosis remains beyond reliable end-to-end automation today.》Time savings exist for support, not full task replacement.
Adoption barriersclaude-haiku-4-5-202510015/5Legal and regulatory frameworks in most jurisdictions require a licensed physician to be responsible for diagnosis; medical liability and standard-of-care doctrine create hard barriers, and diagnostic error carries severe patient safety and legal consequences that prevent autonomous AI substitution.
Adoption barriersclaude-sonnet-55/5Diagnosis is a licensed medical act with strict legal, regulatory, and liability requirements mandating physician sign-off, making substitution essentially prohibited.
Cost vs. human wageclaude-haiku-4-5-202510011/5AI diagnostic systems require significant infrastructure, clinical validation, integration with EHRs, and continuous human physician oversight; the total cost approaches or exceeds the loaded wage of a physician performing the task, especially when accounting for malpractice liability and the need for human sign-off.
Cost vs. human wageclaude-sonnet-52/5AI diagnostic tools require significant integration, validation, and physician oversight, so all-in costs are not dramatically lower than physician time for this cognitively complex task.
Technical feasibility todayclaude-haiku-4-5-202510012/5AI diagnostic support tools (e.g., image analysis, risk prediction) exist in production but typically serve as decision aids rather than autonomous diagnostic systems, with material limitations in handling rare presentations or multimorbidity. No deployed product reliably performs complete diagnosis end-to-end without physician oversight and final judgment.
Technical feasibility todayclaude-sonnet-52/5Diagnostic decision-support tools (e.g., imaging AI, symptom checkers) exist but are narrow in scope and not deployed as autonomous diagnosticians in mainstream primary care practice.

Explain procedures and discuss test results or prescribed treatments with patients.

17

CI 11–23 · exposure 17 · augmentation 63 · importance 4.4/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Adoption of AI for patient-facing explanation remains nascent; most physicians retain this task for themselves due to professional norms, patient preference, liability concerns, and lack of validated, trusted systems in production across healthcare.
Sector adoption velocityclaude-sonnet-53/5Healthcare is adopting AI documentation and patient communication tools moderately quickly (e.g., ambient scribes, portal messaging drafts), but adoption for actual patient-facing clinical explanation remains cautious and physician-mediated.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can assist by drafting explanation templates, pre-generating summaries of test results, or suggesting talking points, thereby reducing preparation time and improving consistency—but the physician typically delivers and customizes the explanation in real time.
Augmentation potentialclaude-sonnet-54/5AI tools (ambient scribes, draft after-visit summaries, patient education generators) meaningfully help physicians prepare and communicate explanations faster while the physician remains the one delivering and validating the discussion.
Task automatabilityclaude-haiku-4-5-202510011/5Current AI systems cannot autonomously explain complex medical procedures or discuss treatment results with patients in a way that meets clinical and legal standards. This requires real-time listening, dynamic adjustment to patient comprehension, emotional intelligence, and accountability—all beyond current deployed capabilities.
Task automatabilityclaude-sonnet-52/5AI can draft explanations of procedures and results, but the live, interactive discussion with a patient—answering unpredictable questions, reading emotional cues, and building trust—still requires substantial physician involvement, so full end-to-end automation is not yet achievable at equal quality.
Adoption barriersclaude-haiku-4-5-202510015/5Legal and professional standards typically require a licensed physician to explain procedures and discuss results directly with the patient; liability, informed consent, and regulatory expectations (e.g., medical board regulations) create hard barriers to full substitution.
Adoption barriersclaude-sonnet-55/5Explaining diagnoses, test results, and treatment plans is a core licensed medical activity with direct liability and informed-consent implications, requiring physician involvement by law and professional standards.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI-generated explanations require significant physician oversight, validation, and liability management, meaning the all-in cost (including human review and error correction) remains close to or exceeds the cost of the physician doing it directly.
Cost vs. human wageclaude-sonnet-52/5AI can cheaply generate explanatory text, but the overall task still requires a licensed physician's time for accuracy, liability, and interactive Q&A, keeping total cost comparable to or only modestly below physician-only delivery.
Technical feasibility todayclaude-haiku-4-5-202510012/5While AI can generate explanatory text or summaries, no mature production system reliably handles the full interpersonal, contextual complexity of explaining procedures or discussing results with actual patients. Chatbot pilots exist but lack clinical validation and physician-patient trust; they are not standard of care.
Technical feasibility todayclaude-sonnet-52/5Patient portals and chatbots increasingly summarize test results or generate patient-friendly explanations, but no deployed product independently conducts full clinical result/treatment discussions with patients at scale without physician oversight.

Conduct research to develop or test medications, treatments, or procedures to prevent or control disease or injury.

16

CI 5–28 · exposure 17 · augmentation 63 · importance 3.7/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5This task occurs in highly regulated research institutions and clinical settings with strong institutional friction against AI-driven autonomy in research design. Adoption of AI in this domain remains limited to analytical tools supporting (not replacing) physician judgment.
Sector adoption velocityclaude-sonnet-53/5AI adoption in biomedical research (drug discovery, literature mining, trial design assistance) is accelerating in pharma and academic settings, but full-scale autonomous research conduct remains at pilot stage rather than routine production use.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can meaningfully assist with literature searches, data analysis, statistical modeling, and pattern detection in large datasets, which does raise productivity for the human researcher. However, assistance is confined to support roles, not transformative end-to-end performance.
Augmentation potentialclaude-sonnet-54/5AI significantly augments researchers by accelerating literature reviews, generating hypotheses, analyzing large datasets, and assisting with protocol drafting, meaningfully increasing research productivity while humans retain control of the scientific process.
Task automatabilityclaude-haiku-4-5-202510011/5This task involves complex hypothesis generation, experimental design, protocol development, and interpretation of results in a tightly regulated environment. Current AI cannot autonomously design novel clinical trials, manage ethical oversight, or synthesize evidence into actionable treatment protocols at the level a physician researcher performs.
Task automatabilityclaude-sonnet-52/5AI can assist with literature review, hypothesis generation, data analysis, and drafting protocols, but the actual research process—designing trials, recruiting patients, conducting clinical testing, and interpreting nuanced results—requires human scientific judgment, ethical oversight, and physical/clinical execution that current AI cannot perform end-to-end.
Adoption barriersclaude-haiku-4-5-202510015/5Regulatory agencies (FDA, IRB, EMA) legally require licensed physicians to design, oversee, and sign off on research protocols. Liability for adverse outcomes, ethical oversight, and informed consent mandates human professional accountability that AI cannot assume.
Adoption barriersclaude-sonnet-54/5Medical research involving human subjects requires IRB approval, physician oversight, regulatory compliance (FDA, etc.), and legal accountability that cannot be delegated to AI systems, creating strong institutional and legal barriers.
Cost vs. human wageclaude-haiku-4-5-202510011/5Physician researchers command high salaries, and the cost of clinical trial infrastructure, regulatory compliance, and specialized research facilities far exceeds any current AI system's operational cost. AI deployment cannot yet replace the human researcher's authority and accountability.
Cost vs. human wageclaude-sonnet-52/5AI can reduce costs for specific sub-tasks like literature review or data analysis, but the overall cost of conducting medical research (trials, regulatory compliance, patient interaction) remains dominated by human labor and infrastructure, keeping AI cost savings modest relative to the full task.
Technical feasibility todayclaude-haiku-4-5-202510012/5While AI tools can assist with literature review, data analysis, and statistical work, no deployed AI system performs the core creative and regulatory work of medication/treatment research end-to-end. Existing tools require substantial physician oversight and decision-making.
Technical feasibility todayclaude-sonnet-52/5AI tools (e.g., for literature synthesis, molecule screening, or statistical analysis) are used in research pipelines, but no deployed product autonomously conducts full medical research studies reliably at scale; human researchers remain central to design and execution.

Plan, implement, or administer health programs in hospitals, businesses, or communities for prevention and treatment of injuries or illnesses.

16

CI 7–25 · exposure 13 · augmentation 63 · importance 3.2/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Healthcare adoption of AI for administrative and program-planning tasks is slower than in other sectors. Regulatory caution, liability concerns, and the entrenched role of physician judgment in organizational decision-making limit rapid displacement of this task.
Sector adoption velocityclaude-sonnet-52/5Healthcare administration adoption of AI is slower than in finance or pure information sectors, with pilots for population health analytics but limited deployment in actual program administration and planning roles.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can meaningfully assist physicians by analyzing population health data, summarizing evidence, modeling program outcomes, and flagging regulatory requirements. However, the core creative and strategic work of designing and administering programs remains largely physician-driven, limiting transformative augmentation.
Augmentation potentialclaude-sonnet-54/5AI tools can meaningfully assist by analyzing population health data, drafting program proposals, tracking outcomes, and summarizing literature, enhancing physician productivity while the physician retains decision-making authority.
Task automatabilityclaude-haiku-4-5-202510012/5While AI can assist in data analysis, evidence synthesis, and draft program frameworks, the task requires substantial human judgment about organizational context, stakeholder needs, regulatory compliance, and implementation logistics that current systems cannot reliably handle end-to-end. Implementation and administration involve complex coordination and accountability that AI cannot independently discharge.
Task automatabilityclaude-sonnet-51/5This task involves strategic planning, stakeholder coordination, resource allocation, and organizational leadership that require human judgment, negotiation, and accountability far beyond current AI capability to execute end-to-end.
Adoption barriersclaude-haiku-4-5-202510014/5Health program planning and implementation in hospitals and communities are heavily regulated and often require physician credentialing, liability assumption, and sign-off. Legal, regulatory, and malpractice considerations create strong barriers to full automation; human physicians must retain responsibility.
Adoption barriersclaude-sonnet-54/5Program administration in healthcare settings typically requires licensed physician oversight, institutional accountability, and regulatory compliance, creating strong barriers against full automation or substitution.
Cost vs. human wageclaude-haiku-4-5-202510012/5AI tools for supporting program planning (data analysis, research synthesis) are modestly cost-effective, but the core task—designing and administering a program—still requires highly paid physician expertise. The cost of AI assistance plus physician oversight remains comparable to physician-only work.
Cost vs. human wageclaude-sonnet-51/5There is no AI substitute performing this whole task, so cost comparison favors humans entirely; any AI use is a minor input cost addition rather than a replacement of the physician's labor.
Technical feasibility todayclaude-haiku-4-5-202510012/5No deployed AI products today can independently plan, implement, or administer health programs at scale. AI tools exist for clinical decision support and literature review, but production systems do not autonomously design or execute organizational health programs; physicians still make all strategic and implementation decisions.
Technical feasibility todayclaude-sonnet-51/5No deployed AI product plans and administers health programs at institutional or community scale; this remains a human administrative and clinical leadership function with AI only used for isolated sub-tasks like data analysis.

Treat internal disorders, such as hypertension, heart disease, diabetes, or problems of the lung, brain, kidney, or gastrointestinal tract.

14

CI 7–20 · exposure 17 · augmentation 75 · importance 4.5/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Despite significant investment in medical AI, adoption in clinical settings remains slow and limited to narrow use cases (radiology reads, administrative coding). Physicians have been slow to adopt autonomous decision-making systems due to liability, regulatory uncertainty, and ingrained workflows, with most AI tools serving as assistive rather than autonomous agents.
Sector adoption velocityclaude-sonnet-52/5Healthcare is a heavily regulated, slower-adopting sector; AI is used for documentation and decision support but production-level autonomous treatment is rare.
Augmentation potentialclaude-haiku-4-5-202510014/5AI demonstrates strong augmentation potential for physicians: diagnostic support (symptom checkers, differential generators), medical literature synthesis, imaging interpretation, and guideline reminders all materially increase diagnostic accuracy and efficiency. Physicians remain in the loop while AI-assisted tools help them work faster and more comprehensively.
Augmentation potentialclaude-sonnet-54/5AI substantially assists physicians via diagnostic suggestions, treatment guideline retrieval, drug interaction checks, and documentation, improving efficiency while the physician remains responsible for care.
Task automatabilityclaude-haiku-4-5-202510011/5This task requires direct patient examination, differential diagnosis reasoning under uncertainty, personalized treatment planning, and ongoing clinical judgment that current AI cannot perform end-to-end. While AI can assist with specific sub-components (e.g., image interpretation), the integrated clinical work of diagnosis and treatment of complex internal disorders demands human physician oversight and cannot meet the 50% time-saving threshold autonomously.
Task automatabilityclaude-sonnet-52/5Diagnosis and treatment of internal disorders requires physical exams, longitudinal judgment, procedural interventions, and accountability for complex, high-stakes decisions that current AI cannot perform end-to-end.'
Adoption barriersclaude-haiku-4-5-202510015/5Medical practice is heavily regulated: only licensed physicians can legally diagnose and treat internal disorders in most jurisdictions. Malpractice liability, standard-of-care requirements, patient consent obligations, and regulatory bodies (FDA, state medical boards) create hard legal barriers to AI autonomous performance of this task.
Adoption barriersclaude-sonnet-55/5Practicing medicine, diagnosing, and prescribing treatment legally requires a licensed physician; strong regulatory and liability barriers prevent AI from independently performing this task.
Cost vs. human wageclaude-haiku-4-5-202510011/5The all-in cost of AI systems (including integration, validation, liability coverage, and mandatory physician oversight) remains substantially higher than the marginal value delivered when compared to physician labor, which is the necessary baseline for patient safety and legal accountability.
Cost vs. human wageclaude-sonnet-52/5AI diagnostic aids are cheap per query, but full treatment involves exams, prescribing, monitoring, and liability coverage requiring a licensed physician, keeping overall cost comparable to or cheaper than only fractionally, not overall.
Technical feasibility todayclaude-haiku-4-5-202510012/5No deployed AI product can reliably diagnose and treat internal disorders independently. Clinical decision support systems exist (e.g., differential diagnosis assistants), but these are narrow aids requiring physician validation; they do not perform the task end-to-end and have material error rates when deployed in real clinical settings.
Technical feasibility todayclaude-sonnet-52/5Clinical decision-support and diagnostic AI tools exist and are used in some settings, but no deployed product independently treats patients with hypertension, heart disease, or similar conditions without physician oversight.

Manage and treat common health problems, such as infections, influenza or pneumonia, as well as serious, chronic, and complex illnesses, in adolescents, adults, and the elderly.

14

CI 7–20 · exposure 17 · augmentation 63 · importance 4.5/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Healthcare adoption of autonomous clinical AI remains slow due to regulatory burden, liability concerns, and physician skepticism; most AI in clinical settings is narrowly scoped assistance (radiology reads, lab interpretation hints) rather than autonomous management of complex illness.
Sector adoption velocityclaude-sonnet-52/5Healthcare overall lags behind finance/tech in AI adoption for core clinical decision-making, though administrative and support functions are adopting faster.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can assist internists with literature search, differential diagnosis suggestions, drug interaction checking, and data synthesis, improving efficiency on specific subtasks. However, the core acts of clinical judgment, patient communication, and treatment adjustments remain human-driven, limiting transformative productivity gains.
Augmentation potentialclaude-sonnet-54/5AI meaningfully assists physicians via clinical decision support, literature synthesis, documentation (ambient scribes), and differential diagnosis suggestions, improving efficiency while the physician remains in control.
Task automatabilityclaude-haiku-4-5-202510011/5Diagnosing and treating complex, serious chronic illnesses requires integration of patient history, physical examination, clinical judgment, and individualized decision-making that current AI systems cannot perform end-to-end. While AI can assist with differential diagnosis or literature review, the physician must ultimately assess, decide treatment, and manage patient safety and comorbidities.
Task automatabilityclaude-sonnet-52/5Diagnosis and management of complex, multi-system illness requires physical exam, longitudinal judgment, and accountability that current AI cannot autonomously perform end-to-end; AI can support parts (documentation, differential generation) but not replace the core clinical management task.
Adoption barriersclaude-haiku-4-5-202510015/5Medical licensure, malpractice liability, and regulatory requirements (FDA, state medical boards) legally mandate that a licensed physician diagnose and prescribe treatment. Patients also expect and legally require a physician's direct care and accountability, creating hard legal and professional barriers.
Adoption barriersclaude-sonnet-55/5Physicians must be licensed and legally accountable for diagnosis and treatment decisions; malpractice liability and regulatory requirements make full substitution essentially prohibited today.
Cost vs. human wageclaude-haiku-4-5-202510011/5Current AI tools for clinical decision support require significant oversight, validation, and integration costs that do not yet approach the cost-offset of a general internist's wage, especially when liability and quality assurance are factored in.
Cost vs. human wageclaude-sonnet-52/5AI tools reduce documentation and lookup time cheaply, but full task cost still requires a licensed physician for exam, judgment, and liability, so overall cost savings are modest.
Technical feasibility todayclaude-haiku-4-5-202510012/5No deployed AI product reliably performs full clinical management of complex illnesses independently; decision-support tools exist but require physician validation at every step. Clinical deployment remains experimental or narrow (e.g., screening aids), not production-grade autonomous management.
Technical feasibility todayclaude-sonnet-52/5AI diagnostic decision-support tools exist and are used in some clinics, but no deployed product independently manages patient treatment plans reliably across the full range of acute and chronic conditions.

Monitor patients' conditions and progress and reevaluate treatments as necessary.

14

CI 7–20 · exposure 17 · augmentation 75 · importance 4.3/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5While healthcare has digitized workflows, active adoption of AI agents for autonomous monitoring and treatment reevaluation remains limited to pilots and research settings. Regulatory caution and liability concerns slow production deployment in this domain.
Sector adoption velocityclaude-sonnet-52/5Healthcare is a comparatively slow-adopting sector for clinical decision-making automation due to regulation, liability, and integration complexity, though EHR-based alerting is spreading gradually.
Augmentation potentialclaude-haiku-4-5-202510014/5AI tools are already augmenting physicians through EHR data synthesis, predictive alerts, pattern detection in labs and imaging, and treatment suggestion support, materially raising productivity in monitoring workflows while physicians retain decision authority.
Augmentation potentialclaude-sonnet-54/5AI-based clinical decision support, risk-prediction models, and data summarization tools meaningfully help physicians track patient status and flag reevaluation triggers, improving efficiency while the physician remains the decision-maker.
Task automatabilityclaude-haiku-4-5-202510012/5AI can assist with data synthesis and pattern detection in vitals/labs, but the task requires integrating complex clinical judgment, patient-specific context, and real-time reassessment that depends on nuanced human-physician expertise and accountability. Current systems cannot reliably perform full end-to-end monitoring with treatment reevaluation at the 50% time-saving threshold.
Task automatabilityclaude-sonnet-51/5This requires longitudinal clinical judgment, physical examination, patient interaction, and accountability for treatment decisions that current AI cannot perform end-to-end without a physician driving the process.
Adoption barriersclaude-haiku-4-5-202510015/5Treatment decisions and patient progress monitoring are core functions that require physician licensure, malpractice liability, and regulatory authority under medical practice acts. Legal and fiduciary duties place hard barriers on autonomous performance of this task.
Adoption barriersclaude-sonnet-55/5Only a licensed physician can legally reevaluate and modify treatment plans, and medical liability for missed or wrong decisions is high, creating strong regulatory and legal barriers.
Cost vs. human wageclaude-haiku-4-5-202510011/5The integration cost, oversight requirements, and liability exposure for autonomous AI monitoring far exceed the marginal cost of a physician reviewing patient data. All-in, current AI solutions are more expensive than direct physician monitoring.
Cost vs. human wageclaude-sonnet-52/5AI tools can cheaply flag data points, but the actual monitoring and treatment reevaluation still requires a licensed physician's time and judgment, so overall cost savings are modest given required oversight.
Technical feasibility todayclaude-haiku-4-5-202510012/5Some EHR-integrated monitoring tools exist (dashboards, alert systems), but they are narrow in scope and function primarily as assistive alerts rather than autonomous decision-making systems. No deployed product reliably performs independent patient monitoring and treatment reevaluation without physician oversight.
Technical feasibility todayclaude-sonnet-52/5Products exist for flagging abnormal labs, vitals trends, or decision support alerts, but no deployed system autonomously monitors patients and reevaluates treatment plans without physician oversight.

Prescribe or administer medication, therapy, and other specialized medical care to treat or prevent illness, disease, or injury.

12

CI 3–21 · exposure 17 · augmentation 75 · importance 4.5/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5While large healthcare systems are adopting clinical decision-support tools, autonomous prescribing automation is nearly absent in production; adoption remains slow and cautious due to regulatory, liability, and cultural resistance in medicine.
Sector adoption velocityclaude-sonnet-52/5Healthcare overall adopts AI unevenly and cautiously for core clinical decision-making due to regulatory, liability, and safety constraints, despite faster uptake in documentation and diagnostics support.
Augmentation potentialclaude-haiku-4-5-202510014/5AI significantly augments physician prescribing through real-time drug interaction alerts, dosing recommendations, evidence-based guideline summaries, and contraindication screening, allowing physicians to work faster and more safely while retaining final decision authority.
Augmentation potentialclaude-sonnet-54/5AI clinical decision support, drug interaction checkers, and diagnostic aids meaningfully help physicians choose and manage treatments, though the physician remains the decision-maker and administrator.
Task automatabilityclaude-haiku-4-5-202510012/5AI can assist with medication selection and contraindication checking, but cannot replicate the full end-to-end task of diagnosing complex patients, assessing contextual factors, communicating with patients, and making nuanced clinical judgments that meet the 50% time-saving threshold at equal quality.
Task automatabilityclaude-sonnet-51/5Prescribing and administering treatment requires physical presence, physical examination, legal authority, and real-time clinical judgment integrating patient-specific factors that current AI cannot execute end-to-end.
Adoption barriersclaude-haiku-4-5-202510015/5Legal and regulatory barriers are absolute: only a licensed physician can prescribe medication in most jurisdictions; liability for adverse events rests with the prescriber, and malpractice risk is severe, making any fully autonomous AI system infeasible regardless of technical capability.
Adoption barriersclaude-sonnet-55/5Prescribing controlled substances and administering treatment is tightly regulated, requiring a licensed physician's authorization, with high liability and legal barriers preventing AI substitution.
Cost vs. human wageclaude-haiku-4-5-202510011/5The AI infrastructure for clinical decision support is relatively expensive compared to automating purely administrative tasks, and the loaded physician wage remains high, making the cost ratio unfavorable for full substitution.
Cost vs. human wageclaude-sonnet-51/5AI cannot substitute for the licensed act itself, so there is no comparable AI-only cost; any AI use adds to rather than replaces the human's cost of service.
Technical feasibility todayclaude-haiku-4-5-202510013/5Decision-support systems and clinical AI tools exist in production (e.g., drug interaction checkers, dosing calculators), but they operate narrowly and require substantial physician oversight; no system reliably replaces the prescribing task end-to-end without a licensed physician in the loop.
Technical feasibility todayclaude-sonnet-51/5No deployed product independently prescribes or administers medication or therapy; AI tools exist only as decision-support inputs to a physician who retains full responsibility.

Provide and manage long-term, comprehensive medical care, including diagnosis and nonsurgical treatment of diseases, for adult patients in an office or hospital.

11

CI 3–20 · exposure 13 · augmentation 75 · importance 4.4/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510012/5Healthcare is a heavily regulated, risk-averse sector with slow adoption of autonomous AI. While diagnostic AI pilots are emerging, actual displacement of physician tasks in comprehensive care management remains minimal, and adoption velocity is well below technology and professional services sectors.
Sector adoption velocityclaude-sonnet-52/5Healthcare has historically been a slow adopter of AI at the point of clinical decision-making due to regulation, liability, and integration complexity, despite growing pilot programs.
Augmentation potentialclaude-haiku-4-5-202510014/5AI significantly assists physicians today through diagnostic decision support, literature summarization, clinical guideline retrieval, administrative documentation, and drug interaction checking. These tools raise physician productivity and reduce cognitive load while the physician retains full decision-making authority.
Augmentation potentialclaude-sonnet-54/5AI tools (e.g., ambient documentation, diagnostic decision support, literature synthesis) meaningfully boost physician efficiency and information access while the physician retains full clinical responsibility.
Task automatabilityclaude-haiku-4-5-202510012/5While AI can assist with diagnostic support and treatment protocol suggestions, the task requires complex clinical judgment, patient rapport, real-time physical examination, and ongoing care management. Current AI systems cannot reliably perform the full end-to-end task of comprehensive long-term care management with the required safety and quality standards.
Task automatabilityclaude-sonnet-51/5Comprehensive long-term care requires physical exams, longitudinal relationship management, complex judgment across comorbidities, and legal accountability that current AI cannot perform end-to-end.
Adoption barriersclaude-haiku-4-5-202510015/5Legal and regulatory frameworks require a licensed physician to perform diagnosis, prescribe treatment, and maintain responsibility for patient outcomes. Medical licensing boards, state laws, and liability frameworks all mandate human physician accountability for internal medicine practice.
Adoption barriersclaude-sonnet-55/5Medical licensure, malpractice liability, and legal requirements mandate a licensed physician to diagnose and manage treatment, creating hard regulatory and liability barriers to automation.
Cost vs. human wageclaude-haiku-4-5-202510011/5The total cost of AI systems (infrastructure, validation, oversight, malpractice liability) plus the still-necessary physician oversight far exceeds the cost of direct physician care. AI cannot reduce the need for a licensed physician in this context.
Cost vs. human wageclaude-sonnet-52/5AI can cheaply assist with documentation or triage, but the full scope of longitudinal diagnosis/management still requires physician time, oversight, and liability coverage, keeping overall costs comparable to human-delivered care.
Technical feasibility todayclaude-haiku-4-5-202510012/5No deployed product reliably performs comprehensive internal medicine care independently. AI tools exist for narrower functions (diagnostic suggestions, literature review, documentation), but production systems do not manage the full scope of long-term patient care with comparable outcomes to human physicians.
Technical feasibility todayclaude-sonnet-51/5No deployed product independently diagnoses and manages ongoing patient care; AI tools remain decision-support adjuncts used alongside physicians, not substitutes.

Direct and coordinate activities of nurses, students, assistants, specialists, therapists, and other medical staff.

6

CI 0–11 · exposure 8 · augmentation 38 · importance 3.8/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5Healthcare remains heavily regulated with strict credentialing and liability requirements; there is no meaningful adoption of AI-driven staff direction in production clinical settings, and regulatory barriers make such adoption unlikely in the foreseeable future.
Sector adoption velocityclaude-sonnet-52/5Healthcare administrative coordination is adopting AI tools slowly for scheduling/documentation, but leadership/coordination of clinical teams shows minimal automation in production.
Augmentation potentialclaude-haiku-4-5-202510012/5AI could marginally assist with scheduling optimization, communication logistics, or task tracking, but current systems offer limited augmentation to the core judgment and interpersonal work of directing clinical teams. Most augmentation remains low-impact administrative support.
Augmentation potentialclaude-sonnet-53/5AI can assist with scheduling, task tracking, and communication support among care teams, moderately easing coordination burdens without replacing the physician's directive role.
Task automatabilityclaude-haiku-4-5-202510011/5Directing and coordinating medical staff requires real-time human judgment, interpersonal negotiation, conflict resolution, and contextual decision-making that current AI systems cannot perform autonomously. The task fundamentally depends on hierarchical authority and accountability that cannot be delegated to AI.
Task automatabilityclaude-sonnet-52/5Directing and coordinating multidisciplinary staff involves real-time judgment, interpersonal leadership, and accountability that current AI cannot execute end-to-end; at best AI can support scheduling or communication logistics.'
Adoption barriersclaude-haiku-4-5-202510015/5Legal, regulatory, and professional standards require a licensed physician to direct clinical care and supervise medical personnel; no AI can substitute for this licensed responsibility. Liability and accreditation standards explicitly mandate human leadership in clinical coordination.
Adoption barriersclaude-sonnet-55/5Directing medical staff requires licensure, legal accountability, and clinical authority that only a physician can hold, creating hard regulatory and liability barriers.
Cost vs. human wageclaude-haiku-4-5-202510011/5An AI system providing meaningful staff direction would require significant specialized infrastructure, customization, and oversight—likely exceeding the cost of a physician's time spent on coordination, especially since the physician must remain accountable.
Cost vs. human wageclaude-sonnet-51/5There is no AI substitute performing this coordination role, so cost comparison favors the human physician entirely.
Technical feasibility todayclaude-haiku-4-5-202510011/5No deployed AI system can autonomously direct clinical staff or make binding coordination decisions; this remains a human leadership function in all production medical settings. AI tools may assist with scheduling or task tracking, but not the actual direction and coordination.
Technical feasibility todayclaude-sonnet-51/5No deployed product autonomously directs clinical teams or coordinates staff activities in production; this remains a human managerial and clinical leadership function.

Provide consulting services to other doctors caring for patients with special or difficult problems.

4

CI 0–7 · exposure 5 · augmentation 63 · importance 4.1/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5Healthcare adoption of AI for core clinical decision-making remains in pilot and research phases; active substitution of physician consultation is virtually nonexistent, with only supportive tools (e.g., diagnostic aids) in limited use. Regulatory, liability, and professional culture barriers keep adoption extremely slow.
Sector adoption velocityclaude-sonnet-52/5Healthcare overall has slow, cautious AI adoption for high-stakes judgment tasks, with pilots for diagnostic support but not for full consultative replacement.
Augmentation potentialclaude-haiku-4-5-202510013/5AI-powered literature search, differential diagnosis checklists, and evidence synthesis can assist a consulting physician in preparing or formulating recommendations, improving speed and thoroughness. However, augmentation is limited to knowledge retrieval and organization; the core intellectual and interpersonal work of consultation remains fundamentally human-dependent.
Augmentation potentialclaude-sonnet-54/5AI tools can help consulting physicians quickly pull relevant literature, flag differential diagnoses, and summarize patient records, meaningfully speeding preparation and analysis even though the human remains central.
Task automatabilityclaude-haiku-4-5-202510011/5This task requires nuanced clinical judgment, synthesis of complex medical information across specialties, and interactive dialogue with other physicians to address patient-specific diagnostic or management dilemmas. Current AI cannot reliably replace the human expertise and real-time interactive consultation essential to the task.
Task automatabilityclaude-sonnet-51/5This requires integrative clinical judgment, accountability, and nuanced synthesis of complex patient data across specialties, which current AI cannot perform end-to-end at equal quality with major time savings.
Adoption barriersclaude-haiku-4-5-202510015/5Consulting physicians are required by law and medical ethics to sign off on clinical recommendations; malpractice liability, licensing requirements, and the standard of care all mandate that a licensed physician provide the consultation. Patients and referring clinicians expect human professional judgment and accountability.
Adoption barriersclaude-sonnet-55/5Medical licensure, liability for consulting advice, and legal/ethical requirements mean only a licensed physician can provide this kind of consultation and bear responsibility for it.
Cost vs. human wageclaude-haiku-4-5-202510011/5The loaded cost of an internal medicine physician providing consultation far exceeds any current AI inference or oversight cost, but the task itself cannot be substituted; the economic comparison is moot because AI cannot yet perform it reliably at the quality required.
Cost vs. human wageclaude-sonnet-51/5There is no viable AI substitute performing this task, so no meaningful cost comparison favors AI; human specialist consultation remains the only real option.
Technical feasibility todayclaude-haiku-4-5-202510012/5While AI can assist with differential diagnosis suggestions and literature synthesis, no deployed product reliably performs end-to-end consulting—which requires understanding the full clinical context from another physician, offering thoughtful recommendations, and being available for follow-up discussion. AI tools exist to support these elements but not to replace the consultant physician.
Technical feasibility todayclaude-sonnet-51/5No deployed product provides physician-to-physician specialist consultation on difficult cases in production; AI clinical decision support tools exist but do not replace this consultative role.

Immunize patients to protect them from preventable diseases.

0

CI 0–0 · exposure 0 · augmentation 38 · importance 4.0/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5There is no meaningful adoption of AI for performing immunizations because the task fundamentally requires human medical licensure and physical capability. Adoption velocity is not applicable when legal and physical constraints prevent automation entirely.
Sector adoption velocityclaude-sonnet-51/5Physical, hands-on clinical procedures in healthcare delivery show minimal AI displacement; adoption is confined to administrative/documentation support, not the physical task itself.
Augmentation potentialclaude-haiku-4-5-202510013/5AI can assist physicians with immunization through reminder systems, patient eligibility screening, documentation support, and contraindication checking, moderately enhancing workflow efficiency. However, the human physician must remain in direct control of clinical assessment and vaccine administration.
Augmentation potentialclaude-sonnet-52/5AI can help with scheduling, patient reminders, screening for contraindications, and documentation, but offers little assistance for the physical act of immunizing a patient.
Task automatabilityclaude-haiku-4-5-202510011/5Immunization requires direct physical contact with patients, fine motor control for injection administration, and real-time clinical judgment about patient eligibility and contraindications. Current AI systems cannot perform the injection itself or make the contextual medical decisions required at the point of care.
Task automatabilityclaude-sonnet-51/5Physically administering an immunization and managing the clinical encounter (screening, consent, monitoring for reactions) requires physical presence and hands-on action that current AI cannot perform.
Adoption barriersclaude-haiku-4-5-202510015/5Immunization is legally restricted to licensed healthcare providers (physicians, nurses, pharmacists) in virtually all jurisdictions. Regulatory and licensing requirements create hard barriers to any substitution of human decision-making and administration with autonomous systems.
Adoption barriersclaude-sonnet-55/5Vaccination requires licensed medical/nursing personnel, informed consent processes, and liability for adverse reactions, creating hard regulatory and legal barriers to automation.
Cost vs. human wageclaude-haiku-4-5-202510011/5The cost comparison is not meaningful since AI cannot perform this task at all. The labor cost of a physician administering vaccines cannot be replaced by AI systems that cannot physically perform injections or independently manage clinical decision-making.
Cost vs. human wageclaude-sonnet-51/5AI has no capability to substitute for the physical act of injection, so cost comparison favors the human by default; robotics for this are not commercially deployed.
Technical feasibility todayclaude-haiku-4-5-202510011/5No deployed AI system can autonomously administer vaccines or make independent immunization decisions in clinical practice. While AI can assist with reminder systems or documentation, the core task of physically immunizing a patient remains exclusively within human medical practice.
Technical feasibility todayclaude-sonnet-51/5No deployed AI product administers vaccines; this remains entirely a manual clinical/nursing task requiring physical execution.

Operate on patients to remove, repair, or improve functioning of diseased or injured body parts and systems.

0

CI 0–0 · exposure 0 · augmentation 50 · importance 3.7/5 · click for rater detail

Panel ratings per dimension for this task
DimensionModelRatingRationale
Sector adoption velocityclaude-haiku-4-5-202510011/5Surgical AI adoption remains confined to supportive roles (imaging analysis, surgical planning) within human-controlled procedures. Autonomous or semi-autonomous surgical performance has not achieved meaningful real-world deployment despite decades of robotics development; adoption velocity is extremely slow.
Sector adoption velocityclaude-sonnet-51/5Surgical practice remains a physical, hands-on medical specialty with minimal AI autonomy adoption; robotic assistance adoption is slow and still fully human-directed.
Augmentation potentialclaude-haiku-4-5-202510013/5AI assists surgeons through preoperative imaging analysis, intraoperative guidance systems, and robotic platforms that enhance precision and reduce fatigue, but these are assistive augmentations rather than transformative—the surgeon remains the decision-maker and primary executor of the procedure.
Augmentation potentialclaude-sonnet-53/5AI and robotic-assisted surgical platforms (e.g., image guidance, robotic arms) enhance precision and visualization for surgeons, though this occupation profile (general internal medicine) rarely performs surgery directly.
Task automatabilityclaude-haiku-4-5-202510011/5Surgical operation requires precise physical manipulation of tissue, real-time intraoperative decision-making based on visual and tactile feedback, and immediate response to complications. Current AI systems cannot perform end-to-end surgical procedures independently; they lack embodied dexterity and the sensorimotor integration necessary for safe tissue handling.
Task automatabilityclaude-sonnet-51/5Surgical operation requires physical manipulation, fine motor skill, and real-time judgment that current AI systems cannot perform end-to-end; this is a physical-world task outside language/vision model capability.:
Adoption barriersclaude-haiku-4-5-202510015/5Legal and regulatory barriers are absolute: only licensed physicians can perform surgery, and malpractice liability, informed consent requirements, and medical licensing laws create hard barriers to autonomous substitution. Patient safety regulations and credentialing boards explicitly require a licensed surgeon to direct all surgical procedures.
Adoption barriersclaude-sonnet-55/5Surgery legally and professionally must be performed by a licensed physician/surgeon, with strict liability, malpractice, and regulatory frameworks requiring human accountability and physical presence.
Cost vs. human wageclaude-haiku-4-5-202510011/5The infrastructure, maintenance, training, and surgical oversight required for AI-assisted surgery remain extremely expensive compared to a surgeon's labor, especially when accounting for liability, regulatory compliance, and the still-necessary human surgeon's involvement throughout.
Cost vs. human wageclaude-sonnet-51/5There is no viable AI substitute for surgical operation, so cost comparison is moot; any AI-assisted robotic system adds substantial capital and oversight cost rather than reducing it.
Technical feasibility todayclaude-haiku-4-5-202510011/5No deployed AI system can autonomously perform surgical operations on patients. While surgical robots exist (da Vinci), they are operated by licensed surgeons who maintain full control and decision-making authority; the human surgeon remains the primary agent, not the AI.
Technical feasibility todayclaude-sonnet-51/5No deployed AI product autonomously performs surgery; robotic surgical systems exist but are human-operated tools, not autonomous agents performing the procedure.

Related occupations — Healthcare Practitioners & Technical

How to read this

A high substitution score does not mean this job disappears — it means a large share of its current tasks face replacement pressure, so the mix of tasks is likely to change. High augmentation alongside substitution typically means the occupation reorganizes around the protected tasks. Wide confidence intervals mean the rater panel disagreed: treat those scores as open questions, not verdicts.

What would change this score

New model capabilities (automatability, feasibility), falling inference costs (cost ratio), regulation and licensing shifts (barriers), and measured sector adoption (velocity) all re-enter at every index release. Each release is recomputed, versioned and kept queryable — scores are claims with a date on them, not permanent labels.