Medical Scientists, Except Epidemiologists
19-1042.00Median wage $103,410/yr172,340 employed (US)Rank #605 of 923 scored · top 66% by substitution
Conduct research dealing with the understanding of human diseases and the improvement of human health. Engage in clinical investigation, research and development, or other related activities.
Sub-scores
0–100 · band = confidence interval from rater disagreement
Substitution23
Exposure21
Augmentation66
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
14 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%21
panel mean rating 1.9/5 → substitution pressure 21/100
Technical feasibility todayw 20%21
panel mean rating 1.9/5 → substitution pressure 21/100
Cost vs. human wagew 15%21
panel mean rating 1.8/5 → substitution pressure 21/100
Adoption barriersw 20%inverted — strong barriers lower the score27
panel mean rating 3.9/5 (barrier strength) → substitution pressure 27/100
Sector adoption velocityw 10%27
panel mean rating 2.1/5 → substitution pressure 27/100
Task breakdown (14 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.
Write applications for research grants.
42CI 30–55 · exposure 42 · augmentation 88 · importance 4.0/5 · click for rater detail
Write applications for research grants.
42| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Medical scientists operate in highly regulated, credential-dependent sectors where personal reputation and institutional review are paramount; while some labs use AI for drafting, adoption remains cautious and supplementary, not displacing the core grant-writing role. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Academic and biomedical research settings show growing but uneven adoption of AI writing tools; usage is common informally but institutional policies and NIH/NSF guidance on AI-assisted proposals are still evolving, so deep production-level integration is not yet universal. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI tools substantially augment grant writing by accelerating first drafts, suggesting literature synthesis, improving clarity, and freeing scientists to focus on strategic scientific framing and institutional positioning—a clear productivity gain while the scientist remains the authoritative decision-maker. |
| Augmentation potential | claude-sonnet-5 | 5/5 | AI substantially speeds up drafting, editing, literature summarization, and formatting for grant applications, letting scientists focus more time on scientific content and strategy while staying fully in the loop. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can draft sections of grant applications (literature review, methodology summaries), the task requires deep domain expertise, novel scientific hypothesis framing, and strategic positioning of the research contribution—elements that currently demand substantial human oversight and revision rather than achieving 50% time savings at equal quality. |
| Task automatability | claude-sonnet-5 | 3/5 | AI can draft substantial portions of grant text (background, significance, methods boilerplate) given inputs, but crafting a competitive, novel, scientifically sound proposal requires human expertise, original hypotheses, and strategic framing that current AI cannot fully replicate end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Grant agencies and institutional review boards typically require the principal investigator's signature, certification of originality, and demonstrated scientific accountability; liability for misstatements of prior work and regulatory/funder trust in human authorship create meaningful legal and professional barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 2/5 | No licensing requirement for grant writing itself, but funding agencies generally expect PI authorship, integrity certifications, and some grant bodies have disclosure rules on AI use, creating moderate friction. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI writing tools cost $20–100/month, while a medical scientist's labor cost for grant writing (10–20 hours per application at ~$40–80/hour loaded) ranges $400–1600; even with modest time savings, the cost-per-output remains dominated by human labor, making full substitution economically marginal. |
| Cost vs. human wage | claude-sonnet-5 | 3/5 | AI drafting assistance costs very little per use, but the human scientist's time for review, original research design, and iteration remains substantial, making the effective cost roughly comparable rather than order-of-magnitude cheaper for full task completion. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | AI writing tools (ChatGPT, specialized scientific writing assistants) exist and are used to draft portions of grant applications, but real-world grant success depends on personalized strategy, funder alignment, and scientific novelty that AI cannot reliably execute end-to-end; tools remain aids rather than independent performers. |
| Technical feasibility today | claude-sonnet-5 | 3/5 | Products like ChatGPT, Claude, and specialized grant-writing tools are used in production by researchers to draft sections and polish language, but they require heavy human editing and fact-checking, and no product reliably produces a fundable application unassisted. |
Standardize drug dosages, methods of immunization, and procedures for manufacture of drugs and medicinal compounds.
36CI 15–57 · exposure 41 · augmentation 88 · importance 4.3/5 · click for rater detail
Standardize drug dosages, methods of immunization, and procedures for manufacture of drugs and medicinal compounds.
36| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 3/5 | Large pharmaceutical firms and contract research organizations are piloting AI-assisted protocol standardization, but adoption remains selective and usually augmentative rather than replacive. Smaller biotech and specialty manufacturers lag significantly, and regulatory conservatism slows sector-wide rollout. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Pharmaceutical R&D is adopting AI for drug discovery and data analysis, but the specific task of standardizing dosages and procedures remains slow to change due to regulatory and safety constraints. |
| Augmentation potential | claude-haiku-4-5-20251001 | 5/5 | AI excels at rapidly surfacing relevant prior data, flagging safety thresholds, and generating candidate standardization protocols from published literature and trial databases, substantially accelerating the expert scientist's ability to synthesize and refine dosage and manufacturing standards. This high-impact assistance keeps the human scientist in full oversight and decision authority. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can meaningfully assist scientists by analyzing large datasets, modeling pharmacokinetics, and drafting standardization documents, significantly speeding up parts of the research and documentation process while humans retain final judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 4/5 | AI systems can now analyze pharmacological data, clinical trial results, and regulatory guidelines to propose standardized dosages and procedures at scale, with minimal human intervention needed for validation. Current tools can cross-reference manufacturing standards, safety thresholds, and population parameters to generate compliant protocols that meet or exceed 50% time savings on routine standardization tasks. |
| Task automatability | claude-sonnet-5 | 2/5 | AI can assist with data analysis, literature review, and drafting protocols, but the core scientific judgment, experimental validation, and regulatory decision-making required to standardize dosages and manufacturing procedures cannot be fully automated today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | FDA and international regulatory bodies require that drug dosages, manufacturing procedures, and immunization protocols meet formal approval and often demand licensed expert sign-off on final standardization decisions. Liability asymmetry and mandatory human accountability create significant friction against full automation, even where AI could perform the underlying analysis. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Drug dosage and manufacturing standardization is tightly regulated (e.g., FDA), requiring credentialed scientists and formal approval processes, making this a hard legal/regulatory barrier. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 4/5 | Once integrated into a firm's data infrastructure, AI systems can process and standardize dosages and procedures at marginal cost per task, while pharmaceutical scientists command high salaries. The all-in cost of AI-assisted standardization (inference + regulatory validation oversight) is substantially lower than the loaded cost of a full-time scientist for high-volume routine work. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI tools can lower costs for data analysis and literature synthesis, but the overall task still requires expensive human expertise, lab work, and regulatory processes, keeping the all-in cost comparable to human-led work. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Deployed pharmaceutical informatics platforms and AI-assisted protocol generators exist in production at major pharma companies, but they typically require significant expert human review and are narrowly scoped to specific drug classes or regulatory frameworks. Error rates on novel compounds or rare populations remain material, limiting full autonomous deployment. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously standardizes drug dosages or manufacturing procedures; this remains a research- and expert-driven process requiring clinical trials and regulatory oversight. |
Prepare and analyze organ, tissue, and cell samples to identify toxicity, bacteria, or microorganisms or to study cell structure.
30CI 25–35 · exposure 30 · augmentation 63 · importance 4.5/5 · click for rater detail
Prepare and analyze organ, tissue, and cell samples to identify toxicity, bacteria, or microorganisms or to study cell structure.
30| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | While digital pathology adoption is growing in large academic and clinical institutions, the sectors performing this task (clinical diagnostics, biotech) are adopting AI slide analysis slowly at scale; most smaller labs and quality-control labs still rely on manual expert review. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Life sciences R&D labs are adopting AI tools for image analysis and data interpretation, but wet-lab and cell biology work overall remains a slower-adopting, physically-grounded domain compared to information-sector automation. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI-assisted image analysis can help pathologists and microbiologists screen or categorize samples faster, reducing the manual microscopy workload; however, the augmentation is limited to image interpretation rather than the full preparation and diagnostic workflow. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI-powered image recognition, pattern detection, and data analysis tools significantly speed up identification of toxicity markers, bacteria, or cell structures, meaningfully augmenting scientists' throughput and accuracy while they remain in the loop. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with image analysis of prepared samples (histopathology classification), the physical preparation of tissue samples, slide creation, and hands-on microscopy examination remain manual. Current AI systems cannot perform the full end-to-end wet-lab work of sample preparation and achieve 50% time savings on the complete task. |
| Task automatability | claude-sonnet-5 | 2/5 | Physical sample preparation and much of the microscopic/histological analysis require hands-on lab work and expert judgment that current AI cannot fully replace end-to-end, though image analysis portions can be partially automated. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Medical sample analysis is heavily regulated; clinical labs typically require licensed laboratory professionals (CLIA certification) to validate results, and liability for misidentification of pathogens or toxins creates strong legal and organizational barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 3/5 | While no strict licensing mandates a human perform this, quality control, reproducibility standards, and liability for misdiagnosis in toxicity/bacterial studies create meaningful institutional friction against pure automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI-powered image analysis is becoming cheaper per slide, but the full task includes expensive wet-lab sample preparation (consumables, equipment, technician time) that automation has not substantially reduced, making the all-in cost comparable to or higher than manual expert work. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Wet-lab sample preparation still requires skilled technicians and physical equipment, so AI only reduces cost for the analysis/image-classification portion, not the full task, keeping overall cost roughly comparable to human labor. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 3/5 | Deployed AI products (e.g., digital pathology platforms from vendors like Paige, Proscia) can analyze prepared histology slides for certain toxicity markers and bacterial identification with reasonable accuracy, but their deployment is still subject to material error rates and regulatory constraints, and they require expert validation. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI-assisted image analysis tools (e.g., digital pathology, microscopy classifiers) exist and are used in research settings, but they handle narrow sub-tasks like pattern detection rather than the full sample prep-to-interpretation pipeline reliably in production. |
Conduct research to develop methodologies, instrumentation, and procedures for medical application, analyzing data and presenting findings to the scientific audience and general public.
29CI 25–32 · exposure 25 · augmentation 75 · importance 4.1/5 · click for rater detail
Conduct research to develop methodologies, instrumentation, and procedures for medical application, analyzing data and presenting findings to the scientific audience and general public.
29| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Adoption of AI in medical research is slow and cautious; tools augment workflow in data analysis and literature mining, but wholesale automation of research design and methodology development remains rare and confined to pilots rather than production substitution. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Biomedical research is adopting AI tools (e.g., for literature review, data analysis, drug discovery) at a moderate pace, with pilots and augmentation common but full autonomous research still rare. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI significantly aids medical research productivity through rapid literature synthesis, statistical analysis, figure generation, manuscript drafting, and data visualization; these assistive tools materially accelerate a researcher's ability to design, execute, and communicate findings while retaining human control over methodology and interpretation. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI substantially assists with literature review, statistical analysis, hypothesis generation, drafting manuscripts, and data visualization, meaningfully boosting researcher productivity while humans retain scientific judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with data analysis, literature synthesis, and presentation drafting, the core work—designing novel methodologies, validating instrumentation, and interpreting unexpected results—requires domain expertise, creative problem-solving, and human judgment that current AI cannot reliably replace end-to-end at 50% time savings and equal quality. |
| Task automatability | claude-sonnet-5 | 2/5 | This task combines novel scientific hypothesis generation, experimental design, hands-on instrumentation development, and communication—AI can assist with data analysis and drafting but cannot autonomously conduct the full research cycle at equal quality today. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Regulatory oversight (FDA, IRB requirements), publication standards requiring human accountability, liability for flawed methodology, and professional licensure/credentials create substantial barriers; research institutions legally and ethically require qualified human researchers to design and validate medical procedures. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing requirement to conduct research itself, but peer review, institutional oversight, grant funding processes, and scientific credibility norms create substantial friction against full AI substitution. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI data analysis and drafting tools reduce peripheral costs, but the core research—wet-lab work, instrument calibration, hypothesis refinement—remains expensive and human-intensive; AI cost per full research output remains high relative to the sunk cost of experienced medical scientists. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI can cut costs on data analysis and literature synthesis, the core research design, experimentation, and validation still require expensive expert scientists, keeping overall cost savings modest. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed product autonomously conducts medical research methodology development or instrumentation validation; AI tools exist for literature review and statistical analysis, but the integrated research pipeline remains human-driven in all production medical research settings. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI tools exist for literature review, statistical analysis, and writing assistance, but no deployed product independently develops novel methodologies or instrumentation for medical research in production settings. |
Investigate cause, progress, life cycle, or mode of transmission of diseases or parasites.
29CI 25–32 · exposure 25 · augmentation 75 · importance 3.9/5 · click for rater detail
Investigate cause, progress, life cycle, or mode of transmission of diseases or parasites.
29| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Biomedical research institutions are adopting AI tools for analysis and screening, but adoption of autonomous investigation remains limited; human scientists remain central to research design and validation. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Biomedical research is adopting AI tools (protein folding, drug discovery, literature synthesis) at a moderate-to-fast pace, though full task automation in bench science remains rare and pilot-stage in most labs. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI significantly augments productivity through literature review, sequence analysis, data mining, and hypothesis prioritization, allowing researchers to focus on experiment design and interpretation rather than information gathering. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI significantly augments this task via literature synthesis, hypothesis generation support, bioinformatics analysis, and pattern detection in complex datasets, substantially boosting researcher productivity while humans retain investigative control. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with literature review, data analysis, and hypothesis generation, investigating disease transmission and cause requires experimental design, biological intuition, and iterative hypothesis testing that still requires substantial human direction and hands-on laboratory work. |
| Task automatability | claude-sonnet-5 | 2/5 | This is open-ended scientific investigation requiring hypothesis generation, experimental design, wet-lab work, and novel discovery, which current AI cannot execute end-to-end despite assisting with literature review and data analysis. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Medical research is heavily regulated (IRB approval, biosafety protocols, validated methodology standards) and requires licensed scientists to design, execute, and sign off on investigations; liability for experimental design and safety is non-transferable to AI systems. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing requirement mandates a human perform this specific task, but institutional review, scientific rigor standards, peer review, and accountability for research findings create meaningful friction against pure AI-driven conclusions. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI tools for data analysis are cheaper than human labor per unit, but the full investigation cost is dominated by experimental equipment, human researcher salary, and wet-lab operations, making overall replacement cost-ineffective. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | AI can cheaply accelerate data analysis and literature synthesis, but the core investigative work still requires expensive laboratory infrastructure, human expertise, and experimentation, keeping overall costs comparable to human-led research. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Current AI systems can support components (literature mining, pattern detection in genomic data) but no deployed product performs end-to-end disease investigation autonomously; experimental validation and wet-lab work remain human-dependent. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI tools (e.g., protein structure prediction, literature mining, genomic analysis) are used in research pipelines, but no deployed product autonomously investigates disease causation or transmission dynamics as a complete task. |
Write and publish articles in scientific journals.
26CI 25–28 · exposure 25 · augmentation 63 · importance 4.0/5 · click for rater detail
Write and publish articles in scientific journals.
26| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Academic publishing is relatively conservative; while some researchers use AI tools for writing assistance and outlining, widespread adoption of AI for article drafting remains limited and often discouraged by guidelines. Most medical scientists still rely on traditional writing and peer collaboration rather than AI-driven article production. |
| Sector adoption velocity | claude-sonnet-5 | 3/5 | Academic and scientific publishing has seen notable uptake of AI writing tools for editing and literature review, but full manuscript generation remains rare and cautious due to journal policies and reputational risk. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI writing assistants meaningfully help researchers improve clarity, structure, and grammar in manuscripts, and can accelerate first-draft generation and literature synthesis. However, augmentation is limited to writing mechanics and organization; AI cannot augment the core scientific reasoning, hypothesis testing, or novelty assessment required to publish in peer-reviewed venues. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI substantially helps with literature review, drafting, grammar, formatting citations, and summarizing data, meaningfully speeding up the writing process while the scientist retains responsibility for content and conclusions. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | AI can draft sections of scientific articles (literature review, methods description) and improve writing clarity, but cannot independently generate novel research insights, validate experimental results, or make judgment calls on scientific contribution and novelty that are core to this task. Full end-to-end automation without substantial human scientific judgment would not meet quality standards for peer-reviewed publication. |
| Task automatability | claude-sonnet-5 | 2/5 | AI can draft sections, summarize literature, and assist with figures/statistics, but the full task—generating novel research findings, ensuring scientific validity, and navigating peer review—requires human expertise and judgment that current AI cannot replace end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Journals and institutions require author accountability and expert sign-off; AI-generated text must be reviewed and certified as scientifically sound by human scientists. Ethical guidelines (ICMJE, publisher policies) restrict AI use in article generation, and authorship requires human responsibility for claims and integrity, creating strong institutional and legal barriers to full substitution. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Journals require named human authors accountable for content, many have policies restricting or requiring disclosure of AI-generated text, and scientific integrity/authorship norms create strong institutional and ethical barriers to full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI writing tools cost pennies per article, but a medical scientist's labor in writing, conceptualization, and revision is modest relative to total research cost. The bottleneck is scientific judgment, not writing time, so cost savings from AI assistance are limited and do not approach an order of magnitude advantage. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI can cheaply assist with drafting and editing portions of a manuscript, the core scientific work (experiments, analysis, interpretation) still requires costly human expert labor, so overall cost savings are modest. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | AI writing assistants (Copilot, Claude) exist and are used by some researchers for drafting and revision, but no deployed product independently writes publication-ready scientific articles from raw data. Products lack the domain expertise and scientific rigor needed to verify claims, contextualize findings, or navigate journal-specific requirements reliably. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI writing assistants (e.g., Grammarly, ChatGPT, Writefull) are used in production for drafting and editing, but no deployed product reliably authors publishable original scientific articles without extensive human oversight and content generation. |
Evaluate effects of drugs, gases, pesticides, parasites, and microorganisms at various levels.
25CI 25–25 · exposure 25 · augmentation 75 · importance 4.6/5 · click for rater detail
Evaluate effects of drugs, gases, pesticides, parasites, and microorganisms at various levels.
25| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | While pharmaceutical and biotech firms deploy AI for target screening and data analysis, actual evaluation of drugs and toxins remains a slow-to-digitize domain reliant on physical labs and regulatory protocol. Adoption of AI is limited to augmentation (modeling, analysis) rather than substitution in production workflows. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Life sciences R&D adopts AI for specific subtasks (molecule screening, literature synthesis) but core bench-based physical evaluation work remains largely non-digitized and slow to change. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI substantially assists medical scientists in literature mining, hypothesis generation, data analysis, pattern recognition in large datasets, and predictive modeling (e.g., QSAR for toxicity prediction). These tools meaningfully raise productivity and accuracy in evaluation workflows while keeping the human scientist in decision-making and experimental control. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI tools significantly help with experimental design suggestions, dose-response modeling, literature synthesis, and statistical analysis, meaningfully speeding up parts of the evaluation workflow. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist in analyzing pre-generated experimental data and literature synthesis, the core task of designing and physically executing evaluations of bioactive substances requires hands-on laboratory work, complex decision-making about experimental protocols, and real-time adjustments that current AI systems cannot perform end-to-end. AI tools may expedite data analysis but cannot achieve 50% time savings on the full task without substantial human involvement in experimental design and execution. |
| Task automatability | claude-sonnet-5 | 2/5 | This requires designing and executing physical experiments (lab work, animal/tissue studies, assays) that AI cannot perform; AI can assist with data analysis but not the core empirical evaluation.' |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Strong regulatory and liability barriers exist: FDA, EPA, and international regulations require qualified human scientists to design, conduct, and certify toxicology and efficacy studies. GLP (Good Laboratory Practice) standards mandate human expertise and accountability; AI cannot sign off on safety or efficacy reports. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Regulatory requirements (FDA, IACUC, GLP standards) mandate qualified scientists to design, conduct, and validate such studies, and results often feed into regulatory submissions requiring human accountability. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI tools for data analysis and literature review cost thousands annually, but full evaluation programs (personnel, equipment, regulatory overhead) remain dominated by highly trained scientist labor ($80k–$150k+ loaded cost). AI supplementation has not achieved cost parity or advantage for the complete task. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Wet-lab experimentation, reagents, and specialized equipment dominate costs; AI only reduces costs for the data-analysis portion, not the physical experimentation, so overall cost savings are modest. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed AI product reliably performs independent drug or toxicological evaluation. AI systems exist for literature review, data mining, and predictive modeling (e.g., QSAR), but these are narrow components; actual evaluation requires wet-lab execution, regulatory compliance, and expert interpretation that remains human-driven in production settings. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | Products exist for statistical analysis, literature review, and toxicology prediction but no deployed system autonomously evaluates biological/chemical effects at the bench level reliably in production. |
Consult with and advise physicians, educators, researchers, and others regarding medical applications of physics, biology, and chemistry.
25CI 25–25 · exposure 25 · augmentation 75 · importance 3.4/5 · click for rater detail
Consult with and advise physicians, educators, researchers, and others regarding medical applications of physics, biology, and chemistry.
25| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Medical research and clinical settings are moving slowly on autonomous AI decision-making due to regulatory scrutiny, liability concerns, and entrenched workflows around expert consultation. AI augmentation is emerging, but replacement of the advisory role itself remains rare in production. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Medical science research settings show cautious AI adoption, primarily for literature review and data analysis support rather than replacing expert consultation, since fields prioritize verified accuracy and liability concerns. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI significantly assists medical scientists by rapidly surveying literature, generating hypotheses, summarizing cross-disciplinary findings, and drafting technical explanations. These capabilities enhance a human consultant's productivity and breadth without replacing their judgment or accountability. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI significantly aids scientists in synthesizing literature, drafting explanations, and cross-referencing multidisciplinary knowledge, boosting productivity even though the scientist remains the consulted expert. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | Consultation and advisory work requires nuanced understanding of context-specific medical applications, integration of domain expertise from multiple fields, and adaptive communication with diverse stakeholders. While AI can draft technical summaries or generate candidate insights, the core value of this task—personalized expert guidance—demands human judgment that current systems cannot reliably replace end-to-end. |
| Task automatability | claude-sonnet-5 | 2/5 | This task requires expert judgment, contextual synthesis, and interpersonal trust-building in advisory relationships that current AI cannot replicate end-to-end; AI can support research but not replace the consultative role. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Consultation that shapes medical research and clinical practice carries liability and regulatory weight. Institutional policy, professional ethics codes, and implicit legal responsibility for expert advice create strong friction against full automation. Physicians and researchers expect human accountability and credential verification from their advisors. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Advisory roles in medical/scientific contexts often require credentialed expertise, professional accountability, and trust in high-stakes decisions, creating substantial barriers to full automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | Medical scientist consultants command significant hourly rates ($100–$200+) reflecting their specialized expertise. AI tools (if used for support) cost far less per inference, but full replacement would require the AI system to assume the liability and trust role—making the effective cost of a deployed autonomous system higher than selective human-assisted use. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | Highly specialized PhD-level scientific advisory expertise cannot be substituted cheaply by AI without significant human oversight, keeping cost savings limited despite AI's low per-query cost. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed product reliably performs expert medical consultation at scale. AI systems can assist with literature synthesis and technical explanation but do not independently serve as trusted advisors to physicians or researchers in production settings. The task requires accountability and contextual depth beyond current system capabilities. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI tools (e.g., literature-summarization assistants, clinical decision support) exist to inform such consultations, but no deployed product autonomously provides expert cross-disciplinary advisory consultation to physicians and researchers reliably. |
Use equipment such as atomic absorption spectrometers, electron microscopes, flow cytometers, or chromatography systems.
23CI 16–30 · exposure 20 · augmentation 50 · importance 3.8/5 · click for rater detail
Use equipment such as atomic absorption spectrometers, electron microscopes, flow cytometers, or chromatography systems.
23| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Adoption of AI for instrument operation remains limited; while data-analysis augmentation is emerging in some research settings, production-scale automation of equipment handling in clinical and research labs is still rare due to cost and regulatory constraints. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Lab automation and robotics adoption in life sciences is growing but remains slow and capital-intensive compared to fast-adopting information sectors; full instrument autonomy is still rare. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can meaningfully assist by automating data interpretation, pattern recognition in instrument readouts, and flagging anomalies, thereby raising scientist productivity; however, the human must remain in the loop for instrument operation itself. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can assist with data interpretation, anomaly detection, and result analysis from instrument outputs, meaningfully aiding scientists even though it doesn't operate the equipment itself. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with data interpretation from these instruments, the hands-on operation of complex analytical equipment—calibration, sample preparation, troubleshooting, and real-time adjustments—requires physical manipulation and domain expertise that current AI systems cannot perform end-to-end with 50% time savings at equal quality. |
| Task automatability | claude-sonnet-5 | 2/5 | Operating specialized lab instruments requires physical manipulation, calibration judgment, and troubleshooting that current AI cannot perform end-to-end; AI can assist with data analysis but not the hands-on equipment operation. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Regulatory requirements (FDA, GLP/GCP compliance), safety protocols for hazardous instruments, and the need for credentialed operators to validate results create substantial adoption friction that prevents straightforward substitution of human operators. |
| Adoption barriers | claude-sonnet-5 | 3/5 | No licensing mandate strictly requires a human to physically run the equipment, but safety protocols, expensive equipment liability, and specialized training create meaningful organizational friction against automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI tools for data analysis are relatively inexpensive, but the human expertise required to operate and maintain specialized analytical equipment remains costly; the all-in cost of AI assistance does not yet undercut the loaded wage of a trained medical scientist for this task. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | Physical instrument operation still requires trained human technicians; AI cannot substitute for the hardware interaction, so no cost savings accrue at the automation level assessed. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | Some AI products exist for post-hoc analysis of instrument output (e.g., image analysis from microscopy), but no deployed system reliably performs the full operational pipeline of equipment setup, execution, and troubleshooting without significant human oversight and correction. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously operates atomic absorption spectrometers, electron microscopes, or flow cytometers in production labs; automation here is limited to lab robotics research and instrument software, not general AI agents. |
Teach principles of medicine and medical and laboratory procedures to physicians, residents, students, and technicians.
19CI 13–25 · exposure 17 · augmentation 63 · importance 4.1/5 · click for rater detail
Teach principles of medicine and medical and laboratory procedures to physicians, residents, students, and technicians.
19| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | While medical institutions experiment with AI-assisted learning platforms and simulations, core teaching responsibilities remain with licensed faculty. Adoption is slow because regulatory requirements, accreditation standards, and the centrality of human mentoring to medical training create structural resistance. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Academic medicine adopts AI slowly for core teaching roles, though e-learning tools are increasingly used as supplements; large-scale replacement of instructors is rare. |
| Augmentation potential | claude-haiku-4-5-20251001 | 3/5 | AI can assist by generating practice questions, summarizing literature, creating virtual patient simulations, and drafting lecture outlines, which would help educators prepare more efficiently. However, the core teaching act—live interaction, clinical judgment modeling, and assessment—remains substantially human-dependent. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI can significantly help generate teaching materials, simulate case studies, provide personalized quizzes, and support curriculum development, enhancing instructor productivity substantially. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Teaching medical principles and procedures requires real-time interaction, assessment of comprehension, mentoring, and adaptive pedagogical responses that current AI cannot reliably deliver end-to-end. While AI can generate educational materials, it cannot replace the hands-on demonstration, live feedback, and credible authority that medical education demands. |
| Task automatability | claude-sonnet-5 | 2/5 | Teaching involves live interaction, adapting to learner needs, mentoring, and hands-on demonstration of lab procedures that current AI cannot fully replicate end-to-end.'},'rating rationale continues below in other fields.'} , |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Medical education is heavily regulated by accrediting bodies, licensing boards, and institutional governance. Teaching is often a credential-protected role (MD/PhD required), and learners expect human mentors with clinical authority; regulatory and organizational barriers strongly protect this function from automation. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Medical education often requires accredited faculty, licensed physicians, and institutional certification, creating strong regulatory and credentialing barriers to full AI substitution. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | Medical education requires specialized expertise, regulatory credibility, and live interaction that an experienced physician-educator commands. AI content generation is cheap, but integrating it into accredited medical curricula with proper oversight still requires significant human investment and institutional costs. |
| Cost vs. human wage | claude-sonnet-5 | 2/5 | While AI can cheaply generate lecture content or quizzes, the overall teaching task still requires costly human oversight, supervision of hands-on training, and credentialing, keeping costs comparable to human instruction. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | AI systems can generate educational content and simulations, but no deployed product reliably handles the full teaching task—which requires dynamic assessment of learner understanding, practical demonstration oversight, and credential-backed authority. Narrow tools exist (e.g., virtual simulations) but not comprehensive teaching solutions in production. |
| Technical feasibility today | claude-sonnet-5 | 2/5 | AI tutoring tools and content generators exist but are not deployed as primary instructors for medical/lab procedure training at scale; human instructors remain essential for accreditation and hands-on skills. |
Plan and direct studies to investigate human or animal disease, preventive methods, and treatments for disease.
16CI 7–25 · exposure 13 · augmentation 75 · importance 4.5/5 · click for rater detail
Plan and direct studies to investigate human or animal disease, preventive methods, and treatments for disease.
16| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Medical research institutions are slow to adopt AI for core research direction; adoption remains experimental and limited to supporting tools (bioinformatics platforms, data analysis). The high stakes, regulatory overhead, and publication culture favor human-led design over AI-autonomous direction. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Biomedical research organizations are adopting AI tools for literature synthesis and data analysis, but adoption for actual study planning and direction remains nascent and pilot-stage. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI can meaningfully assist: literature summarization, identifying data patterns, suggesting statistical approaches, and flagging potential correlations in preliminary data all boost a researcher's productivity and scope. However, the human retains the interpretive and directional role. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI substantially assists with literature reviews, hypothesis generation, experimental design suggestions, grant writing support, and data analysis planning, meaningfully boosting researcher productivity while humans retain direction and judgment. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with literature review, data analysis, and study design suggestions, the core task—planning and directing studies—requires human judgment about research direction, ethical considerations, hypothesis formation, and stakeholder coordination. AI cannot autonomously make strategic research decisions that satisfy the >50% time-saving bar for end-to-end execution. |
| Task automatability | claude-sonnet-5 | 1/5 | Planning and directing research programs requires scientific judgment, creativity, funding strategy, ethical oversight, and leadership that current AI cannot perform end-to-end; AI can only assist with sub-components like literature review or protocol drafting. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Study direction in human or animal research faces strong regulatory barriers: institutional review boards, animal care committees, and funding agency approval all require human sign-off and oversight. Liability for research misconduct, fraud, and safety is legally and professionally anchored to the directing scientist. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Research involving human/animal subjects requires IRB/IACUC approval, PI credentials, funding agency accountability, and regulatory compliance (FDA, NIH) that legally requires qualified human oversight and sign-off. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | AI can reduce costs on specific subtasks (literature synthesis, data preprocessing), but the full loaded cost of medical scientists (salary, expertise, accountability) far exceeds current AI inference and integration costs. The expertise premium is too high for cost parity. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no AI system that substitutes for the human role of directing a study, so no meaningful cost comparison for full task replacement exists; the human remains essential and costly to replace. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | No deployed product reliably performs study planning and direction end-to-end. AI tools exist for literature mining and statistical analysis, but directing research—setting priorities, managing teams, navigating regulatory pathways, interpreting unexpected results—remains human-dependent in practice. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously plans and directs disease research studies; this remains squarely a human-led scientific leadership function with only isolated AI-assisted sub-tasks. |
Study animal and human health and physiological processes.
16CI 7–25 · exposure 13 · augmentation 75 · importance 4.1/5 · click for rater detail
Study animal and human health and physiological processes.
16| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 2/5 | Academic and biomedical research sectors adopt AI slowly for specific analytical tasks but retain strong human-centric research cultures. Production-level displacement of medical scientists by autonomous AI is negligible; pilots using AI for literature or data tasks exist but do not materially alter hiring or research direction. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Biomedical research is adopting AI tools (e.g., for literature synthesis, protein structure prediction) but wet-lab physiological study remains slow to digitize and largely unautomated in production settings. |
| Augmentation potential | claude-haiku-4-5-20251001 | 4/5 | AI substantially augments medical scientists through literature summarization, statistical analysis, image recognition (pathology, imaging), and data visualization. These tools demonstrably raise researcher productivity and hypothesis generation speed while keeping the human scientist in control of experimental design, interpretation, and decision-making. |
| Augmentation potential | claude-sonnet-5 | 4/5 | AI substantially assists scientists by mining literature, analyzing complex datasets, generating hypotheses, and modeling biological systems, meaningfully boosting research productivity while humans remain central to the actual studies. |
| Task automatability | claude-haiku-4-5-20251001 | 2/5 | While AI can assist with literature review, data analysis, and hypothesis generation, the core task of studying health and physiological processes requires hands-on experimental design, lab work, animal/human subject interaction, and judgment that cannot be automated end-to-end. AI cannot independently conduct wet lab experiments, manage living subjects, or replicate the intuitive troubleshooting essential to discovery science. |
| Task automatability | claude-sonnet-5 | 1/5 | This is a broad, open-ended scientific investigation task involving hypothesis generation, experimental design, wet-lab work, and interpretation of novel biological phenomena, none of which current AI can execute end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | IRB approval, animal care protocols, biosafety regulations, and institutional liability frameworks legally require qualified human researchers to design and oversee studies involving human or animal subjects. Professional accountability for research integrity and experimental validity creates strong regulatory and organizational barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Research involving human/animal subjects is heavily regulated (IRB/IACUC approval, credentialed PI oversight), and scientific conclusions require accountable expert judgment, creating strong institutional and ethical barriers to automation. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 2/5 | The cost of AI tools (language models, image analysis) plus human oversight is comparable to or higher than the salary of a medical scientist when factoring in equipment, validation, and the irreducible need for skilled human direction. AI does not yet provide order-of-magnitude savings for the full research workflow. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | The task requires physical experimentation, biological specimens, and expert judgment that AI cannot substitute for, so there is no meaningful AI-only cost comparison—human researchers remain essential. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 2/5 | AI tools exist for specific subtasks (literature mining, statistical analysis, image analysis) but no deployed product reliably performs the full task of studying physiological processes. Real-world biomedical research remains dependent on human researchers directing experiments, interpreting anomalies, and making critical design decisions that deployed AI cannot yet match at production quality. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed product autonomously studies animal or human physiology; AI tools are used only as narrow aids (literature review, data analysis) within a human-led research process. |
Confer with health departments, industry personnel, physicians, and others to develop health safety standards and public health improvement programs.
6CI 5–7 · exposure 0 · augmentation 38 · importance 3.7/5 · click for rater detail
Confer with health departments, industry personnel, physicians, and others to develop health safety standards and public health improvement programs.
6| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Health departments and regulatory bodies remain conservative, hierarchical, and credential-dependent. Adoption of AI for core conferencing and standard-setting is minimal; human leadership remains the norm in public health governance. |
| Sector adoption velocity | claude-sonnet-5 | 2/5 | Public health and government-adjacent scientific policy work adopts AI tools slowly relative to sectors like finance or software, with most current use limited to research support rather than deliberative processes. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI could assist by summarizing evidence, drafting meeting agendas, or preparing background materials, but the core task—building trust and consensus among stakeholders—remains firmly human-centered. Assistance value is limited and peripheral to the decision-making. |
| Augmentation potential | claude-sonnet-5 | 3/5 | AI can help synthesize research, draft position papers, summarize regulations, and prepare background materials that support these conferences, meaningfully aiding preparation even though the interactions themselves remain human-led. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | This task requires nuanced dialogue, stakeholder negotiation, and consensus-building across diverse perspectives—activities that demand human judgment, contextual understanding, and interpersonal trust. Current AI cannot meaningfully replace the iterative, relationship-dependent conferencing process. |
| Task automatability | claude-sonnet-5 | 1/5 | This is a live, multi-stakeholder deliberative and relationship-based task requiring negotiation, trust-building, and consensus formation among diverse experts and institutions—something current AI cannot substitute for end-to-end. |
| Adoption barriers | claude-haiku-4-5-20251001 | 4/5 | Health and safety standard-setting often requires legal authority, professional credentials (MD, public health degrees), and formal organizational sign-off. Regulatory and liability frameworks typically mandate human medical scientists and health officials as decision-makers in these conferences. |
| Adoption barriers | claude-sonnet-5 | 4/5 | Public health standard-setting often involves regulatory processes, credentialed expert input, and institutional accountability that require qualified humans to represent and negotiate on behalf of organizations. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | The task involves high-stakes policy development where human expertise commands significant salary and the conferencing itself is the primary value. AI tools (if deployed as assistants) would add marginal cost with limited displacement of the core labor. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | AI cannot perform the actual conferring and consensus-building, so the human expert must still be paid for the full task; any AI cost would be additive, not substitutive. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed product reliably conducts independent stakeholder conferences or develops consensus-based health standards. While AI can draft documents or summarize existing standards, it cannot authentically participate in or lead multi-party deliberative processes at scale. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed AI product conducts or substitutes for interagency consultations to develop health/safety standards; this remains a research-stage capability at best, not a production one. |
Follow strict safety procedures when handling toxic materials to avoid contamination.
0CI 0–0 · exposure 0 · augmentation 25 · importance 4.7/5 · click for rater detail
Follow strict safety procedures when handling toxic materials to avoid contamination.
0| Dimension | Model | Rating | Rationale |
|---|---|---|---|
| Sector adoption velocity | claude-haiku-4-5-20251001 | 1/5 | Adoption of AI/automation for toxic material handling is minimal; regulatory, liability, and safety requirements mean most labs retain human-supervised procedures, and organizational inertia around validated safety protocols is very high. |
| Sector adoption velocity | claude-sonnet-5 | 1/5 | Physical lab safety execution is an area with essentially no AI adoption; robotics for hazardous material handling remains experimental and far from broad deployment in this sector. |
| Augmentation potential | claude-haiku-4-5-20251001 | 2/5 | AI might assist by providing real-time procedural reminders, flagging deviations from protocols, or monitoring sensor data, but the core safety task requires direct human execution and judgment, limiting augmentation potential. |
| Augmentation potential | claude-sonnet-5 | 2/5 | AI can help generate or check safety protocols, SDS lookups, or training materials, but it does not meaningfully assist during the actual physical execution of safety procedures. |
| Task automatability | claude-haiku-4-5-20251001 | 1/5 | Following strict safety procedures for toxic material handling requires real-time environmental monitoring, physical dexterity, contextual judgment, and immediate corrective action in unpredictable lab conditions. Current AI cannot perform the embodied, procedural aspects of physically safe handling end-to-end. |
| Task automatability | claude-sonnet-5 | 1/5 | Physical handling of toxic materials and adherence to lab safety protocols requires embodied action, judgment, and manual dexterity that current AI cannot perform; this is a hands-on physical task, not an information task.6 |
| Adoption barriers | claude-haiku-4-5-20251001 | 5/5 | Regulatory frameworks (OSHA, EPA, laboratory accreditation standards) typically mandate that a qualified human scientist with proper training and licensing must directly oversee or personally execute handling of toxic materials, creating hard legal barriers to full automation. |
| Adoption barriers | claude-sonnet-5 | 5/5 | Handling toxic/hazardous materials is governed by strict regulatory, institutional biosafety, and legal liability requirements mandating trained, certified human personnel to perform and be accountable for the work. |
| Cost vs. human wage | claude-haiku-4-5-20251001 | 1/5 | Autonomous robotic systems capable of handling toxic materials with the safety compliance required are extremely expensive and specialized, far exceeding the cost of a trained human scientist for this task. |
| Cost vs. human wage | claude-sonnet-5 | 1/5 | There is no viable AI substitute performing this physical safety task, so no meaningful cost comparison exists; the human is required at any cost. |
| Technical feasibility today | claude-haiku-4-5-20251001 | 1/5 | No deployed AI system can autonomously execute physical laboratory safety procedures or monitor and respond to contamination risks in real-world lab environments. This task fundamentally requires embodied physical action and human judgment that current systems cannot deliver. |
| Technical feasibility today | claude-sonnet-5 | 1/5 | No deployed AI product physically handles hazardous materials or executes safety procedures in a wet lab; this remains firmly in the domain of human lab technicians and scientists. |
Related occupations — Life, Physical & Social Science
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.