Microsoft study names 20 jobs least likely to be replaced by AI — healthcare and blue‑collar work look safest
Microsoft’s research, using 200,000 anonymized Bing Copilot chats and O*NET task mappings, identifies 20 occupations least matched by current generative AI — largely hands‑on, unpredictable or safety‑critical roles in healthcare and blue‑collar trades, showing AI augments, not replaces, work.
Key takeaways
- Generative AI applicability: Microsoft measured an “AI applicability” score to compare Copilot capabilities with O*NET job tasks.
- Low‑risk roles: Hands‑on, physical and unpredictable jobs (healthcare support, trades, heavy‑equipment operators) scored lowest for current generative AI overlap.
- Augment, don’t replace: High applicability signals tasks AI can assist with (drafting, summarizing), not full occupation replacement.
- Policy focus: Findings point to training, apprenticeships and vocational investment to sustain local labor demand.
Key information
Microsoft measured an “AI applicability” score by comparing roughly 200,000 anonymized, privacy‑scrubbed conversations with Bing Copilot against O*NET occupational task lists to assess how much current generative AI aligns with actual job tasks. Sources include Microsoft, Fortune and an arXiv preprint.
A low applicability score means the tasks of a job do not match what generative AI can do today — typically hands‑on, physical, variable or safety‑critical work where in‑person judgment or manual dexterity matters. See the full Microsoft research paper at Microsoft Research.
Important: Microsoft stresses that high applicability is not equivalent to full replacement — AI generally augments tasks such as drafting, summarizing and research rather than performing whole occupations by itself. (See Microsoft and Fortune.)
How Microsoft measured risk
Researchers mapped the capabilities surfaced in ~200,000 anonymized Copilot chats to O*NET task inventories and computed an AI applicability score for each occupation. Low scores indicate limited overlap; high scores indicate strong overlap. Methods and data are published in an arXiv preprint and explained on the Microsoft Research blog post.
Top 10 occupations least affected by generative AI
Microsoft’s list highlights roles that depend on on‑site judgment, manual skill, or unpredictable environments. (Source: Fortune summarizing Microsoft research.)
- Dredge operators — operate heavy equipment to clear or deepen waterways; work is site‑specific and physical.
- Bridge and lock tenders — manage waterway bridge and lock systems requiring real‑time, local judgment.
- Water treatment plant and system operators — hands‑on monitoring and adjustments at treatment facilities.
- Foundry mold and coremakers — craft precise molds in metal casting; high manual skill.
- Rail‑track laying and maintenance equipment operators — outdoor, variable‑condition heavy machinery work.
- Pile driver operators — run powerful equipment on construction sites for foundation work.
- Floor sanders and finishers — skilled craftsmanship for unique surfaces.
- Orderlies — direct patient handling and physical care in hospitals and nursing homes.
- Motorboat operators — navigation and seamanship in changing weather and waters.
- Logging equipment operators — heavy machinery work in rugged terrain with quick local decisions.
Ten more roles and categories with low AI applicability
To expand the list to a broader top 20, Microsoft groups low‑score occupations into clusters that require physical presence, manual skill, or human care. (Summary source: Fortune.)
- Healthcare support workers — especially home health aides and personal care aides: tasks require personal contact, empathy and unpredictable responses; also among the fastest growing jobs per BLS projections.
- Construction trades — carpenters, electricians, plumbers: varied on‑site work, tool use, and safety decisions.
- Heavy machinery operators and mechanics — diesel mechanics, excavator operators and similar roles.
- Skilled installers and repairers — HVAC technicians and industrial repair workers.
- Forestry and agricultural equipment operators — outdoor, environment‑dependent tasks.
- Waste collection and material movers — on‑site physical labor with changing conditions.
- Protective service roles with hands‑on duties — some correctional, security, and emergency support roles.
- Specialty manufacturing roles — positions requiring manual dexterity such as coiling, molding, or fine assembly.
- Maritime deckhands and small vessel crews — seamanship and in‑person teamwork.
- Utility line workers and field technicians — live‑wire safety, climbing, and situational judgment.
Microsoft cautions that other forms of AI and automation — for example, machine monitoring AI or autonomous vehicles — could affect some of these roles later; the present study focuses on generative AI’s language and content generation capabilities. See Microsoft and Fortune for context.
What jobs are most at risk by comparison?
Jobs that emphasize communication, routine knowledge work or text generation have the highest AI applicability. Examples include interpreters and translators, writers and authors, customer service and sales representatives, and roles centered on drafting and summarizing information — areas where generative AI currently excels. (See Fortune.)
What this means for workers and employers
Augmentation, not wholesale displacement: Microsoft emphasizes that high applicability signals tasks AI can help with — like drafting emails, summarizing documents, or generating code snippets — rather than a prediction that entire occupations will vanish. See the Microsoft blog.
“Train workers to use AI effectively rather than assume jobs will disappear.”
Training and reskilling: Workers in higher‑applicability jobs will need training to adopt AI tools productively. For lower‑applicability jobs, expanding apprenticeships and vocational programs can bring new workers into stable sectors. (See the arXiv methods and recommendations.)
Local demand: State and local labor projections align with the research, showing continued demand for health aides and trades — key considerations for education and economic planning.
Policy and business choices to consider
For employers and policymakers prioritizing stable work and healthy local economies, actions include:
- Invest in career and technical education: Fund community colleges, apprenticeships and trade schools to funnel workers into low‑applicability jobs.
- Promote on‑the‑job upskilling: Help workers in communication‑heavy roles adopt AI tools productively rather than be displaced.
- Monitor broader automation: Keep watch on machine vision, robotics and vehicle autonomy, which are separate from generative AI but could affect physical jobs over time. (Microsoft discusses these distinctions in its blog post.)
Implications for the United States
Economic impact
The U.S. labor market could see steady demand in healthcare support and skilled trades — sectors Microsoft finds have low generative AI applicability. That demand can help shore up wages and local economies in regions with construction, aging infrastructure and eldercare needs. (Sources: Microsoft Research, Fortune.)
Political consequences
Policymakers from both parties may face pressure to fund vocational training, community colleges and apprenticeship programs; conservative audiences favoring work‑based learning can point to these findings as a case for targeted investment. (See arXiv.)
Social effects
Jobs requiring human contact — orderlies, home health aides, personal care workers — are less exposed to generative AI and provide essential services for an aging population. Protecting and expanding these roles supports families, caregivers and local communities. (See Fortune.)
Cultural relevance
The research reinforces respect for skilled trades and hands‑on work. For readers who emphasize self‑reliance, local jobs and family caregiving, the study highlights practical paths where human labor remains central and valued.
Practical applications for communities
State and local leaders can use these findings to shape economic development: incentives for apprenticeships, partnerships with vocational schools, and support for homecare employment could be prioritized. Businesses can assess which tasks to automate and which to keep human — then invest in training so workers use AI as a tool, not a threat. (Further detail: arXiv, Fortune.)
