Microsoft AI study: 20 jobs least likely to be changed by generative AI — healthcare and blue‑collar work top the list
Lede: A Microsoft Research analysis of over 200,000 anonymized Bing Copilot chats finds hands‑on healthcare and many blue‑collar roles are least exposed to generative AI, while communication‑heavy knowledge work shows higher applicability across sectors today and globally.
Key takeaways
- Microsoft built an AI applicability score from 200,000+ privacy‑scrubbed Copilot chats to estimate where LLMs can support work activities (Microsoft Research).
- Lowest applicability includes hands‑on healthcare, repair and blue‑collar trades that require physical presence, dexterity, or immediate safety judgment.
- Not a displacement forecast: applicability shows where AI can help with tasks, not whether jobs will be eliminated (research blog, ArXiv preprint).
Main
What Microsoft measured — and what it didn’t
Microsoft’s paper and blog post emphasize the distinction between applicability — how useful generative AI is for parts of a job based on real Copilot chats — and job displacement. The score maps Copilot use to O*NET work activities and highlights where LLMs can support tasks; it does not predict layoffs or capture robotics and computer‑vision impacts.
The 20 jobs least likely to be impacted by AI
Below are occupations at the bottom of Microsoft’s AI applicability rankings, with brief reasons why each showed low applicability in the study. Sources include GeekWire reporting, the Microsoft Research paper, and Fortune.
- Roofers — work high on site, use physical tools and local judgment.
- Helpers — Roofers — assist with materials and on‑site setup; tasks are manual.
- Tire Builders — operate and adjust manufacturing equipment; hands‑on assembly.
- Gas Compressor and Pumping Station Operators — monitor heavy equipment onsite.
- Industrial Truck and Tractor Operators — drive forklifts and move heavy loads safely.
- Cement Masons and Concrete Finishers — perform manual finishing work on construction sites.
- Supervisors of Firefighters — lead emergency response and on‑scene safety decisions.
- Dishwashers — manual, back‑of‑house tasks in food service settings.
- Machine Feeders and Offbearers — load and unload industrial machinery.
- Packaging and Filling Machine Operators and Tenders — run and tend physical production lines.
- Medical Equipment Preparers — clean, sterilize, and prepare instruments for procedures.
- Surgical Assistants — assist surgeons in the operating room with hands‑on tasks.
- Massage Therapists — provide in‑person therapeutic touch and client care.
- Ophthalmic Medical Technicians — perform eye tests and assist patients with equipment in clinic.
- Home Health and Personal Care Aides — provide daily, in‑home physical support to patients.
- Nursing Assistants and Orderlies — basic, hands‑on patient care in clinical settings.
- Construction Laborers — carry out varied outdoor manual tasks.
- Installation, Maintenance, and Repair Workers (general) — fix physical systems on site.
- Food Preparation and Serving Workers (basic roles) — perform hands‑on kitchen tasks.
- Janitors and Cleaners — on‑site cleaning and basic facility maintenance.
Note: Items 1–14 were explicitly named in Microsoft’s “least impacted” list reported by GeekWire; items 15–20 reflect occupations Microsoft and later reporting describe as low AI exposure, especially in healthcare support and manual work (Fortune).
Why healthcare support and blue‑collar jobs scored low
Microsoft found shared traits among low‑applicability occupations: physical presence, manual dexterity, close personal interaction, and on‑the‑spot situational judgment. Jobs requiring touch, fine motor control, patient lifting, or immediate safety decisions map poorly to current LLM capabilities.
“Lower applicability today does not guarantee permanent safety.” — Microsoft caveat about robotics, sensing, and other advances
How Microsoft built the score
Researchers mapped anonymized Copilot conversations to O*NET work activities, then computed an applicability score estimating how much AI could support tasks in each occupation. The input was practical — based on how people actually used Copilot in 2024 — making the study a window into current AI utility, not a forecast of economic change (Microsoft Research, research blog).
Implications for Los Angeles, California
Los Angeles’ large health‑care, construction, hospitality and entertainment sectors mean the study’s findings carry local weight:
- Jobs and industries: Many frontline healthcare support roles (nursing assistants, home health aides, medical equipment preparers) and blue‑collar jobs (roofers, construction laborers, machine operators, janitors) are less likely to see immediate task support from generative AI (GeekWire, Fortune).
- Health care pressure and growth: LA’s aging population and sprawling health system suggest continued demand for in‑person care; low AI applicability supports ongoing hiring while workers still need training for record keeping and hybrid tech tools (Microsoft Research).
- Construction and housing: Trades insulated today matter for planning and unions; apprenticeship and safety programs remain key if tasks stay hands‑on (GeekWire).
- Service and hospitality: Dishwashers, food prep staff, and janitors are likely to remain essential, though employers may adopt AI for scheduling, payroll, or training (Fortune).
- Risks for communication roles: PR, reporting, marketing, and scriptwriting show high AI applicability and may see increased use of generative tools for drafting, research, and translation (Microsoft Research).
- Policy and workforce response: Los Angeles leaders should pair support for in‑person care and trade apprenticeships with upskilling programs teaching how to use AI tools in office and creative settings (Microsoft WorkLab).
Sources and further reading
- Microsoft Research — “Working with AI: Measuring the occupational implications of generative AI”
- Microsoft Research blog — Applicability vs. job displacement
- GeekWire reporting on Microsoft study
- Fortune coverage of occupational impacts
- Microsoft WorkLab — AI at work
- ArXiv preprint (study data and methods)
