Microsoft study names 20 jobs least impacted by AI — healthcare and manual labor top the list
A 2024 Microsoft study analyzing 200,000 anonymized Bing Copilot conversations identifies occupations least likely to be affected by generative AI, finding healthcare and hands-on manual roles show the lowest AI applicability scores and minimal current overlap.
Key takeaways
- Research basis: Microsoft mapped 200,000 anonymized Bing Copilot conversations to U.S. job activities to compute AI applicability for ~1,000 occupations (Microsoft research).
- Low applicability: Jobs needing physical presence, manual dexterity, or direct care—especially healthcare and blue-collar trades—score lowest and show limited current overlap with generative AI.
- Not a prediction: The score measures present task overlap, not future displacement; AI could augment roles or shift tasks rather than simply replace workers (Microsoft blog).
How the study worked
Microsoft researchers analyzed anonymized transcripts from 200,000 user conversations with Bing Copilot and linked those exchanges to the U.S. Department of Labor’s list of job activities. From that mapping they built an AI applicability score ranging from 0 to 1. Scores near zero indicate little overlap between what generative AI can do today and the tasks those jobs require. The study covered about 1,000 U.S. occupations and aimed to show which jobs are more or less likely to receive immediate support from current AI tools (full paper; Study).
What the numbers show
Clear patterns emerged: knowledge and office jobs present the highest AI applicability scores because they rely on communication, writing, and data work—areas where generative models already assist. By contrast, jobs requiring physical manipulation, complex motor skills, or direct human care registered the lowest scores. The authors emphasize that applicability signals where current generative AI can assist or automate tasks today, not where layoffs are certain.
“AI applicability measures present overlap between tasks and AI outputs; it is not a direct forecast of future job displacement.”
Twenty jobs least likely to be impacted by AI
Microsoft published partial lists highlighting occupations with the lowest AI applicability scores. Across reporting sources, these 20 jobs appear most consistently as least impacted by generative AI (listed with original phrasing, no specific rank except where noted):
- Roofers
- Gas compressor and pumping station operators
- Helpers—roofers
- Tire builders
- Surgical assistants
- Massage therapists
- Ophthalmic medical technicians
- Industrial truck and tractor operators
- Supervisors of firefighters
- Cement masons and concrete finishers
- Dishwashers
- Machine feeders and offbearers
- Packaging and filling machine operators
- Medical equipment preparers
- (Additional low-risk roles include other hands-on trade and healthcare jobs noted in the full Microsoft list.)
Sources noted: GeekWire, Fortune, and Microsoft research.
Why healthcare and manual labor score low
Healthcare roles such as surgical assistants, ophthalmic technicians, and medical equipment preparers require hands-on patient care, precise manual skills, and split-second judgments in physical settings—tasks that current generative AI cannot perform directly. Likewise, blue-collar trades (roofers, cement masons, machine feeders) involve tools, strength, unpredictable environments, and sensory decisions that limit immediate AI assistance. The study highlights this minimal overlap with existing model capabilities (Microsoft research; GeekWire).
Which jobs are most at risk now
In contrast, occupations heavy on written communication, data analysis, and routine office tasks show the highest AI applicability. Roles such as interpreters, historians in certain contexts, clerical staff, teachers, journalists, and office administrators could see substantial AI support in drafting, summarizing, and research—areas where generative models already add value (Fortune; GeekWire).
Caveats and context
Microsoft and outside reporters urge caution: the AI applicability score measures current overlap between tasks and AI outputs, not future automation. History shows adaptation—jobs often evolve rather than vanish. For example, bank tellers shifted toward advisory roles after ATMs. The study notes AI may augment duties, spawn new tasks, or reshape job content over time (Microsoft blog).
Policy and training implications
Where AI applicability is high, education and employers may prioritize skills that complement AI—supervision, critical judgment, and interpersonal capabilities. Where applicability is low, investment in apprenticeships, vocational training, and healthcare education remains essential. The mapping can guide local workforce planning and funding toward occupations likely to remain hands-on.
Implications for United States
Economic: Rural and small-town economies reliant on manual labor and local healthcare may face less immediate AI disruption. Many low-applicability jobs are locally rooted and can sustain family incomes where remote office work is scarce (Microsoft research; GeekWire).
Political: Lawmakers in farm states and rural districts can cite these findings to support apprenticeships, trade-school funding, and healthcare workforce programs—policies that emphasize durable, local jobs over retraining into high-exposure remote knowledge roles (Fortune).
Social and cultural: Regions dominated by hands-on work may experience slower employment shifts, preserving tax bases and local services. Public messaging that values skilled trades and bedside care can align economic policy with community identity (GeekWire).
Practical applications
County officials and school boards can use the study to inform curriculum choices—expand vocational programs, strengthen employer partnerships, and fund certified apprenticeships. Rural healthcare providers might prioritize training for roles such as surgical assistants, ophthalmic technicians, and medical equipment preparers who are likely to remain essential to local care systems (Microsoft study).
