TLDR: A new study by Microsoft Research, leveraging data from 200,000 anonymized Copilot user conversations, identifies occupations most susceptible to generative AI’s influence and those least likely to be affected. The research highlights a significant shift in the nature of work, emphasizing augmentation and the emergence of hybrid roles rather than widespread job displacement.
The advent of generative artificial intelligence (GenAI) has sparked widespread discussion about its potential impact on the global job market. Addressing this critical question, a recent study from Microsoft Research, led by Kiran Tomlinson, provides a data-driven perspective on which occupations are most and least vulnerable to the capabilities of GenAI. The findings are detailed in their paper, ‘Working with AI: Measuring the Occupational Implications of Generative AI,’ published on August 13, 2025.
The research methodology involved a comprehensive analysis of 200,000 anonymized and privacy-scrubbed conversations between users and Microsoft Bing Copilot, a publicly available generative AI system. This dataset was collected between January 1, 2024, and September 30, 2024. The team utilized a GPT-4o-based large language model (LLM) classification pipeline to identify intermediate work activities (IWAs) within these conversations, aligning them with the U.S. Department of Labor’s O*NET database, which standardizes nearly 1,000 U.S. occupations. Each occupation was then assigned an ‘AI applicability score’ to measure the potential impact.
The study reveals that knowledge and communication-focused roles exhibit the highest susceptibility to generative AI. Occupations such as interpreters and translators, historians, passenger attendants, sales representatives of services, writers and authors, customer service representatives, and news analysts, reporters, and journalists ranked high on the AI applicability scale. Other roles like CNC tool programmers, telephone operators, ticket agents, travel clerks, brokerage clerks, political scientists, mathematicians, technical writers, proofreaders, and editors also showed significant overlap with AI capabilities. These findings suggest that tasks involving information processing, content generation, and routine digital communication are prime candidates for AI augmentation.
Conversely, jobs requiring significant physical dexterity, real-world judgment, or direct human interaction and empathy were found to be least affected by current GenAI capabilities. The study identified 40 occupations largely beyond AI’s reach. These include dredge operators, bridge and lock tenders, water treatment plant and system operators, foundry mold and coremakers, rail-track laying and maintenance equipment operators, pile driver operators, floor sanders and finishers, orderlies, motorboat operators, and logging equipment operators. Other resilient professions include phlebotomists, nursing assistants, hazardous materials removal workers, embalmers, oral and maxillofacial surgeons, automotive glass installers, ship engineers, tire repairers, prosthodontists, highway maintenance workers, medical equipment preparers, massage therapists, surgical assistants, and roofers. These roles often demand physical presence, complex problem-solving in dynamic environments, or a high degree of human connection that AI cannot replicate.
Also Read:
- Generative AI Poses Unprecedented Threat to White-Collar Employment Amid Economic Downturns, IndexBox Analysis Reveals
- New Zealand Poised to Adapt as AI Redefines ‘Low-Value’ Work Across Global Industries
While the research highlights significant AI applicability in certain sectors, Microsoft and independent analysts caution that this does not necessarily equate to widespread job loss. Instead, the prevailing outlook suggests that generative AI tools will primarily augment workflows, leading to the creation of hybrid roles where human judgment and oversight remain crucial. This is particularly anticipated in sectors like financial services, healthcare, and creative industries. The study underscores that the long-term impact on employment remains uncertain, with outcomes largely dependent on how businesses and workers adapt to evolving tools and workflows. As the paper’s abstract states, ‘understanding the effects of AI on the economy is one of society’s most important questions,’ emphasizing the ongoing need for real-world usage trend analysis to inform policy and workforce planning.
