人工智能会取代你的工作吗？或许未来十年不会

Mark A. Wynne and Lillian DerrJune 03, 2025 Recent rapid improvements in the capabilities of artificial intelligence (AI) have raised concerns about thesetechnologies’ impact on employment, specifically, the rollout of generative AI models such as ChatGPT that cancreate new work product from existing inputs. Unease about new technologies displacing workers isnot new.It can be traced back at least to the earliest days ofautomation during the Industrial Revolution. Technologies such as the steam engine and the dynamo inspiredsimilar fears in their day, as did information technology when computers were first introduced. But the jobs AI is expected to touch in the years to come are different from those impacted by previous waves oftechnological change. The ultimate effects of AI on the workforce will depend on the extent to which AI augments (or complements) ratherthan automates (or substitutes for) workers’ tasks. Will this new technology aid workers or replace them? To understand AI’s possible occupational implications, we explore the workforce effects of the last technologicaladvance thought to put jobs at risk: computerization. We look at the impact on occupations believed vulnerable tocomputerization 10 years ago and what that analysis may say about job vulnerability to generative AI today. First, computers loomed over occupations Before the advent of AI, the main workforce concern was whether computerization would displace jobs. A widelycited University of Oxfordstudyby Carl Benedikt Frey and Michael Osborne, first circulated in 2013, examined thesusceptibility of different occupations to computerization. It ultimately determined that 47 percent of total U.S.employment was at risk in 2013. The authors assigned a probability of computerization to each of 702 occupations. Jobs deemedmostat riskincluded telemarketers, title examiners, sewer workers, mathematical technicians, insurance underwriters, watchrepairers, cargo and freight agents, tax preparers, photographic process workers, new account clerks, librarytechnicians and data entry keyers, all with a 99 percent likelihood of computerization. Among those deemedleastat risk were recreational therapists, occupational therapists, health care and socialworkers and choreographers. All had a less than a half-percent likelihood of computers taking over. Move over, here comes AI A decade later, with the November 2022 release of ChatGPT and subsequent rapid advance of generative AI sincethen (DeepSeek, Stargate, Manus, for example), AI has overtaken computers as a leading risk to long-standingoccupations. A recentstudyby Edward W. Felten, Manav Raj and Robert Seamans attempts to identify occupations mostvulnerable to automation from generative AI from both the perspective of language modeling (learning language andcreating sentences after analyzing huge amounts of existing text) and image generation (creating images based onassimilation of large data sets) in a similar fashion to the earlier work by Frey and Osborne. Both studies identifyoccupations by the same occupation codes, making these two measures easily comparable. The AI study authors created two scales for automation risk: from language modeling and image generating AI,labeling hundreds of occupations accordingly. The jobs identified as most susceptible to automation by languagemodeling AI include telemarketers, many types of teachers, sociologists, political scientists and arbitrators. The jobs identified as most susceptible to automation by image generating AI include interior designers, architects,chemical engineers, art directors, astronomers and mechanical drafters. Computerization, AI threaten different jobs Comparing the occupations thought to be most likely to be computerized a decade ago with those most likely to beautomated by AI today, the lists vastly differ (apart from telemarketers) (Chart 1). Downloadable chartChart data The risks to specific occupations from earlier computerization and AI language modeling appear little correlated.The results for image-generating AI are similar. Today’s at-risk occupations aren’t the same ones threatened earlier. Bureau of Labor Statistics offers 10-year employment projections Every year, the Bureau of Labor Statistics (BLS) issues 10-year projections of employment in detailed occupations.The most recent set of projections was published in August 2024 covering 2023 to 2033. The BLS assumes that“labor productivity and technological progress will be in line with the historical experience” but recognizes thatrecent AI advances could cause the future to differ greatly from the past. These projections have historicallybeenfairly accurate. We looked at previous projections to better understand when they proved more or less accurate amid majortechnological change. We asked whether there is any evidence that the BLS’ misses by occupation were somehowrelated to the susceptibility of that occupation to computerization, as estimated by Frey and