了解人工智能对就业和行业转型的影响

© Oliver WymanIn an intriguing twist of historical irony, the product that introduced generative AI to theworld is aptly named ChatGPT. Within this context, GPT stands for “generative pre-trainedtransformer,” a term for a class of large language models (LLMs). However, in the annals ofscience and technology, “GPT” has long held another meaning: “general purpose technology.”These are innovations with applications so sweeping thatthey affect entire economies orcivilizations. Examples of GPTs in this sense include the wheel, the printing press, the internalcombustion engine, and the internet. The question now arises: will generative AI becomeanother GPT, with similarly far-reaching implications and impacts? If so, how will it affecttoday’s industries, and what can business leaders do to anticipate its impacts?The impact of AI on various jobs and industries has been a topic of much discussion.Business leaders face hard decisions on how much to invest in deploying AI, where to focusthe efforts, and how to manage the risks. For private equity investors and other financialsponsors, understanding how their portfolio is exposed toAI’s risk and opportunitiesiscrucial so the right investment decisions can be made and the right guidance given toportfolio companies.To simplify this complex topic, we believe two dimensions are worth considering: first, thedegree to which AI is likely to impact a particular industry, and second, the extent to whichtotal demand in an industry is likely to expand given greater productivity (and thereforelower costs) due to the deployment of AI.HOW AI WILL IMPACT AN INDUSTRYThe first dimension has been a topic of extensive academic research, examining thenature of jobs in an industry and their substitutability with AI. While many recenttechnological innovations have displaced lower earners or more blue-collar jobs — thinkself-order kiosks at fast food restaurants — generative AI’s impact is expected to begreatest on middle-skilled, white-collar jobs. Nor will the effects of AI only be felt in thejobs themselves but in management structures as well: fewer middle managers will beneeded if (for example) 10 AI-turbocharged programmers can soon do the work of 50today. Some jobs have more “exposure” to the potential impacts of AI than others, andthis can be extrapolated to the industry level based on the job mix in a given industry.To quantify these differences, we have leveraged theAI Industry Exposure (AIIE) indexcreated by Edward Felten, Manav Raj, and Robert Seamans in their academic research firstin 2021 and updated to reflect the impact of LLMs specifically in 2023. Their work foundthat janitorial services, meat processing, and coal mining (to take three examples) shouldsee less impact from AI, while accounting, commercial banking, and legal services will seefar more. © Oliver WymanBut we also need to consider the second-order impacts on industries of “AI shock.” How longwill enhanced productivity boost margins before being competed away? And what industrieswill grow most as a result of higher productivity and, therefore, lower costs? Mechanicalweaving, for instance, did not just make clothes cheaper; it allowed demand for them toexplode. Instead of owning just one set of work clothes and a “Sunday best,” a typical workercould afford to have a different outfit each day, leading to today’s closets full of clothes thatmay not even be remembered or worn. © Oliver WymanAI-DRIVEN PRODUCTIVITY GAINS AND DEMAND EXPANSIONTo estimate how “elastic” an industry is to growth from a positive productivity shock,Oliver Wyman examined the correlation between labor productivity and real output for eachindustry in the NAICS hierarchy from 1987 through to 2019 (to avoid COVID noise). Rankedin descending order from highest to lowest correlation, a higher score on the y-axis can beinterpreted as more potential for industry growth in total output due to higher productivity.By plotting industries according to both their expected AI impact and their historicalrelationship with labor productivity, (figure below), valuable insights can be gleaned intowhat might be expected in different industries over the comingdecades.Exhibit 2: Industry-level potential impact of AI on productivity and total outputJanitorial servicesPoultry processingSawmills andwood preservationCoal miningNewspaper, periodical,book, and directorypublishersApparel accessories andother apparel manufacturingVeterinary servicesLegal servicesOther accountingservicesCommercial bankingTax preparation servicesFiber, yam, andthread mills050100150200250300350400450500550−2.0−1.5−1.0−0.50.00.51.01.52.02.5Productivity and Output Correlation (Rank)AI Industry Exposure (AIIE)Source: “Occupational Heterogeneity in exposure to Generative AI” by Felten, Raj, Seamans, Bureau of Labor Statistics,Oliver Wyman analysis © Oliver WymanIndustries on the far left, ranging from dry cleaning to janitorial services, will experiencelimited direct AI impacts. For these busin