概述
Customer service agents, administrative assistants, translators, bookkeepers, paralegals, and some marketing and programming roles are examples. Exposure measures how many of a job's tasks AI could speed up. It does not measure whether the job will disappear, and many exposed jobs are augmented rather than eliminated.
深入探討
Researchers measure exposure by breaking occupations into tasks and asking which tasks a technology could do or speed up. How they do this has changed over time. An influential 2013 study by Carl Frey and Michael Osborne estimated that about 47 percent of US employment was in occupations at high risk of computerization. Critics pointed out that it scored whole occupations, even though jobs are bundles of tasks. Later OECD work that used task-level data found a much smaller share of jobs that could be fully automated. In 2023, Tyna Eloundou, Sam Manning, Pamela Mishkin and Daniel Rock published "GPTs are GPTs." They rated US O*NET tasks using both human annotators and GPT-4. Their estimate was that roughly 80 percent of US workers are in occupations where at least 10 percent of tasks are exposed, and around 19 percent are in occupations where at least half are. In a reversal of earlier automation waves, which mostly hit routine manual work, higher-wage and more educated occupations tended to be more exposed. A 2023 International Labour Organization study found clerical support work to be the most exposed group worldwide and concluded that augmentation was more likely than full automation for most occupations. In 2024 the IMF estimated that about 40 percent of global employment is exposed, with a higher share in advanced economies. Exposure is not the same as replacement. Whether jobs are lost depends on adoption costs, regulation, customer preferences, and how demand responds when output gets cheaper. A frequently cited historical example is that US bank teller employment did not collapse after ATMs spread, because banks opened more branches and tellers took on different work. The most common misconception is that exposure scores predict layoffs. They are estimates of what is technically feasible, not forecasts.
戰略影響
風險與安全
災難性和日常的人工智慧危害都取決於誰了解風險以及誰能夠採取行動。
更明確的決策
民眾和專業素養決定強而有力的安全政策在政治上是否可行。
突破炒作
清晰的解釋可以減少炒作、實驗室公關和模糊道德劇場的影響。
The Future of Which Jobs Are Most Exposed to AI
Exposure estimates will keep changing as models gain new abilities such as tool use, computer control and better reasoning, which may extend exposure into tasks that were rated low before. How much of that turns into job losses is uncertain. Past technologies have usually taken years to spread through organizations, and effects have shown up first in slower hiring for entry-level roles rather than mass layoffs. Researchers are increasingly pairing exposure scores with real usage and labor market data to tell apart what AI could do from what it is actually doing. Watch that evidence rather than headline percentages.
現實世界的實施
A customer support center adds an AI assistant that drafts replies. Agents close more tickets per hour, and the company responds by hiring fewer new agents rather than laying off current staff.
A freelance translator finds that clients increasingly want post-editing of machine translation instead of full translation, often at a lower rate per word.
A paralegal uses AI to summarize thousands of discovery documents. It saves time, but it also cuts the number of junior hours billed on the case.
A software developer uses a coding assistant for boilerplate code and unit tests. The job remains, but more of it shifts to code review, system design and debugging.
風險與防護欄
將存在風險視為科幻小說，同時能力複合。
混淆了表面產品安全與高度自治下的對準。
只給非英語和非專業觀眾留下低品質的資源。
實施路線圖
單獨的產品危害、誤用和失控/失調風險。
詢問哪些證據會改變您對時間表和嚴重性的看法。
比起行銷主張，更喜歡主要來源和具體評估。
確定一條行動路徑：職業、政策、資金或技能——而不僅僅是意識。
不斷探索
Free newsletter
Get the daily AI briefing
Three verified AI stories every weekday morning, written in plain English. Free forever, no ads.
One email each weekday. Unsubscribe in one click. We never sell or share your address.
Test yourself
Take the Which Jobs Are Most Exposed to AI quiz
Instant feedback on every answer, and a shareable certificate with a verifiable ID once you pass a course.
Support free AI education. AI Understanding is a 501(c)(3) nonprofit — no ads, no paywall, ever. Make a donation
常見問題
What is Which Jobs Are Most Exposed to AI?
The jobs most exposed to AI are those where a large share of the work is reading, writing, analyzing or producing information on a computer. Customer service agents, administrative assistants, translators, bookkeepers, paralegals, and some marketing and programming roles are examples. Exposure measures how many of a job's tasks AI could speed up. It does not measure whether the job will disappear, and many exposed jobs are augmented rather than eliminated.
What does an AI exposure score actually measure?
Exposure is about technical feasibility at the task level. It does not predict layoffs, which depend on costs, regulation, demand and how organizations respond.
What was the main criticism of the 2013 Frey and Osborne estimate?
Critics argued that treating an occupation as a single unit overstated risk. Task-level analyses, such as the OECD's follow-up work, found far fewer fully automatable jobs.
In the "GPTs are GPTs" study, which workers tended to be more exposed than in earlier automation waves?
Earlier automation mostly hit routine manual and clerical work. The 2023 study found that generative AI exposure was often higher for well-paid knowledge work.
According to the 2023 ILO study, which occupation group was most exposed worldwide?
The ILO identified clerical support as the most exposed group and concluded that augmentation was more likely than full automation for most occupations.
Why does the guide mention bank tellers and ATMs?
Teller employment did not collapse after ATMs because banks opened more branches and tellers' work shifted. Exposure alone did not determine the job outcome.
繼續學習
相關指南
為此主題精選的更多指南