Arguments with opposite conclusions came out one after another in the same week. David George, a general partner at the venture capital firm a16z, wrote in an essay that the prediction that AI will drive out white-collar workers and create a permanently unemployed class is bad economics and fiction. On the other side, diagnoses continued that employment indicators for young American college graduates are clearly worsening. The two arguments are not looking at the same thing.
What George took aim at is the "lump of labor fallacy." The term refers to the premise that the total amount of work the world needs is fixed, so that when a machine takes one job, people lose one, and he considered that premise wrong from the start. He offered as evidence the case of agricultural mechanization in the early 20th century, which wiped out a third of U.S. employment while that workforce moved into factories and offices. Torsten Slok, chief economist at Apollo Global Management, also came down on the side of employment growth, citing the Jevons paradox, the idea that as a technology gets cheaper, demand for it rises instead.
Aggregate statistics favor this side. Studies by the NBER and the Federal Reserve banks have found no meaningful change in total employment figures since AI was adopted. The Carnegie Endowment for International Peace sorted participants in the AI labor debate into the alarmed, the patient and the enthusiastic, and placed a16z among the enthusiastic. The stake involved, optimism from the side that profits from AI investment, is also noted.
But once the aggregate is stripped away and the numbers are broken down by age, the picture changes. A Stanford University study found that employment in "AI-exposed occupations," including software development, fell 13%. Employment among people aged 22-25 declined 6%, while employment among older workers rose. It means that even when the same technology enters the same company, the shock lands on a particular age group.
The unemployment rate for U.S. college graduates aged 22-27 rose from 4.1% in 2022 to 5.6% this February. By the New York Federal Reserve's measure, the underemployment rate for college graduates is around 43%. In a survey by the freelance platform Upwork, 64% of corporate executives said they plan to cut or halt entry-level hiring because of AI. Britain's IPPR institute found that the first shock of AI's spread is landing on entry-level office jobs, where the shares of women, young people and low-wage workers are high.
The scene on the job market has changed as well. Bunmi Omisore, 22, a Duke University graduate, said she plants key words in her applications using ChatGPT to get past AI document filters. Eamon Morton, 23, landed at YouTube after sending out 453 internship applications.
The share of U.S. entry-level job postings that require AI skills has roughly doubled in recent years. Whether the cause of this change is AI, or the combined result of other factors such as layoffs at technology firms during the rate-hike period, is hard to separate out.
There is also a view that treats replacement and redesign separately. A BCG report, "AI redesigns jobs rather than replacing them," put jobs that disappear entirely at 10-15% and jobs likely to have their task composition rearranged at more than half. The countries covered by the estimate and the forecast year lie outside what was made public.
Hwang Hyung-jun, head of BCG Korea, ran a column carrying this argument in that day's paper. Warnings from economists including Anton Korinek that labor itself becomes optional once AGI arrives are another branch of the same debate.
Both sides agree that AI lowers the barrier to entry for coding and analysis. Only the reading is opposite. One side reads it as anyone being able to become a developer, the other reads the lowered barrier as having erased the work that new hires once learned on and got by with. What has actually shrunk is, more than the total amount of work, the time in which beginners could build skills while working clumsily.
Confirming the same picture with Korean data would require putting the youth unemployment rate and the number of entry-level postings side by side, and that work has not been done. On the 29th of this month, an academic conference on labor after technology will be held at Dongguk University's Seoul campus. Lee Kwang-seok, a professor at Seoul National University of Science and Technology, Hong Nam-hee, a professor at Yonsei University, and Park So-hee, a labor standards inspector at the Ministry of Employment and Labor, will speak.
The question of whether technology replaces people is too big. What actually needs an answer is whether a 23-year-old still has the chance to learn by doing three years of clumsy work at a first job. The 453 applications are a record of how far the price of that chance has climbed.
