AI’s experience bias shocks labor market, hitting unemployed youth first

Home > Opinion > Columns

AI’s experience bias shocks labor market, hitting unemployed youth first

Published: 10 Mar. 2026, 00:03

Audio report: written by reporters, read by AI

Kim Hyun-cheol

The author is a professor at Yonsei University College of Medicine and director of the Institute for Population and Talent.

The atmosphere in research laboratories has changed. Analyses that once required a week of work from graduate-level researchers can now be completed by artificial intelligence in a few hours. Yet I have not reduced the number of assistants in my lab. Instead, I use AI to conduct more research.

Workers in fields where AI is widely used often feel a sense of crisis, fearing their jobs may soon be replaced. By contrast, those in sectors less affected by AI often believe their roles are safe. Economists have begun addressing this issue through more than 20 empirical studies published between 2025 and 2026.

Young job seekers attend a job fair held at Busan City Hall in Buan on Nov. 10, 2025. [NEWS1]

So far, overall labor market indicators appear relatively stable. One study tracked 25,000 Danish workers for three years and found no significant changes in income or working hours (Humlum & Vestergaard, 2025). The AI company Anthropic also measured “AI exposure” by combining usage data from its chatbot Claude with U.S. labor statistics (Massenkoff & McCrory, 2026). Unemployment rates in highly exposed occupations were no different from those in less exposed sectors.

There are three main reasons for this stability. First, many firms have adopted AI but use it only modestly. A survey of 6,000 executives across four countries found that although 70 percent of firms had introduced AI, managers used it only about one and a half hours per week on average (Yotzov et al., 2026).

Second, gains in individual productivity do not immediately increase overall output. After companies adopted AI coding tools, individual productivity rose by about 8 to 9 percent, yet there was no measurable change in total output or employment (Chen & Stratton, 2026).

Third, a wide gap remains between theoretical capability and real-world application. In theory, AI could perform up to 94 percent of tasks in computer and mathematics occupations. In practice, however, it currently covers only about 33 percent of those tasks (Massenkoff & McCrory, 2026).

The AI revolution, therefore, needs time before productivity gains appear in economic statistics. That transitional gap is the moment we are experiencing. Still, there is reason for concern because the long-term possibility of substitution remains high. Recent data already show early warning signs in youth employment.

In software development jobs, employment among workers aged 22 to 25 has declined by about 20 percent. Employment among workers aged 30 or older in the same occupation grew by 6 to 13 percent during the same period (Brynjolfsson et al., 2025).

Anthropic’s recent research shows a similar pattern. Among workers aged 22 to 25, the rate of new employment in AI-exposed occupations declined by 14 percent, while no comparable pattern appeared among workers older than 25 (Massenkoff & McCrory, 2026).

Korea shows a similar trend. Over the past three years, jobs held by young workers declined by about 211,000. Of that decline, 98.6 percent occurred in industries with high exposure to AI. In contrast, employment among workers in their 50s increased by 209,000 in the same sectors, according to the Bank of Korea.

AI-driven labor market change is occurring mainly through reduced hiring rather than layoffs. Firms find it difficult to dismiss existing workers. Instead, they slow or suspend recruitment. As a result, the impact falls largely on young people who have not yet entered the work force.

Wage patterns show a similar divide. After AI adoption, the number of junior workers declined by 7 to 12 percent while the number of senior workers increased by about 6 percent (Hosseini & Lichtinger, 2025). Starting salaries for junior employees fell by 6.3 percent while senior wages rose slightly (Azar et al., 2025). This form of technological change that favors experience is unusual.

How AI is used also matters. When AI replaces call center agents, employment falls. When it assists them, employment can remain stable. In occupations where AI mainly automates tasks, youth employment declines, while in jobs where AI augments workers, employment grows by 8 to 12 percent (Brynjolfsson et al., 2025). The intensity of unemployment, therefore, depends on how the technology is applied.

Economists including Daron Acemoglu have raised similar concerns this year. Current AI development is heavily oriented toward automation because corporate incentives emphasize labor cost reduction. These scholars argue AI should evolve toward “pro-worker AI,” which enhances human expertise and creates new tasks.

Counseling desks for middle-aged job seekers are empty at an employment and welfare center in Mapo District, western Seoul, on Feb. 11. [YONHAP]

What should governments prepare for? First, labor markets must be monitored systematically in real time. Researchers at Anthropic detected early signals because they could link multiple labor statistics with AI usage data. Korea still lacks such an integrated infrastructure. Employment insurance data, job postings and business surveys should be combined to track AI exposure and employment changes each quarter.

Second, young people should not face this transition alone. Governments could offer tax incentives to firms that maintain entry-level hiring or expand internships and apprenticeships. Studies show youth employment rises in occupations using augmentation-based AI (Brynjolfsson et al., 2025). If young workers lose opportunities to learn on the job today, the industrial work force may face a missing middle layer a decade from now.

Third, the gains from AI should be shared more broadly. One projection estimates that AI adoption could increase corporate productivity by 1.4 percent while reducing employment by 0.7 percent (Yotzov et al., 2026). If companies capture all the benefits while society bears the costs, inequality will deepen. Measures such as a robot tax, stronger corporate taxation or digital taxes could help distribute the gains.

Fourth, governments should prepare fiscal scenarios in advance. Policymakers must consider how income support systems would function under unemployment rates of 10 percent, 30 percent or even 50 percent, as well as under demographic changes such as higher fertility or life expectancy exceeding 90 years.

Technological change should be encouraged because countries that lead in AI are likely to create greater economic value. Yet if the benefits concentrate among a few while the costs are shared by everyone, the outcome will not be just. Societies must design institutions now so that communities can absorb the coming shock together.

This article was originally written in Korean and translated by a bilingual reporter with the help of generative AI tools. It was then edited by a native English-speaking editor. All AI-assisted translations are reviewed and refined by our newsroom.