Is the claim that "AI is taking away jobs" becoming reality?
A review of domestic and international research shows it is premature to conclude that generative artificial intelligence is simply "replacing" workers in the labor market. However, studies consistently detect early warning signs among young workers, particularly those just entering the workforce.
South Korea: Youth Employment Declining in Accounting, Customer Service, Media
According to a report titled "Generative AI High-Exposure Occupations and Recent Youth Employment" released by the National Assembly Budget Office on the 26th, declining youth employment was observed in certain jobs with high exposure to generative AI.
The report classified "AI high-exposure occupations" based on the International Labour Organization's occupational AI exposure index, including accounting and bookkeeping clerks, finance and insurance clerks and specialists, customer service and monitoring agents, computer and software developers, and writers and media-related professionals.
Examining employment changes from ChatGPT's launch in November 2022 through the first half of last year, youth employment declined in accounting and bookkeeping, customer service and monitoring, and writing and media-related professions.
In contrast, computer systems and software specialists, finance and insurance specialists, and human resources and management specialists showed clear increases. This indicates that AI-related fields did not uniformly experience job losses.
More notably, regression analysis found no statistically significant causal relationship between generative AI proliferation and reduced employment in high-exposure occupations.
In fact, following ChatGPT's release, employment growth rates for young workers aged 15-34 were 1.2 percentage points higher in AI high-exposure occupations than in other jobs. Workers aged 35-49 showed rates 0.66 percentage points higher, while those 50 and older were 1.4 percentage points higher. This makes it difficult to conclude that "AI reduced employment."
However, detailed hiring data revealed youth recruitment declines in specific categories including telemarketers and accounting clerks. Conversely, youth hiring increased for finance and insurance clerks and specialists, translators and interpreters, and writers.
The report concluded that "no clear evidence was found that generative AI proliferation reduced employment in high-exposure occupations," while noting that additional research on long-term effects is needed.
U.S. Shows Different Pattern: 13% Drop in Employment for Ages 22-25
More direct changes were observed in the United States. Research published in September last year by Stanford University's Digital Economy Lab found that employment of entry-level workers aged 22-25 fell 13% in occupations with high AI exposure.
Software developers aged 22-25 saw employment drop approximately 20% from the late 2022 peak through July 2025. Similar patterns were confirmed in customer service positions.
Notably, results varied by age within the same occupations. Employment of experienced workers aged 35-49 actually increased, revealing a phenomenon of "fewer entry-level hires, more experienced hires."
Rapid improvements in AI performance are cited as a contributing factor. According to the AI Index Report, the proportion of coding problems AI can solve on SWE-Bench, a software engineering benchmark, surged from 4.4% in 2023 to 71.7% in 2024. This indicates AI has reached a level where it can handle a substantial portion of routine coding tasks.
Researchers also noted that how AI is used affects employment differently. Occupations using AI as an "automation" tool to replace human tasks saw entry-level hiring decline. Conversely, occupations using AI as a "complementary" tool for learning, verification, and collaboration tended to see employment increase.
Even after controlling for company characteristics and economic factors, workers aged 22-25 in high AI-exposure occupations showed relatively greater employment declines compared to those in low-exposure occupations. No significant decreases were observed in other age groups.