How many people go to work when you pour in a billion won?
Every time the government announces it will pour trillions of won into the AI industry, one phrase is never missing from the press release: "job creation." But is that really true? To answer the question, we first have to dust off an old metric.
The employment multiplier
It measures the number of wage earners created, directly and indirectly, across an industry and its related sectors when one billion won of final demand arises in that industry. Calculated by the Bank of Korea on the basis of its input-output tables, it is the most intuitive yardstick we have for how many livelihoods a given industry actually supports.
As of 2023, the employment multiplier for Korean industry as a whole stood at 6.2. In other words, every billion won of new demand drags 6.2 jobs along behind it.
The trouble is that this number varies wildly from one industry to the next.
Semiconductors: 2 jobs per billion. Education: 13.
Between services and manufacturing lies a gulf that cannot be bridged. Labor-intensive service industries—education, health and social care, food and lodging—have employment multipliers well above 10. Construction is no different. In industries that need human hands, money arriving quickly turns into jobs.
Advanced manufacturing such as semiconductors and electronic components is another story entirely. Capital-intensive and highly automated, these sectors yield only two or three jobs even when a billion won is poured in. Software and IT services, too, absorb far less labor than traditional services, because they are built to wring high productivity out of a small headcount.
The AI industry pushes this tendency to an extreme. It is a structure that generates hundreds of billions of won in revenue from a handful of GPU clusters and a small team of engineers. The ratio of revenue to employees at an AI company is barely even comparable to that of a traditional industry. According to figures the Bank of Korea released in December 2025, Korea's specialized AI workforce was estimated at roughly 57,000 people as of 2024—a strikingly small number relative to the scale of an industry attracting trillions of won in investment.
It is already happening
Let's look at reality rather than theory. A report published by the Bank of Korea's employment research team in October 2025, "The Spread of AI and the Contraction of Youth Employment," proves the point with numbers.
In the three years after ChatGPT launched (July 2022 to July 2025), 211,000 jobs for young people (ages 15 to 29) disappeared. Of these, 208,000 occurred in industries with high AI exposure—98.6 percent of the total decline.
The picture grows sharper up close. In computer programming and systems integration, including software development, youth employment fell 11.2 percent. Publishing dropped 20.4 percent, information services 23.8 percent. Even professional services such as law and accounting shed 8.8 percent.
What is striking is the movement among people in their fifties. Over the same period, jobs for this age group grew by 209,000, with 146,000 of them in highly AI-exposed industries. The Bank of Korea called this "seniority-biased technological change": as AI replaces the routine work of junior staff, the organizational and judgment skills of senior workers have, if anything, become scarcer and more valuable.
To sum up: the AI industry creates few jobs in itself. At the same time, in the other industries where AI spreads, it cuts existing jobs—and it pushes out young people trying to enter the labor market first of all.
We are measuring a new era with an old yardstick
And yet the government's industrial policy still leans on the formula "investment → growth → employment." Trillions of won for AI semiconductors, trillions for AI data centers, hundreds of billions to nurture AI startups. These investments will surely do something to lift GDP. But the employment multiplier is far too low to justify hanging a "job creation" sign over them.
The employment multiplier itself has limits as a metric. Because it is grounded in the input-output tables, it cannot properly capture AI's ripple effects, which cross the boundaries of industrial classification. The process by which AI raises manufacturing productivity and cuts jobs while simultaneously creating entirely new occupations in some other domain is impossible to measure with the existing multiplier.
The point is this: if the nation wants to measure the employment performance of its industrial investment, the old employment multiplier alone is not enough.
We need a new yardstick
Employment policy in the AI era needs at least three supplementary metrics.
First, an "employment displacement coefficient." We need to measure how much investment in one industry reduces or transforms employment in others. If a billion won invested in the AI industry creates three jobs in AI while wiping out five jobs in traditional services, the net employment effect is negative. We have to abandon the practice of calculating an investment's employment effect only within the industry that received it.
Second, an "employment quality index." We should look at the wage levels, job security, and skill demands of the jobs being created. If AI jobs are few but offer high pay and stability, that calls for an assessment different from a simple head count. Conversely, if the jobs AI replaces are being converted into platform labor or gig-economy work, that is not job "creation" but job "degradation."
Third, a "generational employment impact assessment." As the Bank of Korea's research showed, AI's employment shock appears in opposite directions depending on the generation. For seniors in their fifties it becomes an opportunity; for young people in their twenties, a barrier. If investment policy ends up blocking a particular generation from entering the labor market, that is not merely a failure of industrial policy but a social risk.
Change the criteria for investment, not its direction
This is not an argument against investing in AI. Falling behind in the race for AI technology would put the future of entire industries in jeopardy. That much is clear.
But the practice of measuring the payoff of AI investment by "how many jobs did it make" is now an anachronism. Slapping the label "job creation" on an industry whose employment multiplier is a mere two or three is dishonest—to the public and to the policy itself.
It is time, instead, to redesign the performance metrics for AI investment policy honestly. We should put indicators like contribution to productivity gains, an industrial-competitiveness index, and technological self-sufficiency front and center, and separate the employment question into its own distinct policy track. A more realistic approach is to design a structure that redistributes part of the added value AI creates into fields with high employment multipliers—education, care, and health.
How many people go to work when you pour in a billion won? If this old question still holds in the AI era, then we have to change the very way we answer it. You cannot measure a new era with an old yardstick.
