EC[ON]OMY

The future of work: how AI shapes employment trends

Artificial intelligence has moved beyond the lab and into the core of the global economy. In just a few years, it has evolved from early automation to self-learning systems. AI can now write text, generate code, analyze data, and guide business decisions. But alongside these capabilities comes a new risk – the large-scale loss of jobs.

Economists call it “technological unemployment.” The idea isn’t new – it dates back to the industrial revolutions – but today it carries fresh meaning. Modern AI doesn’t just perform tasks; it learns from people. Every mistake corrected and every interaction feeds the algorithm. The more workers use it, the faster it improves. Yet this strength hides a threat: once AI reaches a certain level, it begins to replace the very people who trained it.

An economic model highlights three possible futures. The first is partial adoption of AI. Productivity triples, but employment drops by about 23%. Half of these losses happen within the first five years. The second is unlimited adoption, where AI learns so quickly and efficiently that it creates new tasks and supports job growth. In this world, technological unemployment never takes hold. The third is non-adoption. AI stalls, fails to spread, and the labor market returns to business as usual.

Why such different outcomes? Because AI follows a “learning-by-using” path. Workers help it improve, but in doing so they make themselves replaceable. This creates a paradox: productivity surges, while the risk of unemployment grows at the same time.

The danger is most evident in the partial adoption scenario. The faster AI learns and the fewer errors it makes, the more aggressively firms replace people with machines. Jobs disappear, but those who remain employed see their productivity and wages rise. Society, however, bears the cost. The share of income going to labor shrinks, inequality deepens, and social tension rises.

The transition period is the hardest. Economies enter cycles: more workers boost AI learning, but once the technology improves, replacement accelerates. Jobs vanish, learning slows, and the cycle repeats. Researchers compare this to a predator-prey model: more workers mean faster AI growth, but stronger AI means fewer workers. The economy spirals through ups and downs until it stabilizes in a new equilibrium.

If AI learns extremely fast and avoids errors, the second scenario – unlimited adoption – becomes reality. Here, productivity and employment rise together. Job growth continues, and technological unemployment never arrives. But this path depends on a fragile balance: rapid learning, low failure rates, and moderate bargaining power for firms.

The third option, non-adoption, is less likely today but still possible. If AI fails to progress or struggles to find real-world use, the labor market simply returns to familiar patterns. Stability remains, but growth potential is lost.

Labor market frictions amplify these dynamics. In theory, if job searches were costless and instant, AI would only reallocate employment. In reality, searching takes time, vacancies are expensive to create, and layoffs carry real costs. These frictions magnify technological unemployment and make the outcomes harder to predict.

Even traditional rules of labor market efficiency fall short. Take the well-known Hosios condition, which ensures a “fair” split of surplus between firms and workers. In the AI era, it no longer guarantees efficiency. AI introduces new externalities: jobs vanish faster than new ones appear, AI slows its learning when fewer workers are around, productivity rises unevenly, and the job-matching process itself changes. Old frameworks no longer fit the new reality.

The model points to one solution: subsidies for jobs most at risk of being replaced by AI. These subsidies should grow over time, easing the transition and reducing social costs. The benefits are not only social but also economic. In the short run, welfare could improve by about 26%. In the long run, the gain could exceed 50%. In rare fashion, social policy and economic efficiency align.

Why are expert forecasts so divided? Some predict strong job growth, others expect steep declines. The answer lies in parameters. If AI learns quickly and rarely fails, employment may rise. If learning is slower and firms hold more bargaining power, partial adoption dominates – and job losses follow.

For business, the message is caution. Companies should prepare for volatility, rethink hiring strategies, and invest in reskilling. For governments, the challenge is building new tools. Traditional labor policies are insufficient; flexible approaches tailored to AI are essential. For society, the lesson is clear: uncertainty will be the new normal.

Artificial intelligence is no longer just a technology – it is a macroeconomic force. It can accelerate growth or trigger a jobs crisis. The first five years will be decisive. If the transition is managed well, the economy will benefit. If not, technological unemployment could become reality.

Sultan Valikhanov, expert of the EconomyKZ.org portal

Scroll to Top