EC[ON]OMY

Balancing AI adoption and employment: the role of government subsidies

Automation has always carried a double edge. On one side, it drives efficiency, lowers costs, and accelerates economic growth. On the other, it reduces jobs, squeezes incomes, and widens inequality. The 20th century saw this with the mechanization of farming, industrialization, and later with computers and robots. In the 21st century, the challenge has taken a new form. Artificial intelligence is entering the workplace, and its impact on the labor market is far more complex and contradictory.

The main difference is that AI evolves through “learning-by-using.” Traditional machines didn’t change based on who operated them. A tractor or assembly line did its job the same way regardless of the worker. Algorithms are different. Their quality improves through human interaction. The more errors people correct, the more tasks they expose the system to, the faster it learns. That means workers both train AI and eventually make themselves replaceable.

Economists have long relied on the Hosios condition to define labor market efficiency. It says a market is efficient if the share of gains going to firms and workers reflects their contribution to job creation. When this balance holds, there’s no need for government intervention. But in the AI era, this rule no longer works.

AI introduces new externalities. Jobs vanish faster than new ones are created. Remaining workers see their productivity rise, but society as a whole loses balance. When fewer workers remain, AI learns more slowly, which paradoxically can slow down progress. At the same time, algorithms change how hiring works by reshaping vacancies, negotiations, and the speed of job matching. Even if the Hosios rule looks satisfied on paper, the equilibrium is still inefficient.

That inefficiency means the market cannot balance itself. Firms act rationally: they adopt AI where it saves the most costs, replacing workers aggressively. Jobs disappear, new ones arrive more slowly, and technological unemployment grows. The model shows that half of all long-term job losses happen within the first five years of adoption. This is the critical window where government policy matters most.

The solution is a system of optimal subsidies. The idea is to support jobs most at risk of replacement by AI. At first glance, it sounds counterintuitive. Why preserve positions that will eventually disappear? But in reality, it buys time. Keeping workers in place longer allows AI to keep learning and productivity to keep rising, while softening the social shock.

Subsidies act as a buffer. They slow down the speed of displacement and let the economy adapt more gradually. Firms keep incentives to adopt AI but also to retain employees. The result is a smoother transition, fewer social costs, and higher overall welfare. And the model shows the numbers are striking. In the short run, optimal subsidies can boost welfare by about 26.6%. In the long run, the gain exceeds 50%.

The logic is simple. When governments help keep more workers in the system, AI has more interactions to learn from. More learning means faster improvements in productivity across the economy. The paradox is that preserving jobs that look redundant actually makes the technology itself stronger. Everyone wins: firms get better tools, workers keep their jobs longer, governments avoid sharp unemployment spikes, and society enjoys sustainable growth.

But it’s not just about providing subsidies — it’s about designing them properly. Support must be dynamic, not fixed. At the start, when displacement is still limited, subsidies can be modest. Over time, as AI’s impact deepens, subsidies need to grow. This rising support smooths unemployment peaks and balances the pace of adoption with job preservation.

Another key factor is bargaining power. When firms claim too much of the surplus, they adopt AI more aggressively and replace workers faster. When workers hold stronger positions, adoption slows. Policy must account for this balance. Subsidies can be a tool to shift incentives, preventing extreme outcomes on either side.

The practical importance is clear. The first years of AI adoption shape the long-term path. If governments do nothing, the social costs will be heavy. Sharp job losses will cut consumption, reduce trust in technology, and increase social tension. This in turn will slow AI’s adoption and weaken its benefits. But if targeted subsidies are introduced early, the transition becomes smoother. People start seeing AI not as a threat, but as an enabler of growth.

For business, the message is that government can be a partner, not just a regulator. Subsidies allow firms to modernize while keeping workers on board, avoiding a social backlash. For governments, it is a chance to secure both stability and growth. For society, it is proof that AI does not have to mean mass unemployment but can drive new development.

Optimal subsidies are not just social policy. They are part of a new economic architecture in which AI is a system-wide force. Supporting jobs under automation is not a drag on growth — it enhances it. History shows that countries that managed the costs of technological transitions fared better in the long run. Today, governments have the tools to do it deliberately.

The point of the model is that AI is no longer just a technology or a business tool. It has become a macroeconomic force, reshaping employment structures, bargaining rules, and income distribution. Old rules are not enough. Left on its own, the market won’t manage the transition. But the right policies can turn risk into opportunity.

Optimal subsidies are the answer to how this transition can be managed. They buy time, reduce social costs, and accelerate productivity growth. They make AI adoption beneficial for all sides. In the coming years, such measures will decide whether artificial intelligence becomes a source of crisis — or the foundation of a new economic boom.

Alen Serik, expert of the  portal EconomyKZ.org

Scroll to Top