Robotize this: Despite apocalyptic forecasts, AI will make society richer

Lessons learned

Humanity’s historical experience shows that new technologies and automation do not bring about total unemployment. Studies of long-term labor market trends confirm the benefits of automation.

In the U.S., for example, in the year 1900, 41% of all workers were employed in the agricultural sector, but by 2000, this figure had fallen to 2%. Without question, the appearance of combines and other specialized machinery drove people out of the fields, but they did not simply become unemployed. Instead, they moved to factories and the service sector, which were growing rapidly at the time. In addition, mechanization sharply lowered food prices, leaving more money to be devoted to satisfying other needs.

In another example, the invention of the standard shipping container in the 1950s cut the cost of loading one metric ton of cargo from $5.86 to $0.16, and the profession of dockworker disappeared almost entirely. However, shipping became so cheap that its positive impact on world trade exceeded that of all postwar international agreements combined.

After that, the advent of personal computers led to the closure of typing pools, and with the appearance of automatic telephone exchanges, telephone switchboard operators disappeared. At the same time, however, tens of millions of new jobs were created: for programmers, systems analysts, marketers, and designers.

Data gathered by Our World in Data show that from the start of the common era up to 1820, average world income per person remained within the range of 1,100–1,300 international dollars a year. The turning point came during the Industrial Revolution — thanks to the mechanization of labor, output outpaced population growth for the first time. As a result, in the two centuries that followed, global GDP per capita grew by a factor of 15, to exceed $18,000.

In the countries that spearheaded industrialization (Great Britain, and then the U.S. and Western European nations), the gap turned out to be even more impressive, with per capita GDP growing 25–30 times from 1820 to today (from $2,000–$3,000 to a present-day $60,000–$75,000).

Contrary to stereotypes, statistics show that the multiple growth of output over the past two centuries was accompanied not by increased “exploitation of workers,” but by a steady reduction in working hours and an increase in paid leisure.

In the 19th century, the prevailing opinion was that factory production had to rely on continuous and exhausting labor. In the 1870s, the working day in Europe and the U.S. usually lasted 10–12 hours, with a six-day workweek and no right to vacation. Workers agreed to such conditions simply because the alternative of peasant labor in the fields was even less attractive.

However, as mechanization and automation progressed, this trajectory changed. Today, in most developed economies, average working time has almost halved, with a standard five-day 40-hour workweek (and as low as 35 in some countries). Benefits appeared that did not exist in the preindustrial era: paid vacation and maternity leave, sick leave, public holidays, a ban on child labor, and old-age pensions.

So what is the origin of the fear that the development of algorithms will leave almost half of the planet’s population without work? This is a misunderstood estimate released in 2013 by University of Oxford researchers, who argued that 47% of jobs in the U.S. are at risk of disappearing because of AI. However, the researchers themselves stressed that 47% is merely the share of work tasks that computers are technically capable of performing, and not at all a forecast that half of people would actually be laid off. In other words, researchers examined the fundamental possibility of replacing a human with an algorithm in individual operations, not real job losses.

Later, OECD researchers approached the question by breaking down each occupation into specific repetitive operations. Having studied the structure of work of thousands of workers in 21 developed countries, the researchers found that people holding seemingly identical positions often perform fundamentally different tasks. Their calculation showed that the real risk of full replacement by machines applies not to 47%, but to only 9% of jobs, ranging from 6% in South Korea to 12% in Austria. For the vast majority of people, technology does not abolish an entire profession but takes over routine functions and complements human labor.

The real risk of full replacement by machines applies not to 47% but to only 9% of jobs, ranging from 6% in South Korea to 12% in Austria

In the history of the world, no country has achieved sustained growth in its population’s incomes without regularly adopting major technical inventions, such as the steam engine, electricity, the internal combustion engine, and computers. But technology is not the only growth factor, as it delivers results only when combined with quality education, intelligent investment, and functioning government institutions. After all, a transition period is always accompanied by difficulties that require effective management.

How AI is changing the labor market

Artificial intelligence largely repeats the path of earlier technological revolutions, but it features several fundamental differences. The main similarity is that technologies rarely destroy professions instantly. Any job consists of many small operations, and a machine that performs two or three of them cannot replace the human specialist completely.

Experts at the Organisation for Economic Co-operation and Development (OECD) stress that automating individual tasks changes the structure of the workday, as its calculations showed that 27% of jobs in developed countries are in occupations with a high share of potentially automatable operations. However, even in those, algorithms take over only 20% to 40% of specific routine duties, while tasks requiring critical judgment, communication, and personal responsibility remain with humans, who continue to be responsible for goal-setting and decision-making in disputed or nonstandard situations.

Humans continue to be responsible for goal-setting and decision-making in disputed or nonstandard situations

The link between falling costs and rising demand is clearly visible in the market for language translation and digital services. According to industry associations, the appearance of machine translation systems cut the cost of rough processing one page of text by more than 70%. However, rather than destroying the translator’s profession, it led to a significant increase in the volume of material processed: companies began translating dozens of times more internal documentation, instructions, and customer reviews. As a result, overall employment in localization and editing grew, shifting from mechanical translation to review, semantic adaptation, and fact-checking.

The falling price of services such as writing software code and performing legal and financial consulting tasks makes them accessible to those who could not afford them in the past. As a result, the market expands, and demand for labor remains.

Just like previous technologies, AI needs time to lead to an explosive rise in macroeconomic indicators. Economist Paul David described this pattern in a 1990 study by using the electrification of factories as an example. Three decades passed between the appearance of reliable electric motors in the 1880s and the surge in labor productivity in U.S. manufacturing. The share of electrified factories exceeded 50% only by 1920, and for many years annual growth in factory output did not exceed 1.5%.

At first, enterprises simply replaced the old steam boiler with a single electric motor, keeping the existing layout and belt drives in place. The real leap in productivity — to a record 5% a year — began only when factory owners rebuilt the environment from scratch, erecting spacious single-story shops, connecting an individual motor to each machine, and training managers in new work schemes.

The good news for us is, the adaptation process for today’s office and service workers is far milder than the upheaval of past eras. And this is an important difference between the adoption of AI and that of classical technologies. In the 19th century, the mass departure of illiterate peasants to factories was accompanied by severe disruption of everyday life and required the creation of school education.

Today, the threshold for changing career paths within the services and information sector is incomparably lower. Most specialists already possess basic literacy, as well as business communication and information-search skills, allowing them to master adjacent fields without the need to spend years retraining.

AI also differs from earlier technologies in that the algorithms themselves help people master new duties, taking on the role of a personal instructor. Studies of software assistants in corporate support showed that by using model prompts, even the least experienced employees tend to reach the level of qualified specialists in just two months rather than the previous transition time of six months, and the speed and success of their task resolution rose by 34%.

A micro effect and faith in the future

To assess the impact of AI on the modern economy objectively, researchers divide the available data into three different levels.

The first is direct measurements inside individual enterprises, where it is possible to empirically compare the performance of employees with and without a neural network.

The second is general statistics of national labor markets in the first years after the appearance of mass chatbots, showing changes in real wages, hours worked, and the hiring process.

The third is predictive macroeconomic calculations that model the technology’s impact on aggregate productivity and the country’s GDP years ahead.

The most accurate evidence of real productivity growth today has been obtained at the first level. A study published in the Quarterly Journal of Economics analyzed the work of a tech giant’s customer support department. The authors observed 5,172 employees and processed 3 million conversations during the phased rollout of a software assistant based on the GPT-3 model. In the baseline calculation, the number of successfully resolved inquiries per hour rose by 15%. The program essentially accumulated the best phrasing of experienced specialists and passed it on to newcomers.

PwC calculated that productivity growth is 40% higher in the companies that are most exposed to AI. Since 2022, the leaders in corporate AI use have tripled their lead in labor productivity growth over those that have barely adopted it. A link was also established with growth in wages and headcount, which both grow faster in companies where AI is widely used.  

If we turn to labor market data for the first years of ChatGPT’s existence, the macroeconomic picture looks more restrained. Researchers studied survey data and the official state employment register in Denmark, presenting their findings in a paper from the National Bureau of Economic Research.

The analysis showed that although employees began actively using algorithms for tasks and rated their convenience positively, this had no statistically significant effect on their earnings or recorded working hours over the first two years.

At the same time, a study by the Stanford University Digital Economy Lab recorded a worrying trend in the U.S. labor market. While no mass layoffs have been observed, employment of people aged 22 to 25 in the occupations most exposed to automation has been lagging noticeably behind that of other fields.

By June 2026, the hiring gap with occupations less affected by the technology had reached 19%, though the Stanford authors stress that this link is descriptive and does not in itself prove the negative impact of algorithms. However, it still suggests that many young specialists are having a problematic start to their careers.

In long-term macroeconomic scenarios, researchers’ conclusions vary substantially. The OECD’s 2024 model suggests that the use of AI could bring the U.S. economy an additional annual labor productivity gain of 0.4–0.9 percentage points in ten years’ time. However, economist Daron Acemoglu offers a much more cautious estimate. By his baseline calculations, the increase in total factor productivity (TFP) over the decade will not exceed 0.66% — that is, less than 0.1% a year. Still, taking into account investment in fixed capital, the impact on overall GDP should be somewhat higher, Acemoglu notes.

To compare these models, it is important to use a single indicator: the OECD’s corresponding TFP calculation projects from 0.25 to 0.6 percentage points of additional growth each year — noticeably higher than Acemoglu’s estimate.

This discrepancy in figures illustrates the subject of present-day debates. Economists no longer argue about whether algorithms can speed up specific intellectual operations. After all, at the level of individual work tasks, the productivity gain has already been reliably recorded and measured. The main question now is how quickly and to what extent the benefit from innovative software at individual workplaces can translate to the national level.

In aggregate indicators — real wages, the number of jobs, and GDP growth — the effect has been barely noticeable. This inconsistency forces researchers to be more conservative in their forecasts, but this does not mean that the technology will not become the basis for powerful long-term development.

How robots close the demographic hole

The optimistic scenario for adopting artificial intelligence does not mean that the labor market will automatically adjust to the new conditions. A positive result is possible only through a combination of objective demographic processes and well-considered actions by the government and business, which can soften the loss of jobs and help people find new roles in the economy.

The demographic crisis in developed countries has become the most important backdrop for current automation. According to Employment Outlook 2025, the working-age population has begun to decline in many OECD countries. According to demographic forecasts, by 2030 the number of people aged 20 to 64 in European Union countries will fall by more than 10 million. In Japan and Germany, a shortage of workers is already holding back business development.

Under these conditions, adopting software algorithms helps maintain production volumes despite the growing shortage of personnel. However, the deficit of employees is distributed extremely unevenly across sectors and regions, affecting health care, home nursing care, construction, and technical trades most acutely. Meanwhile, generative models are most actively adopted in the corporate sector, marketing, basic programming, and workflow management.

Generative models are most actively adopted in the corporate sector, marketing, basic programming, and workflow management

An acute shortage of nurses or workers does not guarantee that a laid-off office analyst or lawyer can easily change profession and fill such a vacancy, meaning that the development of solutions to the imbalance in workforce availability across sectors requires a targeted government policy. Restrictions on the adoption of AI are not the answer, as attempts to ban technologies or force enterprises to keep unneeded jobs only slow productivity growth across the whole country.

An effective strategy consists of supporting specialists during their transition by providing guaranteed income insurance upon job loss, free retraining programs, and compensation of relocation costs to regions with high demand for personnel. For example, in Denmark about 20% of workers change profession or employer every year, but thanks to targeted payments and short-term retraining courses, this process does not lead to prolonged unemployment or a social crisis.

Young specialists require special attention. Since neural networks are taking over the drafting of simple reports, data collection, and the writing of initial code, the customary chain of internships is beginning to break. As studies show, when entry-level tasks disappear, companies need to rebuild their on-the-job training and mentoring programs. Instead of routine manual work, students and interns are taught to check generated results, set tasks for algorithms, and coordinate work chains.

Finally, a key role in distributing the benefits brought by new technologies belongs to market competition. If advanced models and infrastructure end up concentrated in the hands of a few monopolies, the savings from lower costs will remain inside these corporations in the form of excess profits. By contrast, active rivalry between companies forces businesses to lower selling prices for goods and services for end consumers.

Economic history shows that the decline in the cost of computing and communications in the late 20th century was the result of market competition — consumers got higher-quality and cheaper hardware and services, which in turn expanded the overall market and created new jobs in adjacent sectors.

In 2004, William Nordhaus calculated that thanks to market competition, the creators of new technologies capture only about 2% of the total economic value of their innovations, while the remaining 98% “flows” to consumers, who get much cheaper and better products — and therefore a higher purchasing power.

The authors of another study proved mathematically that the fall in prices of computer components (caused by progress in semiconductors and competition among manufacturers) triggered an avalanche-like accumulation of capital across the entire economy: in banking, logistics, trade, and media. This did not reduce aggregate demand for labor but led to an employment boom in adjacent sectors like distribution and applied software development.

How productivity turns into prosperity

A person’s well-being is determined by their real, not nominal, income. If algorithms lower the cost of a service without reducing its quality, the consumer benefits even if they do not receive a pay raise, because the same amount of money buys more. Here it is more appropriate to speak of relative cost: specific goods and services become cheaper compared with what they would have cost without the adoption of new tools.

Whether cost savings are reflected in the consumer price depends on external conditions — above all market competition, the absence of artificial scarcity, and the buyer’s real ability to switch to another supplier.

As already mentioned, from the end of the 19th century to the middle of the 20th, working time fell sharply in developed countries, but then the reduction slowed and almost stopped as society on the whole opted to direct further efficiency growth not toward a larger share of free time but toward the volume and quality of consumption (sophisticated medicine, higher education, electronics, improved housing, and the service sector).

High productivity opens room for wage growth. However, the median wage of an ordinary worker will not necessarily rise at the same pace. An OECD study showed that in 24 developed countries between 1995 and 2014, the gap between growth in labor productivity and growth in the real median wage averaged 13%. Productivity rose by about 28%, while median pay for workers rose by only 15%.

The study notes a fall in labor’s share of national income and a rise in market concentration — that is, high returns on capital go to a narrow circle of those who control the algorithms, computing power, databases, or the business that adopts the technologies. On the other hand, the development of technology creates new areas of activity that complement automated systems. If demand for people with scarce skills in such adjacent tasks grows, this helps sustain the overall level of employment and pay.

The fact that automation does not simply replace people but creates complementary tasks was shown by an MIT study. About 60% of all jobs in today’s developed economies did not exist at all in 1940. In fact, they arose directly as a result of technological growth.

A win for everyone or just for the chosen few?

Debates about who will benefit from the spread of artificial intelligence are dominated by two opposing views. Some consider growing inequality an inevitable consequence of the use of algorithms, while others attribute it solely to government policy. In reality though, the result depends on the joint action of technology players and public institutions.

The aforementioned Acemoglu and fellow economist Pascual Restrepo studied the impact of automation on the labor market in detail. In their 2022 paper, they showed that 50% to 70% of the growth in wage inequality in the U.S. over the period from 1980 to 2016 was linked to technological displacement — a situation in which machines and software replaced people in routine operations.

Computer systems do indeed devalue repetitive labor, but at the same time they raise the value of other skills, such as the ability to set a task correctly, to double-check a machine’s result, to take on legal or professional responsibility, and to communicate effectively with people. The adoption of algorithms stimulates demand for specialists who prepare quality data, train and adapt systems for narrow industries, or coordinate complex automated processes.

Another important change concerns the concentration of capital. Training large models and maintaining server infrastructure require billions in investment, which gives a gigantic head start to a small number of technology corporations. This monopoly, of course, is not guaranteed to last. The emergence of open and lightweight architectures, as well as a sharp drop in the cost of computing, creates accessible alternatives and prevents a narrow circle of players from closing off the market completely.

The viability of open-source architectures proves that total monopolization of the AI market is unlikely, and the gap in performance between commercial giants and open systems has rapidly narrowed on key benchmarks (MMLU, HumanEval). This democratization is confirmed by statistics: by March 2026, total downloads of open AI models on Hugging Face reached 2 billion — an almost sixfold rise from the 340 million recorded a year earlier.

Today there is no need to pay for a ChatGPT subscription or other popular neural networks every month. Powerful artificial intelligence can be downloaded from the internet and run on a laptop retailing for $1,500. Such a program will work solely on your computer’s resources and deliver results on a par with paid services. The rule here is simple: the more powerful your “hardware,” the smarter and more complex a model you can install on it.

This opens huge opportunities for business. Any company can install free neural network models on its servers and build powerful, customized AI tools for its tasks without depending on giant corporations or purchasing expensive access. Everything works on-premises, securely and under full control. Thus open source forms a powerful market counterweight, turning basic AI tools into publicly available infrastructure.

Any company can install free neural network models on its servers and build powerful, customized AI tools for its tasks

Under such circumstances, antitrust regulation decides whether competition ultimately forces companies to lower prices for buyers or allows them to retain technological rent in the form of excess profits. Labor market rules are equally important — where there are strong trade unions, a transparent system of collective agreements, and a clear minimum wage level, the gain from productivity reaches ordinary workers faster. The experience of the Nordic countries shows that, with the same level of digital tool adoption as in the U.S., developed labor institutions keep income inequality at a significantly lower level.

The transition to new kinds of activity requires support from the tax system and social insurance. Accessible retraining programs allow people whose professional tasks are being automated to master adjacent fields without a catastrophic drop in income. An example is the Danish flexible employment system (flexicurity), in which ease of hiring and firing is combined with high government payments during the job search and mandatory training in new specialties.

Finally, rules of professional responsibility and quality control are critically important in industries such as health care, education, and law. Clear standards prevent service providers from lowering quality to save money, which protects consumers from the hidden costs of automation.

Studies show that high levels of technology adoption do not necessarily lead to income polarization, as inequality is consistently lower in high-tech states than in commodity-based countries. A high Economic Complexity Index, reflecting the diversity and technological knowledge intensity of the goods produced, usually means a lower level of income inequality. A high-tech economy requires distributed human capital, whereas a commodity-based one concentrates wealth around narrow rent-extraction chokepoints.

Countries with the highest density of technology and complex industrial production (Germany, Japan, South Korea, Taiwan, the Nordic states) keep the Gini coefficient after taxes and transfers in the range of 0.26–0.33. However, large resource exporters (the oil monarchies of the Persian Gulf when migrants are counted, commodity economies of Latin America, and also Nigeria and Angola) show a Gini coefficient at the level of 0.42–0.55 and higher, with a colossal gap between the top decile and the rest of the population.

Overall, the history of technological development shows one stable pattern: the advent of more efficient tools has invariably made society richer as a whole. Even when entire professions became a thing of the past and individual workers had to change occupations or go without work, the resulting standard of living, the availability of goods, and the quality of everyday life for the majority invariably rose.

The foundation for well-founded optimism about artificial intelligence is built precisely on this proven mechanism. When new solutions make it possible to achieve the same or a better result at lower total cost, the economy gets a real surplus of resources. Even if it does not take advantage of the opportunity, society at least gains the chance to direct this surplus where it is needed most: raising the quality of medicine and education, increasing the affordability of goods, developing public services, and increasing free time. Attempts to hold on to every familiar job and preserve obsolete labor operations have historically always led to a dead end. Instead, adaptability is the key to prosperity.