By Zisis Sourlas
The following text is the introduction by Zisis Sourlas from Greece, to the discussion titled “The invisible cost of AI”, held during the 33rd camp of Antinazi Zone / YRE (Youth Against Racism in Europe), the antifascist and antiracist campaign built around Xekinima (Greek section of ISp). The discussion was also introduced by Leonard Τraistariu from GAS, the Romanian section of ISp and Yannis Kapsalis from Greece, whose introduction will be published in the coming days.
________________________
In recent years, massive investments have been made in the field of Artificial Intelligence (AI) with the promise of skyrocketing productivity, the replacement of many jobs, and a new era of automation. Some industry executives even go so far as to speak of impacts comparable to those of the Industrial Revolution. Achieving these goals is the only way for these investments to actually deliver their projected returns.
There are various and conflicting views on the matter. Recently, we have indeed seen a number of tech giants carry out mass layoffs of thousands of employees, on the grounds that they will be replaced by AI (for example, 16,000 employees at Amazon, 21,000 at Oracle, over 1,000 at Klarna, and so on). On the other hand, a study by MIT last year found that although $30–40 billion had been invested in generative AI in the first half of 2025, in 95% of cases, the return on investment was nearly zero, suggesting that the future promised by major tech companies is not as certain as it seems.
A closer look at the situation, however, reveals a more complex picture. The wave of layoffs was followed by a wave of rehiring or strategic re-evaluations at many companies, such as Klarna and Duolingo. Furthermore, these layoffs cannot be attributed entirely to AI and may also be related to a readjustment within the industry following the pandemic. Key causes included both AI’s failure to achieve results of the same quality as those produced by humans, as well as rising costs, as AI companies replaced the fixed-subscription model with a pay-per-use model, which caused costs to skyrocket, especially for the most advanced models. A prime example is Uber, which spent its entire AI budget for 2026 in less than four months.
Although skeptics present these setbacks as proof of the failure of the AI, they do not mean, that there is no risk for workers. The aforementioned MIT study did reach the main conclusion mentioned above, but it also pointed out that with the development of AI agents -that is, systems that, in addition to chatbots, also feature a range of tools, commands, and memory systems- we can speak of a new type of internet where autonomous agents will communicate with one another and complete tasks independently. For example, an agent that, upon detecting shortages of raw materials in a manufacturing facility, communicates with and negotiates orders with the suppliers’ agents and autonomously handles the coordination and orders. Systems of this kind are currently being developed by all major technology companies, as well as by many smaller ones.
Consequently, the impact of AI on the labour market may not be immediate, but in the medium to long term, the risk of massive job losses is real.
To assess this risk, we need to answer two key questions. The first is to assess the extent to which AI will be capable of replacing workers, and the second is (as many economists point out) the extent to which and the speed at which it will be able to be integrated into the production process.
Regarding the first question, the answer is not obvious, although there are certain serious obstacles to the development of AI.
One has already been mentioned, and that is the rising cost. The approach -more and more data, larger and larger models, and more and more computing power- will reach its peak. Data centers, microchips, and the volume of data, among other things, cannot keep growing indefinitely. Furthermore, this approach will not be economically or ecologically sustainable due to the enormous costs associated with energy and infrastructure. However, various Chinese models, such as Deepseek and Kimi K3, show that there is another path for AI development, one that requires much less computing power while achieving an equally high level of AI performance.
Another issue concerns the AI’s inability to learn from experience. As we speak with ChatGPT, it does not retrain itself based on our conversation; its “mind” remains unchanged by the experience*, unlike that of a human. For a model to “evolve,” it must go back to the lab so that the next version can be built. Creating a model that can self-improve is a problem that has not yet been solved by advanced AI labs, and it remains an open question whether it can be solved with existing technology. This observation also appears to influence how AI is integrated into the production process. A study by the Federal Reserve Bank of Dallas found that while AI is highly effective at replacing young workers who have only theoretical training, it is only capable of augmenting -not replacing- workers with on-the-job experience. On the contrary, it has even led to wage increases for experienced workers in sectors exposed to AI. This naturally raises the contradiction: how will experienced workers be replaced when they retire if not enough young people are being hired to gain experience?
Estimates regarding the integration of AI into production and the resulting increase in productivity are contradictory. However, unlike in previous years, many economists have begun to question this trend more and more. The projections of tech giants, who expect an immediate and rapid increase in productivity over the next few years, are likely overly optimistic and unrealistic. Other estimates are less optimistic. Economists at Goldman Sachs predict the first tangible benefits around 2030, with peak performance occurring around 2034, while the OECD is less optimistic, forecasting substantial macroeconomic returns in the mid-2030s. The extent to which productivity will increase is also uncertain and remains a point of disagreement among economists.
In conclusion, we can say that AI is a technology with a proven ability to perform intellectual work and that it has the potential to transform the production process. However, the extent to which it will be able to replace humans -as well as the speed at which this can be achieved- is unclear and likely falls short of the promises and expectations of the tech giants, who are currently fuelling a bubble.
Whichever perspective proves to be more accurate, the direction set by the capitalists in this sector is clear. Driven by the logic of profit, they pursue and peddle promises of the mass replacement of workers by autonomous systems. To achieve this, they have created these monstrous models, which, in the final analysis, are a vast melting pot of centuries of collective human knowledge.
They are commercializing this knowledge and attempting to turn it into a weapon against workers. In this effort, environmental protection, safety, and even the quality of the final product are not taken into account. The extent to which AI directly threatens us is one debate; the fact that the profit-driven system will use it against us is another—and it is a given.
The significance of these technologies, as well as their very nature, underscores the need for them not to be subject to the logic of profit. They should be under the control of society and the workers themselves to ensure transparency and safety. This would allow them to be used for research and the development of solutions that benefit society, rather than for autonomous weapons, surveillance, and control. We have seen evidence that AI can accelerate research in a range of fields, such as medicine and pharmaceuticals. We need only consider the possibilities that are opening up to realize that these advances must be collectively owned, not commodities or patents.
It is both feasible and necessary for the development of nanotechnology not to consume these unimaginable amounts of energy. Smaller, more efficient models, tailored to specific tasks, could be researched and developed—a very different direction from these behemoths that blindly pursue a vague notion of “general intelligence” on a human level. Cooperation rather than competition in this field could lead to a human-centered, ecological AI, with society reaping the benefits in the form of shorter work hours without a reduction in pay, the elimination of routine and dangerous tasks, cheaper and faster research, and so on…
The only thing standing in the way of this future we’ve described is not the existing technological capabilities, but their subordination to the logic of profit; and it is against this trend that the Left and social movements will need to wage struggles in the coming period.
