Dossier No. 37, June 26, 2025; this “dossier” is a collection of short articles by the New Anticapitalist Party-Revolutionaries (NPA-R) in France (translated from French)
Capitalists around the world are now betting on artificial intelligence (AI) in the hope of significant gains in labor productivity. The goal remains to replace some human labor with machines that require less labor to produce. It does not matter that thousands of workers are condemned to unemployment as a result of restructuring. Even if the labor savings achieved were to free up labor to develop new sectors of production, as some economists claim, this is not an immediate prospect and in no way guarantees that those who are laid off today will find work in the short term.
Eliminating “intellectual” jobs, further alienating labor
AI is therefore perceived by many workers as a threat, rather than as a new tool at their disposal to reduce their working hours and improve their working conditions. Take Hollywood screenwriters, for example, who went on strike for 148 days in 2023, partly to regulate the use of AI in their profession. Unlike previous waves of automation, this primarily affects intellectual professions, executives, and artists. So-called “generative” AI (which produces text, images, or sound) can already replace some of the tasks performed by translators, computer developers, journalists, and lawyers, while other AI can process vast amounts of information in seconds to aid decision-making by consultants at McKinsey (global management consulting company and similar firms, doctors (for diagnoses), human resources (HR) managers (for tracking applications) and traders. Professions that were once considered prestigious are now on the front line, along with call center agents, who have been partially replaced by chatbots and callbots.
This is where artificial intelligence gets its name: from the ability of these technologies to replace so-called “intellectual” work. However, AI also covers tasks that we would not intuitively categorize as intellectual, such as automated driving, or mixed tasks such as robot control, which pave the way for machines that are increasingly “freed” from human labor… while further alienating workers, who are becoming mere “machine servants.”
Profitability not (yet) in line with capitalism
The main obstacle to the widespread adoption of these systems remains the high cost of operating them for certain tasks. AI requires billions of calculations and therefore huge warehouses of computer servers managed by a few large groups, such as OpenAI and Microsoft. Once the buzz surrounding the world of start-ups has died down, truly profitable applications for capital remain to be found, which will undoubtedly need to be more economical in terms of calculations, but also more reliable. This is all the more difficult as objectives in these areas are more easily achievable by humans.
The horizon for profitability could therefore be further away than the financial world had anticipated, with the risk of a bursting of the speculative bubble around tech, similar to the “new technologies” bubble at the beginning of the century, leading not only to financial losses, but also to business closures and unemployment.
In the medium to long term, depending on the industry, workers must nevertheless prepare for the use of radical innovations introduced by AI to enable bosses to lay off workers while increasing pressure on wages, working conditions, and competition between workers. The solution to this does not lie in a principled rejection of this technology, nor even in its illusory regulation within the framework of capitalism, but above all in strengthening the capacity of workers to resist collectively in order to demand a ban on layoffs, the sharing of work among all, and control over their work processes.
What are we talking about?
The term was coined in the 1950s, when mathematician Alan Turing examined the idea of intelligent machines, i.e., broadly speaking, machines capable of performing tasks that we humans do with our intelligence.
From the outset, several areas of research coexisted, including “learning” AI, which is the most widely used today. This is based on artificial neural networks, a highly simplified version of biological neurons. These “neurons” transmit signals of varying strength to each other, which lead to the expected result. A training phase using data is necessary to adjust the strength of the signals exchanged, a bit like a mixing desk with various buttons to adjust the sound: this is machine learning. For example, if we give the program an image of a dog and it says “cat,” each button is adjusted to produce “dog” instead. If done correctly, the settings will eventually give the right result immediately for images that the program has never seen before. In modern AI, there are millions of neurons, and therefore millions of knobs, and it takes millions of images to adjust them correctly.
While the first algorithms of this type date back to the 1970s, techniques have since emerged, such as deep learning, which became established in the 2010s thanks to the explosion of computing power and the exploitation of huge databases.
A “chatbot,” such as ChatGPT or DeepSeek, works on this principle. Its goal is to write a response that is correct in both syntax and (ideally) meaning based on a question asked. To do this, the AI tries to guess the answer one word at a time, choosing from the most likely options based on the texts it has been trained on, without however being able to distinguish between true and false.
Since the texts on which the chatbot is trained are produced by humans, with all their biases and approximations, programmers have to introduce certain corrections manually. Thus, the Grok AI, produced under the aegis of Elon Musk, will not always say the same thing as an AI concerned with the interests of workers. We are far from “mathematical neutrality,” even if Grok has sometimes spectacularly contradicted its master, so outrageous were its statements!
Marc Laverne
