AI and the Two Antinomies of the Question of Work According to Luc Ferry.

In the chapter of his book AI: Replacement or Complementarity? devoted to the impact of artificial intelligence on employment and the labor market, Luc Ferry clearly summarizes the terms of the debate. It actually takes the form of two antinomies — or more precisely two theses that confront one another in the Kantian sense — which do not operate on the same level of analysis. On the one hand, there is a question of fact: are we, or are we not, moving toward the end of wage labor, replaced by artificial intelligences? On the other hand, there is a question of normativity: would such an evolution be good news or bad news? These two questions thus oppose two camps and two antagonistic theses, which is precisely what the term “antinomy” designates.

From the standpoint of the factual question, one finds on one side those who believe that AI will have little to no impact on employment, and on the other those who consider that it will produce substantial — if not disastrous — effects, triggering an unprecedented wave of technological unemployment in human history. If we now shift to the normative question, the dividing line moves: some see the end of work as a desirable evolution, while others perceive it as a “genuine catastrophe.”

According to Ferry, one must begin with the question of fact before turning to the question of normativity. He therefore analyzes the arguments of those who defend the thesis of continuity with past technological revolutions, and those who, on the contrary, support the idea of a radical rupture, viewing AI as a revolution of an entirely new kind.

What resurfaces with the recent developments in AI — and with the hypothesis of the near advent of artificial general intelligence (AGI), whose cognitive capacities would far exceed those of human beings, with Elon Musk suggesting that this milestone could be reached before 2030 — is the question of the end of work. This issue had already been widely debated by Jeremy Rifkin in The End of Work (1995), published a few years before the decision of the French socialist government to introduce the 35-hour workweek. Rifkin would revisit and deepen this thesis in The Zero Marginal Cost Society (2014), warning about the growing automation of labor driven by the increasing use of artificial intelligence. Today, this perspective seems, to some extent, validated by entrepreneurs such as Elon Musk and Sam Altman.

The factual question concerning AI’s impact on work. — It is therefore necessary to begin by examining the theses advanced by those who consider the end of work and the generalized automation of a large share of tasks and professions to be a chimera. According to them, AI would produce, at worst, effects comparable to previous industrial revolutions and, at best, beneficial outcomes for productivity, growth, and consequently job creation.

• AI will have little effect on labor productivity. The first argument maintains that productivity gains generated by AI would be too weak to produce significant effects on employment. An economist such as Daron Acemoglu, professor at MIT and coauthor with Simon Johnson of Power and Progress, estimates that they would not exceed 0.5%. This estimate, however, is contested by other studies, notably those from McKinsey and Goldman Sachs.

• Few jobs are fully automatable. The second argument claims that the number of professions that can be entirely automated is in fact very limited. This is the position, for example, of Philippe Aghion, economist and co-chair of the Artificial Intelligence Commission.

• AI fits within the continuity of previous industrial revolutions. The third argument rests on the idea that the AI revolution belongs to the continuity of past technological revolutions. It would follow the logic of “creative destruction” theorized by Joseph Schumpeter. Like any technological revolution, AI would eliminate certain jobs and professions while creating new ones, ultimately leading to net job creation. This position is notably defended by Philippe Aghion in his report AI: Our Ambition for France, submitted to the government on March 15, 2024.

• Generative AI will disappear. A fourth argument suggests that generative AI itself may be short-lived. Two AI researchers, Yann LeCun and Thomas Wolf, argue that large language models (LLMs) may no longer exist within five years, replaced by so-called “representational” AI capable of contextual reasoning, memory, and planning. Such a scenario would reinforce the idea that work will not disappear, since generative AI would not have had enough time to produce major productivity effects.

Opposing these advocates of historical continuity and the persistence of work are those who believe, on the contrary, that we are heading toward a massive acceleration of automation, leading to a literally “workless” world. They advance several major counterarguments against the Schumpeterian analytical framework.

• Even jobs once considered non-automatable will become so. They argue first that AI will replace jobs previously deemed resistant to automation. The classic distinction between routine and non-routine tasks is being challenged. Economists such as Daniel Susskind emphasize that current AI systems learn autonomously and perform tasks once considered uniquely human, including creative ones.

• Productivity is not the only factor that matters. They also challenge the idea that productivity gains alone determine employment outcomes. Several studies highlight significant gains linked to generative AI, which could translate into massive job losses, since it may become possible to produce as much — or more — with less labor.

• The AI revolution differs fundamentally from previous ones. They argue that the AI revolution is profoundly different from earlier technological shifts. There is no guarantee that destroyed jobs will be replaced by new ones. AI now affects activities once reserved for humans, such as reasoning or mastery of language, and must be understood within a globalized context that renders traditional analytical frameworks obsolete.

• Representational AI could worsen the situation. Finally, they maintain that so-called “representational” AI would only exacerbate the problem. The disappearance of generative AI would merely postpone the issue, as more powerful systems endowed with memory and planning capabilities could expand the range of automatable tasks even further. Some add that Retrieval Augmented Generation (RAG) techniques, by adapting general models to specific contexts, could multiply use cases and accelerate the automation of professions.

The normative question and the hypothesis of the end of work. — Let us now imagine, as Sam Altman and Daniel Susskind suggest, a world in which the quantity of available work represents only 20 to 30 percent of what it is today. What would such a world look like? What would our children and grandchildren do? And above all, how should we prepare them for it? These unprecedented questions call for answers now. A fundamental issue then arises: would a world without work, in which a large share of the production of goods and services is ensured by machines, be desirable? Once again, two positions confront one another. Some believe that humans would finally be liberated from the obligation to perform often unwanted jobs, viewing work as a form of alienation. Others consider such a world to be a catastrophe, arguing that work is an anthropological category that gives meaning to human existence.