Source: French to English Tester Published on: 2026-04-20
Source: The Conversation – in French– By Hugo Spring-Ragain, PhD student in economics / mathematical economics, Center for Diplomatic and Strategic Studies (CEDS)
Artificial intelligence does not so much destroy jobs as it profoundly changes the skills required to perform them. From this confusion between jobs and skills, errors may arise in policies supporting ongoing transformations.
Each major technological wave has produced its share of contradictory predictions about employment. Artificial intelligence (AI) is no exception. But before knowing how many jobs AI will create or destroy, we need to agree on what it actually automates. The answer requires distinguishing three notions that public debate regularly confuses: employment, skill, and task.
The great waves of automation have followed a remarkably stable logic over two centuries: steam, electricity, and industrial robotics have displaced repetitive physical tasks and spared non-routine cognitive work. This empirical regularity has beenformalized by Autor, Levy, and Murnanesince 2003 under the name “task polarization hypothesis.”
A persistent illusion
Automation erodes intermediate jobs, those of skilled blue-collar workers and office employees performing routine tasks, but spares the two extremes. On one hand, non-routine manual tasks, such as plumbing or caregiving, on the other, non-routine cognitive tasks, such as analysis, consulting, or expert writing. The latter constituted the core of skilled tertiary professions, and the conviction was firmly established that they would remain out of reach.
Also to read:
Why AI is forcing companies to rethink the value of work
This conviction was based on a conceptual confusion that must be cleared up first and foremost. It was not the job of a lawyer or financial analyst that was protected, but a set of specific tasks that made up this job and which had until now resisted automation. The distinction between these three levels is fundamental.
A job designates a position held within an organization, with a contract, a salary, and a job description. A skill is a cognitive or technical ability that can be applied in various professional contexts. A task is a specific, definable action, for which it is possible to assess whether or not it can be automated at a given cost. It is at this third level that the ongoing transformation truly takes place, and it is precisely this level that the public debate ignores.
Rupture in the long history of industrial capitalism
Generative AI represents a breakthrough in this long history. For the first time since industrialization, qualified cognitive tasks such as writing, document analysis, synthesis, and production of first drafts are directly exposed.Eloundou, Manning, Mishkin and Rockestimate that about 80% of the active U.S. workforce could see at least 10% of their tasks affected by large language models, and that this exposure increases with salary level. This is the exact opposite pattern observed in all previous waves.
The analytical framework developed byAcemoglu and Restrepoallows to go further. Their model distinguishes two opposing effects produced by any wave of automation:
-
The displacement effect, first: workers lose tasks to the benefit of the machine, which mechanically reduces the demand for labor and weighs on the wages of the affected groups;
-
The reintegration effect, then: automation produces new tasks where human value is decisive, generating compensatory demand.
The long history of industrial capitalism can be read as a succession of these two effects, the second generally ending up compensating for the first.
The case of translation allows us to see very concretely how displacement and reintegration combine. Generative AI can produce a first draft in another language in a few seconds, which shifts part of the work previously done by human translators to the machine. But this automation simultaneously reintegrates other tasks or enhances their importance, such as checking for misunderstandings, adapting to the cultural context, harmonizing terminology, quality control, and final validation.
Potential imbalance
What is worrying with generative AI is the potential imbalance between these two dynamics. The shift is happening at a speed that labor markets and training institutions struggle to absorb, while reintegration still largely remains to be built.
However, the most important phenomenon is not sectoral, but it is internal to the professions themselves. In its“Employment Outlook”, the OECDhighlights that the professions most exposed to generative AI are precisely those with a high cognitive density: finance, law, consulting, higher education. Unlike previous waves that affected rural areas and industrial regions, the exposure is now stronger in large metropolitan areas and among highly skilled workers, an unprecedented geographical and social reversal.
Redistribute tasks
This reversal concretely takes place at the level of the task.
In the same position of financial analyst or legal advisor, some tasks are shifting to AI (producing an executive summary, generating an initial contract analysis, synthesizing a literature review), while others are mechanically gaining value: defining the relevant analytical framework, assessing the quality of automated reasoning, detecting factual errors in an output, assuming legal or ethical responsibility for a decision. These are not jobs that disappear. They are bundles of tasks that are redistributed between humans and machines, transforming from within what an employer expects from a qualified employee.
This redistribution of tasks has a direct impact on the skills that will truly be valued in the coming years, and it overturns some of the usual assumptions about professional training.
Train workers to use AI instrumentally, to master a tool, to writepromptsEffective, mastering an interface is useful in the short term, but it is insufficient if the skill truly required tomorrow is not to produce with AI, but to supervise and critique what it produces.
A training challenge
However, effectively supervising an AI output requires exactly what short and technical trainings struggle to develop: a solid general knowledge that allows detecting a fundamental error, an argumentative ability to evaluate the coherence of a reasoning, a knowledge of cognitive biases to identify the blind spots of an automated analysis. These are skills thateducational sciences group under the term metacompetencesTo learn to learn, to exercise critical judgment, to mobilize knowledge in unprecedented situations.
The paradox then becomes the following. As AI automates routine knowledge tasks, it precisely values what generalist training and humanities courses have long cultivated and what debates on employability have tended to disregard in favor of more immediately measurable technical skills.
Not out of nostalgia for the humanities, but out of pure economic logic. If the machine produces the text, the analysis and the synthesis, the marginal value of the human lies in their ability to judge whether this text is true, whether this analysis is relevant in light of the real context, whether this synthesis serves the pursued objective.
Hugo Spring-Ragain does not work for, advise, own shares in, or receive funds from any organization that could benefit from this article, and has declared no other affiliations than his research institution.
–ref. Generative AI, the first cognitive revolution in the history of work –https://theconversation.com/lia-generative-premiere-revolution-cognitive-de-lhistoire-du-travail-279911