We're not bound to be replaced with machines, however "intelligent" these may turn out.
The two professors at Essec Business School, Guillaume Chevillon and Julien Malaurent, specialists in artificial intelligence, explain in an op-ed for "Le Monde" that algorithmic replacement will not happen, because AI is effective at replacing specific tasks, but not entire jobs.
Will we be replaced by machines as ChatGPT and other generative artificial intelligence (AI) systems are deployed? The first announcements of massive layoffs, "officially" justified by the lower cost of algorithmic solutions, are rightly worrying: is technological progress making us relive, a hundred years later, the desertification of the countryside, with today's employees echoing the peasants of the past?
Part of the answer, unfortunately, seems hidden in the question: yes, technological transformations modify the structure of work, and a poorly managed transition can prove dramatic for the generations (and ecosystems) that suffer it. However, it can also be accompanied by very strong social progress, a reduction in poverty and inequalities, and an enhancement of individual freedoms and leisure. Who would really wish that our economic and social structure was still that of pre-WW2? Sixty percent of today's jobs did not exist in 1945, and 85% of employment growth is linked to technological advances.
We are not doomed to be replaced by machines, no matter how "intelligent" they may be. A study from the University of Pennsylvania and OpenAI has shown that although 80% of jobs can be affected by the deployment of generative AIs, large language models, or image generation systems, in reality, within these jobs, on average only 10% of the tasks performed are concerned.
Most professions will only undergo a simple adaptation, potentially very positive: according to an experiment by Harvard Business School on consultants from the Boston Consulting Group, employees can, thanks to generative AI, accomplish 12% more tasks, 25% faster, with a 40% improvement in quality. These results are not limited to so-called "cognitive" workers.
A recent study by the Organization for Economic Cooperation and Development (OECD), to which the Metalab of Essec contributed for the French part, showed that all sectors are affected by AI, including the manufacturing sector: assembly line operators, for example, greatly benefit in their maintenance work from the quality of the failure predictions provided by AI.
A very short-sighted, or "myopic," company might conclude that it is possible, thanks to generative AI, to do better with fewer people, and therefore proceed with layoffs.
This purely "quantitative" approach to work leads it to be deceived on many fronts. Not only does it then not benefit from the "qualitative" improvement related to generative AI, which its competitors (or new players) will not hesitate to exploit to take market share, but in addition, the myopic company does not perceive the iceberg toward which its pilots are directing it: generative AI is not reliable, as its recommendations are random.
Our Irreplaceable Intuition
Users of ChatGPT already know well that the same request never yields the same response. Therefore, supervision by a human expert is necessary to avoid the "hallucinations" of generative AI, a term used for the pure and simple inventions it often provides: a commonly cited example is that of a lawyer who had his pleading written by generative AI only to realize, but far too late, that this algorithm had "simply" invented the jurisprudence it relied on.
Regulation, such as the AI Act presented this summer to the European Parliament, is useful here to encourage the transparency of algorithms and ensure the understanding of their recommendations.
Despite the upcoming advances in AI, human supervision will still be necessary for a very long time, if not always. The positive results of the Harvard Business School study actually correspond to the situation where employees had access to experts who knew the limits of generative AI. Without this human help, the tasks performed were of 20% lesser quality. The day when AI will know its own limits exactly and be able to explain them to us is not coming soon.
The myopic company that thinks it can massively part with its employees will therefore be unable to understand and control the work of generative AI. Let us never forget that, with the departure of the teams that worked on the Apollo projects, NASA lost the capacity to send humans to the Moon and it took about twenty years to rebuild the necessary knowledge! The French nuclear sector is another recent example of a loss of technical capabilities linked to the reduction of investment in people.
It is therefore indeed a matter of AI complementing human work, not substituting for it. We must rely on the unique qualities of humans to strengthen our training curricula and promote social progress through generative AI. The OECD study suggests that it is our behavioral skills, our intuition, and our ability to interpret that make us indispensable and irreplaceable. We cannot just focus on enhancing the technical knowledge of personnel but must also help them to shape the supervision of AI by humans within professions.
What will then be the impact of generative AI on work? The continuation of the machine-human hybridization that we have known since the beginnings of the industrial era, or even since the beginnings of humanity. The studies we mention have indeed shown that certain jobs will be very strongly affected: undoubtedly, 20% of jobs will see half of their tasks potentially replaceable by AI.
Companies will therefore see the emergence of what the authors of the studies call "centaurs," who divide their tasks into equal and distinct parts of human and algorithmic work, or "cyborgs," who integrate a fraction of generative AI into all their professional activities, so that the human contribution cannot be separated from that of the machine.
Let's remember the bursting of the financial bubble in the early 2000s: it made the first generation of companies that symbolized the Internet disappear. Today too, companies must invest wisely in AI to ensure their longevity. Beware of those that do not prioritize the improvement and development of their human capital, generating a virtuous circle of increasingly relevant uses of generative AI, but which, on the contrary, replace the competencies and expertise maturely acquired by tools they do not truly master.
It is not jobs that generative AI can replace, but specific tasks that it can improve. We all, citizens and public authorities, must remain vigilant so that companies do not adopt a "myopic" behavior of investments in AI seeking to replace humans, which would penalize the whole society, but maintain a long-term positive vision to improve the quality of jobs and the fulfillment of everyone at work.
(translated by ChatGPT4 with some human adjustments)
References: NEW FRONTIERS: THE ORIGINS AND CONTENT OF NEW WORK, 1940–2018 David Autor, Caroline Chin, Anna M. Salomons, Bryan Seegmiller, NBER WORKING PAPER SERIES, Working Paper 30389, août 2022.
GPTs are GPTs: An Early Look at the Labor Market Impact Potential of Large Language Models, Tyna Eloundou, Sam Manning, Pamela Mishkin, and Daniel Rock. 22 août 2023.
Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of AI on Knowledge Worker Productivity and Quality. Harvard Business School Technology & Operations Mgt. Unit Working Paper No. 24-013, 18 septembre 2023.
The Impact of AI on the Workplace: Evidence from OECD Case Studies of AI Implementation. OECD SOCIAL, EMPLOYMENT AND MIGRATION WORKING PAPERS No. 288, 24 Avril 2023.