Technology is advancing too quickly for employees to adopt it: for the moment, very few tasks are actually performed using AI.

There's theory... and then there's practice.

In early March, the company Anthropic published a groundbreaking report on the effects of artificial intelligence on employment, based on data from 2 million real conversations with Claude, its generative AI model. The novelty lies in the methodology: like previous studies published to date, the company shows the proportion of a job that could theoretically be replaced by artificial intelligence, but it also compares this with the actual exposure observed. The blue areas, therefore, show, by profession, the proportion of tasks in a given job that could theoretically be performed (or accelerated) by artificial intelligence. In red, we see the proportion of tasks actually completed thanks to data from Claude, Anthropic's AI.

Thus, we observe a significant gap between what could potentially be done with AI assistance and the reality of its use by workers.

Slow-to-Diffuse Technologies

Even though IT and mathematics professions have the highest actual adoption rate (33%), the gap remains wide for the time being compared to theoretical exposure (94%).

There are also large gaps between potential and reality in the media, art, and social sciences (we can assume there is reluctance among those involved), and discrepancies are also visible in legal and management professions.

To explain this gap, the study points out that "some theoretically possible tasks may not not be implemented due to limitations of the model." But that's not the only explanation.

"Others may be slow to spread due to legal constraints, specific software requirements, human verification steps, or other obstacles," the authors add.

An organization moves at the pace of the slowest

This mainly means that technological capabilities advance much faster than their adoption. However, according to AI sociologist Yann Ferguson, employee adoption is crucial to seeing productivity gains. "Giving a high-performance AI to employees who don't know how to use it is like giving a Ferrari to people who don't have a license to drive on a small country road," he explained to BFM Business in this article. Furthermore, according to figures from Benjamin Tannenbaum, a French AI specialist and co-founder of the company Aiso, the vast majority of the world's population (84%) has never used a chatbot. Anthropic, however, envisions a future in which the reality of AI use will reach its full potential. "As capabilities advance, adoption becomes widespread, and deployment intensifies, the red zone will eventually overtake the blue zone," the report predicts. "Things are very much based on beliefs." But according to Yann Ferguson, the companies behind AI tools have every incentive to fuel the narrative of rapid and irreversible adoption. "Especially in the field of AI, things are very much based on beliefs. Major tech providers have an interest in communicating exponential growth curves so that companies invest in their tools," he explains. Anthropic's study also sought to determine whether the emergence of AI had led to an increase in the unemployment rate in the most replaceable jobs, which does not appear to be the case. "We have not seen any systematic increase in unemployment among highly exposed workers since the end of 2022," the report states. Nevertheless, the authors found "indications suggesting a slowdown in the hiring of young workers in the exposed professions." Finally, a whole host of jobs are not affected by the arrival of artificial intelligence. These include manual agricultural work such as pruning trees and operating farm machinery, and legal and human services tasks such as representing clients in court...