Surveys completed by 25,000 employees and 7,000 companies in the years 2023-2024, combined with pay and hour -hour data, did not show statistically significant influence of AI chatbots on earnings or overworked time in any of the eleven “particularly vulnerable” professions – from accounting, through journalism, to programming. The average time savings provided by artificial intelligence amounted to only 2.8-3 percent. – equivalent to a few hours a month.
What's more, this minor improvement in productivity has been almost completely consumed by new duties generated by AI tools themselves. According to the authors of the study “Large Language Models, Small Labor Market Effects” from the University of Copenhagen, 8.4 percent. employees began perform tasks that did not exist before. From writing sophisticated prompts, to auditing the quality of the AI model answers. The authors give an example of teachers who today spend additional hours detecting plagiarism created by chatgpt.
We are dealing with a paradox. Automation was to free time, and in practice we have “more work thanks to automation”. This problem is also clearly visible outside of Denmark. In experiments conducted in the USA or Germany, IT specialists reported that browsing and debugging the code generated by the AI model can be slower than writing it yourself – especially when you need to keep security standards or compliance with the internal knowledge base and organization's policy.
Check also: Each of us will be the head of AI employees. Change comparable to the arrival of the Internet
No work by AI? It doesn't seem to
Looking globally, the forecasts also do not indicate a wave of mass unemployment, though They predict a large rotation. The World Economic Forum estimates that as much as 23 percent will change by 2027. positions, with approximately 69 million new jobs, and 83 million disappear – net is a decrease of 2 %. employment, but not a revolution.
The international work organization goes further, separating “automation” from “work extension”. In high -way countries Only 5.5 percent full -time jobs are fully at risk of replacing, while for 13.4 percent. Generative AI will become a tool expanding competencesnot an eliminating position.
This does not mean a lack of social tensions. The Pew Research Center study conducted in the USA just before the premiere of the GPT -5 shows a giant perceptual gap: 73 percent. AI experts believe that technology will improve the way of working in the horizon of 20 years, but the same optimism shares only 23 percent. adult Americans. Fears drive, among others Risk of algorithmic supervision, blurring responsibility and pressure on continuous availability.
On the other hand, the benefits are easily noticeable: rapid translations between 100 languages, shortening the path from the idea to the first sketch of the report, democratization of access to expert knowledge or immediate personalization of training materials. Danish data suggest that where the employer actively supports AI training, a declared increase in satisfaction and efficiency is 10-40 percent. higher than in companies they don't do.
Read also: Building start-ups using the “to effect” method. This is how failures change into millions of successes
Many roles to redesign
The implementation of the algorithm is currently becoming a big challenge as redesigning processes and roles: determining who assesses the credibility of the model's response, who manages training data and who will be responsible for any mistakes. In practice, we observe a shift of competence towards meta -tabs – from prompt engineering, through validation of content, to ethical judgment. These activities are difficult to include in traditional performance measures, but more and more often included in the descriptions of positions.
The application from the Danish experiment is as follows: generative AI can speed up some of the tasks, but at the same time opens a long tail of duties that require human supervision and interpretation. Productivity is growing more slowly than suggesting marketing presentations, and wages are still indifferent to an algorithmic revolution.
That is why labor market policy – from training systems to responsibility regulations – should focus on facilitating smooth passage of employees between old and new meta -subjects. In this sense, AI does not take the job so much as it redefines its structure, leaving economics (so far) amazingly intact. However, new competences must be obtained – there is no escape from this escape.
Author: Grzegorz Kubera, Business Insider Polska journalist
