AI can expand the capabilities of middle-skilled workers, with important implications for vocational training and the labor market.
Debates about generative artificial intelligence (AI) often focus on the jobs it might replace. Yet AI could also broaden the range of tasks workers can perform. For workers with vocational qualifications, this could mean access to work previously requiring more advanced training and potentially better career opportunities. But can AI actually help them take on these more demanding tasks?
In a recent study, we investigated this question through a field experiment with 673 IT apprentices at twelve German vocational schools. All were in their final year of training, specializing in either application development or system integration. We randomly assigned them to complete tasks in their field either with access to ChatGPT Pro or without AI and without internet access.
The key feature of the experiment was the choice of tasks. One fell within the apprenticeship curriculum. Two others went beyond it, requiring skills associated with software or network engineering. This design allowed us to test whether AI helps apprentices perform familiar work better and tackle work beyond their formal training.
Our results show consistent productivity gains from AI access, both within and beyond the apprentices’ formal training. On the task within the curriculum, access to AI raised performance by 44 percentage points. Crucially, performance also improved by 26–31 percentage points on the tasks that exceeded the scope of the apprenticeship curriculum. AI thus enabled apprentices to successfully tackle work for which they had not been formally trained. Gains tended to be larger on tasks where performance without AI was lower.
Could these improvements come at the expense of learning? After each task, we tested apprentices’ understanding of the material without AI assistance. We found no evidence that AI reduced immediate comprehension. Instead, understanding improved on average. Whether these benefits persist over time remains an open question.
The benefits of AI were not systematically related to apprentices’ prior educational qualifications. They were, however, generally larger among those reporting greater AI literacy, particularly the ability to use AI effectively. This suggests that practical familiarity with AI may matter more for benefiting from these tools than conventional educational credentials alone.
This finding raises an important question for vocational education. Although AI use is already widespread among apprentices, only 16 percent in our sample reported learning how to use it at vocational school. Most develop these skills independently. Unequal access to opportunities to learn about AI could therefore create differences in performance among workers with otherwise similar qualifications.
Our experiment does not establish whether better task performance translates into higher wages or more employment. It does, however, show how AI could broaden the opportunities available to vocationally trained workers. Integrating practical AI literacy into vocational education is a promising way to help more workers benefit from this potential. Such training should equip them to use AI effectively and critically assess its output, while preserving the occupational expertise needed to judge the results.
© Hanna Brosch, Christina Langer, Philipp Lergetporer, Sandra Mueller, Harald Pfeifer, and Nele Tschoepe
Hanna Brosch is a Postdoctoral Researcher at the University of Erlangen-Nuremberg, Germany
Christina Langer is a Postdoctoral Researcher at the ifo Institute, Senior Economist at the Stepstone Group, and a Digital Fellow at the Stanford Digital Economy Lab, Germany
Philipp Lergetporer is an Assistant Professor of Economics at the Technical University of Munich, Campus Heilbronn, Germany, and IZA@LISER Research Fellow
Sandra Mueller is a Researcher at the Federal Institute for Vocational Education and Training (BIBB) in Bonn, Germany
Harald Pfeifer heads the “Economics of VET” section at the BIBB and is Professor at ROA, Maastricht University, Netherlands, and IZA@LISER Research Fellow
Nele Tschoepe is a Doctoral Researcher at the “Economics of VET” section at the BIBB in Bonn, Germany
Please note:
We recognize that World of Labour articles may prompt discussion and possibly controversy. Opinion pieces, such as the one above, capture ideas and debates concisely, and anchor them with real-world examples. Opinions stated here do not necessarily reflect those of the LISER.
Related World of Labour content:
https://wol.iza.org/articles/artificial-intelligence-and-labor-market-outcomes by Nick Drydakis
https://wol.iza.org/articles/innovation-and-employment-in-the-era-of-artificial-intelligence by Marco Vivarelli and Guillermo Arenas Díaz
https://wol.iza.org/articles/evaluating-apprenticeship-training-programs-for-firms by Samuel Muehlemann and Harald Pfeifer