AI can speed up some tasks but its impact on learning new skills is not well understood.

A new study investigated the effects of AI use on learning new computer programming skills.

AI use significantly reduces new skill formation.

AI use can speed up many tasks; for example, analyzing complex data. However, using AI to complete a task at work or for a hobby could lead to so-called cognitive offloading. Cognitive offloading means that someone who uses AI to do a task is not as mentally engaged in a task as someone who completes the task all by themselves. While this may not matter too much for the outcome of the task, cognitive offloading may be problematic when it comes to learning something new. When a task is performed with the aim of learning it, offloading it to AI may lead to worse learning success as opposed to conducting the task 100% with their own brain. However, psychological research on whether or not cognitive offloading has negative effects on learning success is sparse so far.

A new study on how AI use affects skill formation when learning how to code

A new study investigated how cognitive offloading due to AI use affects the long-term formation of coding skills (Shen and Tamkin, 2026). The study, entitled “How AI Impacts Skill Formation,” was published on the preprint platform arXiv. As it is a preprint that has not undergone peer review by expert scientists yet, the results should be considered preliminary.

In the study, 52 volunteers participated in two groups: a control group and a treatment group. All participants were either professional or freelance computer programmers. Both groups had to learn a complex computer programming skill and complete various computer programming tasks. The study was divided into three parts. First, both groups had to do a warm-up programming task without AI use. Then they had to program the main task. For this task, the treatment group was allowed to use AI for programming, while the control group was not allowed to use any AI to complete the task. Afterwards, both groups needed to answer a quiz about the software tool they were using to complete the task to find out how much they learned about it.

What were the results of the study?

The results of the study were quite striking. While the group that was allowed to use AI was not faster than the group that did not use AI, they performed much worse in the quiz. The average score of the AI group was 17% lower than that of the control group, a statistically significant difference. A subgroup analysis revealed that this effect was found for beginner, intermediate, and expert programmers alike. Thus, it did not depend on the experience level of the participant. A further analysis revealed that study participants who wholly relied on AI to generate code scored particularly badly in the quiz. Participants who asked the AI to generate code and also provide explanations about the generated code performed better.

Takeaway: Relying heavily on AI dramatically reduces skill formation

The results of the study have major implications for the psychological debate about using AI in schools and universities. They imply that relying heavily on AI while learning a complex task like computer coding comes at a major cost: it considerably reduces the learning of new skills.

Learners who rely on AI simply do not learn as much as those who do not. While being able to use AI tools proficiently will likely be a major skill in the job market of tomorrow, the finding suggests that students should first learn basic skills on their own and AI tools should only be introduced into the learning process once a good level of proficiency has been reached.

Shen, J. H., & Tamkin, A. (2026). How AI Impacts Skill Formation. arXiv preprint arXiv:2601.20245.

© Psychology Today