Study: How AI Use Worsens Problem-Solving Skills

Having AI assistants complete tasks directly can negatively impact problem-solving skills and mental stamina. This is according to a study by a research team from the Universities of Oxford, MIT, UCLA, and Carnegie Mellon. The study found that in trials, the effect was evident after short work sessions with AI of around 10 minutes: as soon as the AI was withdrawn, participants were significantly more likely to give up on tasks and achieve poorer results than individuals in a control group who were not allowed to use AI from the outset.

However, the researchers qualify that it depends on how the AI is used: The deteriorations occurred when participants had their task solutions generated directly. When AI was used for assistance or to clarify questions, no disadvantage was observed. Nevertheless, at 61 percent, the majority of participants preferred the direct AI solution.

What if Chatty suddenly disappears?

The research team investigated the effects on both mathematical problem-solving skills and reading comprehension. In the first experiment, 354 people were randomly assigned to either the group with initial AI use or the control group without AI. Initially, they had to solve 12 fraction problems, with the AI group able to use GPT-5 as an assistant.

This was followed by three more tasks where the AI was suddenly switched off. The previously higher solution rate of the AI group plummeted significantly during this phase, with an average of 57 percent clearly below that of the control group (73 percent). The rate of skipped tasks was also significantly higher for the AI group at 20 percent compared to the control group's 11 percent.

A second round of trials with 667 participants essentially confirmed the effect that the AI group's performance declined after AI withdrawal and the dropout rate increased. The solution rate without AI was then 71 percent, below the control group's 77 percent.

A third trial with 201 participants, focusing on reading comprehension, also led to the same effect: Without AI, the AI group performed worse at 76 percent compared to the control group (89 percent) and abandoned tasks more frequently at 8 percent (control group 1 percent). The researchers conclude from this that the observed negative effects of AI assistance are not limited to specific tasks but are a general consequence of AI-supported problem-solving.

Support or Undermine?

According to the researchers, this raises urgent questions about the impact of daily AI use on users' mental abilities. They warn that systems optimized for short-term utility could undermine the very human capabilities they are intended to support. Other studies have already suggested that AI use could impair critical thinking skills.

(axk)