Study: AI Assistance Reduces Cognitive Persistence and Independent Performance

A research study submitted to arXiv on April 6, 2026, suggests that while artificial intelligence can boost immediate productivity, it may cause significant long-term damage to human cognitive persistence and independent performance.

The paper, titled AI Assistance Reduces Persistence and Hurts Independent Performance, was authored by Grace Liu, Brian Christian, Tsvetomira Dumbalska, Michiel A. Bakker, and Rachit Dubey. The researchers found that AI assistance provides a short-term performance gain but introduces a heavy cognitive cost that manifests when the technology is no longer available.

According to the findings, users can become dependent on AI tools after as little as 10 minutes of assisted problem-solving. Once the AI is removed, these individuals performed worse and were more likely to give up on tasks compared to those who had worked independently from the start.

Experimental Methodology and Results

The researchers conducted a series of randomized controlled trials involving a total of 1,222 participants. The study focused on reasoning-intensive cognitive labor, which includes common AI use cases such as coding, writing, and brainstorming.

In the first experiment, 350 Americans were tasked with solving fraction-based equations. Half of the participants were given access to a specialized chatbot built on OpenAI’s GPT-5, while the other half worked without assistance. Halfway through the exam, the researchers revoked AI access for the assisted group.

The result was a steep decline in the number of correct answers provided by the group that had previously used the AI. Many of these participants simply stopped attempting the problems.

This pattern was replicated in a second, larger experiment with 670 people. A third experiment focusing on reading comprehension questions, rather than mathematics, yielded the same results: a drop in both performance and perseverance following the removal of AI assistance.

Once the AI is taken away from people, it’s not that people are just giving wrong answers. They’re also not willing to try without AI. People’s persistence drops.

Rachit Dubey, assistant professor at the University of California and study coauthor

The Erosion of Persistence

The authors argue that current AI systems act as short-sighted collaborators because they are optimized to provide instant, complete responses without requiring the user to struggle through the process. This differs from human mentors who scaffold learning by prioritizing a student’s growth over immediate results.

The study posits that AI conditions users to expect immediate answers, which denies them the necessary experience of working through challenges independently. Because persistence is a primary predictor of long-term learning and skill acquisition, this dependency is viewed as a critical barrier to competence.

The researchers compared the effect of sustained AI use to the boiling frog analogy, suggesting that the erosion of motivation and persistence happens gradually. By the time the cognitive decline becomes visible, the authors suggest the effects may be difficult to reverse.

Implications for Education and Innovation

Rachit Dubey warned that the rapid integration of AI into the education sector could create a generation of learners who are unaware of their own capabilities. He stated that this trend could eventually dilute human innovation and creativity.

The study did identify one positive interaction method. Participants who used AI tools for hints and clarification, rather than using the bot to generate direct answers, had a significantly easier time continuing their work after the AI was removed.

This suggests that AI development may need to shift toward prioritizing the scaffolding of long-term competence rather than focusing solely on immediate task completion.

Broader Cognitive Trends

This research aligns with other recent observations regarding the impact of AI on the human brain. Some reports have highlighted a phenomenon described as AI brain fry among heavy users of the technology.

In the professional sphere, research has indicated that employees utilizing AI may experience intensified workloads compared to those who do not use the tools. In education, other studies have linked AI reliance in schools to lower test scores among children who depend on chatbots.

The researchers noted that the study on persistence has not yet undergone peer review.