Using AI for Just 10 Minutes Erodes Your Ability to Persist at Hard Tasks

Spending just 10 minutes relying on an artificial intelligence tool significantly impairs a user’s ability to focus and persist with difficult tasks, according to a peer-reviewed study presented October 7, 2026, at the Conference on Language Modeling. Researchers from UC Berkeley, Carnegie Mellon University, the Massachusetts Institute of Technology, the University of Oxford, and UCLA found that when AI assistance is abruptly removed, participants rapidly lose accuracy and frequently give up entirely, highlighting a sharp cognitive cost to offloading complex problem-solving onto chatbots.

Brian Christian, a research fellow with the UC Berkeley Center for Human-Compatible AI and co-author of the study, noted that the findings provide empirical backing for a phenomenon many scholars and professionals already sense in daily practice. It’s a striking finding, but in a way it supplies ammunition for a story that a lot of us kind of feel in our gut, Christian said, as reported by vcresearch.berkeley.edu. Christian has spent two decades examining how artificial intelligence alters human cognition and authored the 2020 book The Alignment Problem.

Randomized Controlled Trials Test Cognitive Persistence at UC Berkeley

To measure how quickly human focus and persistence decline, the research team designed a series of randomized controlled trials involving 1,222 online participants. In the initial experiment featuring 354 participants, one cohort solved 15 basic fraction problems independently, while a second cohort had access to ChatGPT for assistance and direct answers. Initially, the group with AI access achieved higher accuracy than the unassisted group.

However, after 12 problems, researchers removed the AI tool from the experimental group. Almost immediately, those participants stopped solving questions accurately. A second, larger experiment with 667 participants yielded identical results: when AI assistance was withdrawn, those users provided incorrect answers or abandoned the tasks entirely, whereas the control group persevered and finished with higher success rates.

SAT Reading Comprehension Trials Confirm Focus Drop-Off

Researchers then conducted a third experiment with 201 participants using an SAT reading comprehension prompt to determine whether the fraction findings extended to language tasks. The outcome mirrored the earlier math tests. Removing the AI tool triggered an immediate drop in both participant persistence and reading comprehension accuracy.

Data visualizations from the experiments illustrated a steep decline in solve rates for the AI-assisted group once the tool was taken away, accompanied by a sharp increase in the frequency with which participants skipped questions altogether.

Erosion of Productive Struggle in Research and Academia

The study demonstrates that overreliance on automated answers erodes cognitive fitness by eliminating what the research team defines as productive struggle, the mental friction necessary for individuals to build subject mastery and discover their own research passions. When LLMs generate answers in seconds, human users adapt to that velocity, making any task that requires sustained attention feel inefficient.

Christian expressed concern that widespread AI integration in science and academia could compromise the foundational traits of rigorous scholarship, such as staying immersed in raw data and working through complex problems collaboratively. This is something that is happening to human knowledge and human expertise from top to bottom and from grade-school students all the way to the leading experts in the world.

Frontier AI Adjustments Remain Unresolved

While frontier artificial intelligence companies have acknowledged that offloading tasks to chatbots diminishes baseline subject mastery, developers have yet to establish standardized countermeasures. Christian suggested that future AI systems should avoid defaulting to instant answers and instead function more like instructional tutors, designed to augment rather than supplant human capabilities.

The peer-reviewed paper follows preliminary drafts that circulated earlier in the spring and drew coverage from outlets including WIRED, Psychology Today, and The New York Times. Whether AI developers will successfully modify model architectures to preserve human cognitive stamina remains an open question as adoption expands across global academic and research institutions.