Researchers warn AI overuse could weaken students’ ability to think independently

As artificial intelligence becomes increasingly common in classrooms, Stanford University researchers are examining how much thinking students should delegate to these tools without compromising the learning process.

In a report published on Nov 4, Stanford researchers highlighted the risks of cognitive offloading — the practice of using external tools to reduce the mental effort required to complete a task. While tools such as calculators have long helped people handle routine work, researchers say generative AI can take over entire tasks, including those that require creativity, problem-solving and active engagement.

“If I can completely delegate a task to AI and I never spend time thinking about it, then there’s no learning,” said Guilherme Lichand, assistant professor at Stanford Graduate School of Education.

The researchers cautioned that when students rely on AI to perform core learning activities, they may find it harder to apply their knowledge or complete similar tasks independently after losing access to the technology.

AI use linked to weaker performance without the tool

To examine how cognitive offloading affects students, Lichand and predoctoral fellow Sachin Allums studied Brazilian middle-school students who were divided into four groups with different levels of access to an AI tool.

The students completed two creative tasks. Those who used AI for the first task initially benefited from the assistance, but the advantage disappeared when the tool was withdrawn for the second task.

More concerningly, students who had used AI more extensively performed worse on the subsequent creative task than students who had never used the tool.

Lichand said the negative effects were concentrated among students who relied more heavily on AI during the first task, suggesting that greater dependence on the technology could make it harder to perform independently later.

The findings point to a distinction between using AI to support learning and allowing it to replace the mental effort involved in learning.

Review of AI education research highlights concerns

Stanford researchers also pointed to a wider review of more than 800 papers on AI in education conducted by Lily Fesler and her team. The researchers focused on 20 high-quality causal studies examining the effects of AI use on learning.

Although the evidence remains limited, the studies reviewed indicated that cognitive offloading could affect students' ability to recall information and perform tasks without AI assistance.

Fesler, a senior research associate at the AI Hub for Education within Stanford's SCALE Initiative, referred to research in which students using generative AI to learn new information showed lower recall and brain activity than learners who did not use the technology.

The review also raised concerns about general-purpose chatbots used for exam preparation. Students who rely on these tools may see their test performance decline when they no longer have access to them.

Fesler said AI could improve or harm learning outcomes depending on the type of tool and how it is used.

Researchers see potential in AI-assisted learning

The Stanford researchers also highlighted ways AI could make learning more interactive by allowing students to create, test and refine their own ideas.

Hariharan Subramonyam, assistant professor at Stanford Graduate School of Education, said AI tools could support a more creation-centred approach to education. Students could use them to visualise concepts, build models and experiment with different ideas rather than relying only on reading material or memorising information.

In his courses, Subramonyam asked students to plan their designs before using AI models to create simulations. He said students valued the process of working together, experimenting and learning through trial and error.

The researchers also pointed to AI tools that guide students through questions rather than immediately providing answers. Such tools can encourage learners to think through problems themselves.

Fesler said AI assistants could provide more frequent feedback on student essays, provided they do not write the assignments or entirely replace feedback from teachers.

The researchers said schools and educators need to consider which parts of the learning process can be supported by AI and which require students' direct participation.

They emphasised the importance of preserving the effort, repetition and resilience involved in understanding new concepts. These experiences help students develop the ability to transfer previously acquired knowledge and skills to unfamiliar situations.

Fesler called for more high-quality studies to understand the effects of AI on students and teachers. The researchers said the technology has advanced faster than the evidence about its educational impact.

Lichand also cautioned against rushing to introduce AI tools simply because their adoption appears inevitable. The researchers called for deliberate implementation and safeguards that allow students to benefit from AI without surrendering the thinking and problem-solving processes central to learning.