A review of 173 international studies has found no simple answer to whether generative artificial intelligence improves or damages students' critical thinking. The researchers concluded that the same tools can support deeper reasoning when used to question and compare ideas, yet encourage dependence and superficial work when they merely deliver finished answers.
The review examined research published from 2022 through spring 2025 on AI use in schools and higher education. Hilde Storrøsæter, a doctoral researcher at the Norwegian University of Science and Technology's Department of Teacher Education, worked with researchers from Czech universities to assess effects on critical thinking and independent problem-solving.
Their central finding was a dual effect shaped by pedagogy. AI can help learners test an argument, explore competing viewpoints and develop their own reasoning. When it replaces those steps, however, students lose opportunities to practise the cognitive work the assignment was meant to develop. The researchers therefore place more emphasis on how a tool is used than on the presence of AI itself.
The evidence base also has major limits. Eighty per cent of participants in the reviewed studies were university or college students, leaving much less information about children whose reasoning skills are still developing. Many studies relied on participants describing their own experience rather than objective measurements over time. Research was also unevenly distributed geographically, with parts of Africa, South America and Central Asia especially underrepresented.
Those gaps matter as younger people change how they look for information. A Norwegian Media Authority survey cited in the report found that Google search use among people aged nine to 18 fell from 72 per cent in 2024 to 47 per cent in 2026. It also found that 39 per cent of Norwegian 11- and 12-year-olds use AI. Meanwhile, visits to the youth information service ung.no dropped from 24 million in 2024 to 18 million in 2025, a change attributed partly to AI tools and Google's generated summaries.
The review does not establish that AI alone caused those shifts, nor does it support treating results for adult students as evidence of effects on a 10- or 14-year-old. Instead, it identifies questions that schools need longer-term and age-specific research to answer.
For educators, the authors argue that AI literacy must include understanding which parts of a task students should still perform themselves. Teacher training has two linked responsibilities: helping teachers evaluate and use the technology critically, and enabling them to design lessons where pupils do not surrender the thinking process. Storrøsæter supports restraint with very young pupils and says classroom use should occur on teachers' terms. The review's practical message is that assignments, supervision and source criticism—not the tool in isolation—will determine whether AI becomes a learning aid or an intellectual shortcut.
