Artificial intelligence should be completely banned from education to protect students’ ability to think for themselves, according to Japanese language and brain researcher Kuniyoshi Sakai.
Sakai, a professor at the University of Tokyo, makes the case in an opinion article for Nikkei Asia, arguing that schools are adopting digital textbooks and generative AI without sufficiently considering their effects on fundamental learning skills.
“A complete ban on AI in education is the only way,” Sakai writes.
His intervention comes as Japan moves in the opposite direction. Parliament approved changes to education legislation in June that allow textbooks incorporating digital formats to receive the same formal status as traditional textbooks.
Japan’s education ministry says the aim is not to replace all printed books, but to combine the strengths of paper and digital material.
Japan is also running generative AI pilot programmes in schools and has issued guidelines promoting what it calls human-centred use of the technology rather than a blanket prohibition.
One in five students reported reading nothing
Sakai bases part of his warning on a 2025 survey of 1,062 Japanese university students aged between 18 and 29.
According to the survey, 20 percent said they did not normally read books, newspapers or magazines at all. The definition included textbooks, manga and certain online articles.
Another 10 percent said they did not take notes during lectures or in similar situations.
Sakai says students who had neither reading nor writing habits performed extremely poorly on a high school-level reading comprehension test.
He argues that reduced exposure to long texts and handwritten note-taking may weaken reading comprehension and make it harder for students to understand context and the structure of longer arguments.
The survey as described in his article shows a relationship between such habits and reading performance, however, rather than establishing that digital devices themselves caused the weaker results.
Sakai is also critical of digital textbooks. He argues that physical books provide spatial cues that help readers remember where information appears and understand how different parts of a text relate to one another.
AI raises a different problem
His strongest criticism is reserved for generative AI.
Sakai argues that allowing students to outsource brainstorming, writing and editing risks depriving them of the practice needed to develop those abilities themselves.
A similar concern has led the University of California, Berkeley School of Law to introduce unusually strict rules.
Since summer 2026, Berkeley Law’s default policy has prohibited students from using AI to conceptualise, outline, draft, revise, edit or translate work submitted for academic credit. AI is also prohibited for any purpose during examinations.
Berkeley, however, explicitly says it has not imposed a complete AI ban. Teachers may permit its use when appropriate, and the law school says students also need to learn how to use AI effectively for their future careers.
That distinction reflects a wider debate in education.
The OECD’s 2026 Digital Education Outlook found that general-purpose generative AI can improve the quality of work students produce without necessarily improving what they actually learn.
According to the OECD, handing cognitive tasks over to AI can encourage disengagement and what it describes as “metacognitive laziness”. Some research reviewed by the organisation found that advantages enjoyed by students using AI disappeared, and could even reverse, when they later had to perform without it.
At the same time, the OECD says AI designed and used with a clear educational purpose can improve learning and help develop skills including critical thinking, creativity and collaboration.
Most authorities stop short of a ban
UNESCO has similarly called for strict safeguards rather than excluding AI altogether.
Its guidance recommends privacy protections, age limits and teacher training, while its student competency framework argues that schools should prepare young people to understand, critically assess and responsibly use AI.
Australia has adopted a national framework for responsible generative AI use in schools, while England allows schools to decide how AI is used provided that safeguarding, privacy and educational requirements are met.
Sakai argues that these approaches do not go far enough.
In his view, even limited reliance on AI can reduce opportunities for students to practise thinking, writing and solving problems independently.
The emerging international approach is different: preserve the skills students must be able to perform without AI, while also teaching them how to use the technology critically when it genuinely supports learning.