What Should Children Learn Without AI?


As AI becomes a bigger part of education, more people are starting to push back against learning through screens. Schools and government leaders around the world are thinking again about the role of AI learning tools and other technology. Some are returning to more traditional ways of teaching, using books and paper.
In August, the UAE Ministry of Education announced new rules for using technology in classrooms. Starting in the current school year, public schools will give priority to books, notebooks and pens for sixth-grade English, Arabic, maths and science lessons. Sarah Al Amiri, the education minister, said the decision followed research into how screen time affects children's brain development.
Early results from a study of 2,500 pupils at nine public schools in the UAE found that children who did not use screens improved their school performance across all year groups. The biggest improvements were in maths, science and Arabic. “Those students who were in device-free classes, their retention nearly doubled, their behaviour overall was better, and interpersonal relationships and friendships formed because of the reduction in screen time,” Al Amiri told The National. For now, the new plan only applies to public schools.
Denmark is also reducing some forms of digital learning. It is bringing back printed textbooks and limiting the use of personal devices in schools. Sweden is also returning to physical books, paper and pens to try to improve falling reading and writing skills. According to a recent Unesco Courier article, Sweden's policy of giving every student a laptop has not improved school results. It has even had small negative effects on maths among students from less-educated families. In Australia, the federal government is promoting a “digital detox”. The plan aims to reduce students' screen time and limit AI use when it gets in the way of learning.
These changes come as more people worry about how digital technology and AI affect children's learning. Earlier this year, the OECD Digital Education Outlook 2026 found that generative AI “can scale personalised learning through intelligent tutoring systems” and “support knowledge acquisition by enabling collaborative learning and enhancing creativity.” But relying too much on tools that give direct answers “can reduce students’ active engagement, improving task performance without corresponding learning gains.” In simple terms, students may get better results on a task without actually learning more.
For example, a field experiment involving nearly 1,000 Turkish high-school maths students found that unrestricted access to OpenAI's GPT-4 helped students perform better in the short term. But when they could no longer use it, their performance was 17% worse. However, a version called “GPT-4 Tutor”, which had rules to guide its use, largely prevented this problem.
Even as worries about AI's effect on learning grow, companies are finding new ways to bring it into classrooms. In September, Napster signed a strategic partnership with GEMS Education to develop AI agents and digital versions of people. As part of the deal, Napster will make digital copies of teachers using their teaching materials, courses and papers. These digital copies will be available online. They will become a “copy of yourself that you can then use to scale your time,” explained Samuel Huber, Napster's CEO for the Middle East and Africa. Students will be able to talk to these digital teachers outside class, ask questions, go over ideas and learn at their own speed.
However, a new study from Columbia Business School, published in Science Advances, questions how useful these digital copies really are. Researchers carried out 19 studies that had been planned in advance and collected more than 13,500 responses from people. They found that AI copies made using more than 500 pieces of personal information were only slightly better at predicting what a person would actually say than a general AI model. The study also found that these copies relied on stereotypes about groups of people. They were less accurate when representing people with lower incomes and less education. They also tended to trust technology more and feel more positive about it than the real people they were meant to copy.
The study also found that digital copies showed what researchers called “hyper-rationality”. They often gave “perfect” answers, as if they had complete knowledge and were answering a textbook question. This means they did not think like real people, who have their own habits, gaps in knowledge and imperfect ways of reasoning. The digital copies also showed less variety than real people. The study did not test digital teachers or look at their effects on education.
Even with these findings, lead researcher Olivier Toubia, senior vice dean for curriculum and instruction at Columbia Business School, does not reject the technology completely. Instead, he says that how accurate a digital copy is depends on whether it has been given information that matters to the question it is being asked to answer.
“Human performance is going to be a function of how humans are able to use AI to multiply their talents,” says Toubia. “But what is the base? If you don’t have any human knowledge, any kind of input into the process, then you’re completely useless. If you augment zero, zero times 1,000 is still zero. There has to be some core that comes from the human, which means children need to develop this base, this human touch.”
Toubia says children need to learn two main things. First, they need basic knowledge and important school skills. Second, they need people skills, such as communication, working with others and creativity. “That is going to come from traditional courses and should be acquired without AI because they’re learning fundamental human knowledge. At the same time, they need to know how to multiply this knowledge with AI to become superhumans. So, learning how to use AI to augment and multiply human knowledge.”
This leads to a bigger question: which learning experiences should stay human rather than move to technology? The sixth edition of the ARM Holding Children's Programme set out to explore this question. The Bigfoot Project was an art workshop for children aged five to 17. It ran from the start of the school year until 4 November and brought together children from 80 schools across the UAE. Created by Lithuanian artist Tomas Daukša, it encouraged children to think carefully, understand other people's feelings and solve problems.
“Learning is not simply acquiring knowledge,” says Daukša. “It is also about what we experience in the process of pursuing it. A crucial part of learning is having the freedom to explore without necessarily knowing where an idea will lead, to experiment, make mistakes and discover unexpected things. These experiences shape not only what children know, but also how they think and understand the world.”
Daukša agrees that screens are now a part of everyday life and that technology offers many possibilities. But he says people also learn through touch and physical experience. “We learn not only by seeing and processing information, but also by touching, making, experimenting, imagining and interacting with the physical world around us,” he says. “When we rely primarily on screens to gain knowledge and experience, we risk losing things that are fundamental to how we understand the world and relate to each other.”
In this year's Bigfoot Project, children were asked to imagine the mythical human-like creature as a curious traveller arriving somewhere unfamiliar. This helped them practise understanding other people's feelings and seeing things from different points of view. “What I saw was that children are very eager to imagine, explore and create,” says Daukša, whose work brings together art, science, facts and fiction.
“They needed only a reason, or perhaps not even a reason: simply an opportunity, a chance to play with their imagination. And they went off. This eagerness to imagine and play with ideas is something we often lose as we become adults. I believe that the freedom to imagine and the capacity to translate ideas into reality will be among the most important skills in the near future.”
He points out that AI is taking on more tasks that involve handling information and finding solutions. Because of this, the “ability to ask unexpected questions, imagine, experiment and bring original ideas into the physical world will become extremely valuable.”
