Artificial Intelligence (AI) has changed the way we live and, consequently, the way our societies function. Access to knowledge and education has historically been shaped by inequality in general, the more economic resources people have, the greater their access to quality education, technology, and learning opportunities. What may be new now is that some of the tools that provide access to knowledge can also be harmful to the process of learning itself. If access to these tools, and the ability to use them effectively, is unequally distributed, they could create a new layer of inequality altogether. 

A narrowing gap, for the wrong reason  

Students from high-income backgrounds have the most access to AI and technology, yet they are the ones showing the biggest drop in the latest PISA results in reading, science, and mathematics. Advantaged students still outperform disadvantaged ones, but the gap between them is narrowing, not because low-income students are improving, but because socio-economically advantaged students are falling behind. In science, the performance of disadvantaged students remained unchanged, whereas advantaged students experienced an 8-point decline. In reading, scores fell by 4 points among disadvantaged students, compared with a 20-point decrease among advantaged students. In mathematics, advantaged students’ performance dropped by 14 points, while the change among disadvantaged students was not statistically significant. Advantaged or disadvantaged, these are the worst results in the history of PISA.  

There are multiple possible explanations for this deterioration in student outcomes, and many of them converge on the use of screens, technology, and, more recently, AI. There is evidence showing that technology and AI can harm cognitive processes. Tanil & Young (2020) found that the mere presence of smartphones in the classroom disrupts cognitive capacity. Kosmyna et al. (2025) report that using ChatGPT as a writing aid can impair users’ ability to remember and accurately reproduce their own arguments. Their findings further suggest that this effect may be linked to changes in neural connectivity in the brain.  

Beyond these immediate effects, Daron Acemoglu goes further and talks about a collapse in knowledge, at least as we conceive it. He argues there are no longer incentives to learn and think if information is only one click away, a shift that affects younger generations in particular. From this perspective, the problem is not that people have access to more information, but that the reduced cost of retrieving it may change the incentives to develop the underlying knowledge required to interpret it.  

The gap is not only cognitive or academic; social skills are also at stake. Overreliance on digital communication may limit opportunities for face-to-face interaction, reduce emotional depth, heighten communication anxiety, and contribute to greater feelings of loneliness (Rizqi & Raharjo, 2026). These early differences in how children engage with technology are likely to leave lasting marks on learning, communication, and social skills, and, ultimately, on the opportunities and jobs available to them.  

Different responses  

AI is spreading faster than the personal computer and internet did over comparable time frames, so students in low- and middle-income countries will soon have access to these same tools, and become subjected to the same risks. Some countries are already responding: Norway and France, among others in the Global North, began limiting technology use inside the classroom as early evidence on its cognitive effects emerged. This includes keeping phones off during class and only allowing its use for explicit learning purposes.  

Singapore shows that the same principle can be reached from a different direction. Its students are among the best performers in every subject, and they also use technology more than almost any other country; the difference is that its use has been structured around learning from the outset, rather than left open-ended. What separates outcomes, then, is not technology itself, but whether its use comes with supervision and purpose, whether that arrives as an early restriction, as in Europe, or as a design choice from the start, as in Singapore.  

Acting despite uncertainty 

We do not yet know with certainty what the long-term consequences of AI will be for children's development. What we do know, and what PISA's own data suggests, is that unsupervised use without a defined learning purpose is not producing good outcomes, regardless of a country's income level. Under that uncertainty, doing nothing is not a neutral choice: it defaults to precisely the unrestricted model that is currently failing. 

Access to AI itself is not what separates advantaged students from disadvantaged ones. In fact, advantaged students already have greater access to AI and are experiencing greater declines. What could come to separate them is access to the thinking required to use AI well, the critical judgment to question what a chatbot produces, the discipline to still do the cognitive work oneself, and to the policies countries put in place to build that judgment deliberately, rather than leaving it to chance. If this gap goes unaddressed, thinking itself risks becoming the next resource unevenly distributed among those who already have the least. 

Instagram: @carmenbdg

LinkedIn: https://www.linkedin.com/in/carmenbarreradegante 

Cover photo by Fajar Herlambang STUDIO on Unsplash

Disclaimer 

The views expressed in this opinion piece are those of the author/s and do not necessarily reflect the views or policies of IDS.