AI Ethics & Society Columnist
In many classrooms, the issue is no longer whether students can use AI, but whether AI understands that learning styles, languages, and family rhythms are never uniform. Technology that appears neutral often starts with very specific assumptions: who is considered the ideal student, how progress is measured, and what is defined as advancement.[4][8] This question matters because when digital tools enter education, they bring not only features but also a definition of normality.
UNESCO frames this issue firmly: the use of AI in education should uphold human agency, inclusion, equality, gender equality, as well as cultural and linguistic diversity.[1][4][7][8] The agency also highlights that the digitalization of education introduces issues of access, bias, ethics, protection, and accountability.[4][12] From this perspective, educational AI should not be judged merely by how quickly it provides answers, but by whether it expands learning rights without narrowing who is deemed worthy of help.
The World Bank views the challenge from a more material side.[2][5] To ensure digital education does not widen the gap, countries need to consider affordable connectivity, device procurement, cloud solutions, and multimodal learning delivery.[2][5] This focus reminds us that personalization never stands alone; it requires networks, electricity, devices, and uneven digital literacy. In many low-income or remote regions, the initial problem is not algorithms but whether students can connect at all.[2][5]
Here is where the promise of personalized learning becomes complicated. Adaptive systems can indeed tailor content to individual needs, and such approaches are often praised for delivering quicker feedback and more flexible learning paths.[8][9][11] However, personalization designed from a single educational model can become a subtle filter against those who learn differently. If success metrics are determined by patterns most easily read by machines, students whose learning styles deviate risk being seen as less capable, though the issue may lie with the system’s design.[3][6][8]
Several studies cited in the source package also remind us that personalized classrooms are not inherently fair.[3][6] Teacher or system designer biases can cause certain students to receive fewer challenges, while others are given more room to grow.[3] In environments already pushed toward personalization, such biases can be harder to detect as decisions appear to arise from data rather than human judgment.[3][6] But data is never entirely value-free; it records design choices, social context, and who was visible enough from the start to be included in the dataset.
There is also a frequently overlooked layer: language.[1][4][7][8] UNESCO explicitly stresses the importance of linguistic and cultural diversity because AI used in classrooms easily favors dominant languages and majority communication styles.[1][4][7][8] For students learning in national languages rarely prioritized, or families switching languages between home and school, ‘personalized’ can feel like an imposition of a single standard. In regions such as the Middle East, South Asia, or North Africa, this is not theoretical; it touches how children understand the world and how schools value their identity.
So, who truly benefits from this optimization? In the edtech market, business incentives tend to push systems that can be measured quickly: higher scores, reduced learning time, or increased engagement.[5][10] But education is not a production line. Learning success also encompasses confidence, the ability to question, collaboration, and resilience when facing challenges. If AI is measured only by easily marketable metrics, the most human aspects of education might be sidelined in performance reports.
Here it is important to be honest about what cannot yet be verified in the available source package. We do not yet have sufficient details to compare the impact of educational AI fairly across countries, especially the proportion of data from students in low-income countries or the extent to which the models used are truly trained in multilingual and multimod[1][2][4][5] Large figures about millions of learning records or dozens of studies can be impressive, but readers have the right to ask: who is included, who is excluded, and what standards are considered universal? Evidence that would change our reading includes data disaggregated by region, language, income level, and type of school.[1][2][4][5]
Therefore, the debate about AI in education should shift from 'is this technology smart?' to 'who sets the rules?' Guidelines, infrastructure, teacher training, and product design choices are as important as the model itself.[1][2][4][8] Technology that does not accommodate connectivity limitations, language diversity, and learners' first experiences in their families risks deepening the inequalities it aims to reduce.[1][2][4][5] The future of AI in education might be determined not just by its intelligence, but by how much the system recognizes that access is part of equity, not just an added feature.
References
References
Small numbered tags in the article body point to the sources below.
- UNESCO: Governments must quickly regulate Generative AI in schools
- Digital Technologies in Education | World Bank Group
- [PDF] Equity and Personalized Learning: A Research Review
- What you need to know about AI and the right to education | UNESCO
- Digital Pathways for Education: Enabling Greater Impact for All
- The Impact of AI-Driven Personalized Learning Platforms on ...
- UNESCO's guidance on technology and AI in education now available in
- AI and education: guidance for policy-makers
- AI and education: Guidance for policy-makers (UNESCO, 2021) – educ[AI]tion
- World Bank Document
- [PDF] AI and the future of education - la CUSO
- AI and education: Protecting the rights of learners | UNESCO
PICKUP ARTICLES
Pickup Articles
-
Generative AI & Foundation Models
Africa Didn't Wait for Banks to Redesign Money
Context article on how Africa built a payment and financial inclusion architecture centered on mobile phones, focusing on mobile money, interoperability, credit, regulation, and di
-
Generative AI & Foundation Models
In court, AI hallucinations are no longer a curiosity. They are a liability.
This article examines how AI hallucinations have moved from a technical annoyance to a legal and commercial liability, using court sanctions, legal-research studies, and shifting p
-
Generative AI & Foundation Models
Is Banning Smartphones Enough to Protect Children? Science Is Still Unsure
Recent debates on banning smartphones and social media among minors intersect with a body of nuanced research: some link screen time to increased anxiety or depression, while other