The arrival of artificial intelligence in the educational sphere has revived an old aspiration: that technology might finally make it possible to resolve the structural problems that schools have carried for decades. Platforms capable of personalising learning, automated assessment systems or assistants that ease teachers’ administrative burden all feed the idea of an imminent qualitative leap. The risk is well known. In education, promises of rapid change often collide with complex, unequal systems that are deeply shaped by their context.
The recent history of EdTech offers ample examples of well-intentioned innovations which, when implemented without a systemic reading, left little trace or even reinforced existing gaps. It is within this context that the publication Artificial Intelligence and Education in the Global South is situated, a collective work coordinated from Harvard University that proposes a deliberately cautious perspective. Far from celebrating AI as a solution in itself, the book asks what conditions must be in place for technology to genuinely contribute to improving learning.
Its thesis is clear: artificial intelligence can only make educational sense if it is integrated into a holistic vision, aligned with curriculum, teacher training, assessment and the governance of education systems. Without such coherence, innovation risks remaining little more than rhetorical shorthand.
The starting point: an education system in crisis (especially in the Global South)
Over recent decades, many countries have made notable progress in school enrolment. Millions of children who were previously excluded from education now attend school. Yet this progress coexists with another, far less visible reality: a significant proportion of students do not acquire the basic learning outcomes expected for their age. Reading with understanding, comprehending a simple text or solving elementary mathematical problems remain challenges for far too many learners. Access alone does not guarantee learning.
This crisis is particularly acute in the so-called Global South, where low educational outcomes are compounded by persistent inequalities, deep digital divides and institutional systems with limited capacity. Schools with precarious infrastructure, overburdened teachers and little pedagogical support create a context in which any innovation—including technological ones—rests on fragile foundations.
This is why the book focuses on these countries. Not only because they account for most of the world’s school-age population, but because the margin for error is smaller. Introducing complex technologies into systems still struggling to guarantee foundational learning entails high risks: diverting resources, increasing inequalities or reinforcing existing dynamics of exclusion.
It is within this framework that the core issue is played out. As the Harvard researchers emphasise, in contexts of scarcity, educational decisions are inherently political. Investing in artificial intelligence without a careful reading of priorities, capacities and objectives can prove more costly than not innovating at all. The question we should therefore ask is not what technology can do, but which educational problems we choose to address—and who is left behind in the process.
Thinking of education as a system
To speak of education as a system is to recognise that its main components—curriculum, teaching staff, assessment, school management and governance—do not function independently, but in constant interaction. Changing one without attending to the others often produces limited or contradictory effects. An innovation may improve one aspect while simultaneously disrupting others, diluting its impact or even creating new problems.
This systemic perspective is one of the book’s central contributions. Artificial intelligence, its authors argue, does not operate in a vacuum: it is embedded in pre-existing structures, with their own rules, incentives and professional cultures. A personalised learning system, for example, is unlikely to transform teaching if the curriculum remains rigid, assessment rewards memorisation, or teachers lack the time and training to reinterpret the data generated by technology.
Education systems are also particularly sensitive to so-called second-order effects. Automating administrative tasks may free up time for pedagogical leadership—or generate new bureaucratic burdens. Introducing more sophisticated assessment tools may enrich information about learning—or reinforce control practices if not accompanied by changes in school culture. Outcomes are rarely linear or predictable.
This is why many educational innovations fail not because of technical flaws, but because they are introduced as isolated components within complex mechanisms. Thinking of AI as a lever for transformation therefore requires something more demanding than adopting new tools. It requires asking how each intervention fits within the system as a whole, and what pedagogical, organisational and political adjustments are needed for genuine improvements in learning to take place.
Thinking of AI as a lever for transformation requires asking how each intervention fits within the system as a whole, and what pedagogical, organisational and political adjustments are needed for genuine improvements in learning to take place.
Where AI can contribute… if conditions are right
The book avoids presenting artificial intelligence as a uniform solution and instead offers a tour of the specific areas of the education system where its use might make a difference. We examine them below.
Personalising without changing the logic is not transformation
Approached from a systemic perspective, artificial intelligence opens up real possibilities for improving student learning. The most frequently cited promise is personalisation: systems capable of adapting pace, content or support to individual needs could help address classroom diversity more effectively. Yet the available evidence remains uneven and points to a clear limit: when such tools are introduced without changes to curriculum or teaching practice, they tend to reproduce existing models of instruction rather than transform them.
Teaching with AI is not the same as educating for a world with AI
A similar dynamic applies to curriculum and so-called AI literacy. Teaching students how to use tools is not the same as preparing them for a world permeated by artificial intelligence. This entails understanding how AI works, developing critical thinking about its limits and effects, and situating it within a broader ethical and social framework. Integrating these competencies requires deliberate curricular decisions, not marginal additions to already overloaded programmes.
Better assessment does not depend solely on better algorithms
Assessment is another area where AI could make a significant difference. The ability to analyse written texts, problem-solving processes or learning trajectories opens the door to valuing more complex competencies than mere memorisation. But technology alone does not redefine what counts as learning. If assessment systems continue to prioritise low-level standardised testing, AI is likely to reinforce those same logics, albeit in a more sophisticated form.
When AI supports teachers (and when it overwhelms them)
For teachers, artificial intelligence can become a meaningful ally. Tools that support planning, provide feedback or reduce administrative workloads can free up time for what matters most: teaching and accompanying learning. The risk arises when such solutions are conceived as substitutes for professional judgement or introduced without sustained investment in teacher training.
Governing with data requires more than having data
In school management and governance, AI promises to improve decision-making through more intelligent use of information, optimise resources and anticipate problems. Without clear frameworks for use, transparency and accountability, however, these capacities can lead to new forms of control, exclusion or mistrust.
What emerges from all this is consistent: artificial intelligence can add value on multiple fronts, but only when integrated into systems that are clear about what they seek to improve. Without clear educational priorities, the technological promise fades.
The risks: when innovation can make the problem worse
Debates about artificial intelligence in education often focus on its potential. Much less visible is the analysis of the risks involved in adopting it within fragile or unequal education systems. Ignoring these risks does not make them disappear; on the contrary, it may amplify them.
The most evident risk is the widening of inequalities. In contexts where access to devices, connectivity or pedagogical support is uneven, AI-based solutions tend to benefit first—and sometimes only—those who already enjoy better conditions. The promise of personalisation can thus translate into new forms of educational segmentation.
Added to this are algorithmic biases and a lack of transparency. Many automated systems rely on incomplete or unrepresentative data and operate as black boxes that are difficult to audit. When such tools influence processes of assessment, guidance or resource allocation, the risk is not merely technical but profoundly ethical and political.
Another growing concern is dependence on technology providers. The adoption of closed solutions designed outside local educational contexts can limit systems’ capacity to adapt, regulate or even understand the technologies they use. In doing so, it erodes the local agency of teachers, schools and education authorities.
Finally, there is the risk of distraction. In systems still struggling to guarantee basic learning, investing time and resources in sophisticated solutions can divert attention from more urgent priorities.
What conditions give AI educational meaning
One conclusion stands out from debates on artificial intelligence and education: the impact of technology depends not only on what it can do, but on how and why it is used. For AI to make educational sense, the first step is to prioritise real problems: improving basic learning, supporting teachers, reducing persistent inequalities. Without this hierarchy of objectives, innovation risks dispersing itself in eye-catching but irrelevant solutions.
Designing with equity and context in mind is the second condition. The same tools do not function equally well in systems with intermittent connectivity, overcrowded classrooms or limited resources. Thinking about AI from the Global South means adapting it to those realities, rather than importing models conceived for other environments.
Teacher training is an indispensable requirement. No technology transforms teaching if those in the classroom lack the time, support and skills to integrate it critically into their practice. Without sustained investment in professional development, AI becomes yet another burden.
This is compounded by the need to evaluate impact. Measuring how many schools use a tool says little about whether it improves learning. Ultimately, all of this requires clear governance and regulatory frameworks capable of guiding the use of AI as public policy rather than as a collection of isolated initiatives.
Only then can technology cease to be an end in itself and become a means in the service of education.
Education remains a human project
As Artificial Intelligence and Education in the Global South reminds us throughout its chapters, artificial intelligence can expand capacities, offer new forms of support and open up spaces for improvement in education systems. But it does not replace teachers or the collective decisions that define what it means to educate. Teaching and learning remain profoundly human processes.
In a period of rapid technological acceleration, the decisive question is not how far education can be automated, but what criteria guide those decisions. Schooling is not merely a technical space, but a social and political project that defines what is learned, how and for whom. Artificial intelligence can be integrated into that project and reshape some of its practices, but it cannot replace it. What is truly at stake is not the adoption of a new tool, but the model of education system we choose to build in the age of AI.
