Voices of the System: What Foresight with Hundreds of Kazakhstani Teachers, Parents, and Students Revealed

In the first two materials of the series, we discussed the big picture: why AI in school is a matter of governance rather than simple implementation, and why the world into which schools send children is changing irreversibly. But any big picture risks remaining an abstraction until one comes down to earth — to those who stand at the board every day, check homework, pick children up from school, and sit at a desk themselves.

That is why we decided not to guess what Kazakhstan’s education system thinks about the arrival of AI, but to ask it directly. This article is about what we heard. And the main conclusion, if reduced to one phrase, is this: the dispute has long ceased to be about whether AI is needed in school; it is about the conditions under which it should be allowed in. On those conditions, there is no agreement — and it is precisely in these disagreements that the most important issues are hidden.

In autumn 2025, from September 19 to 26, the TALAP Center for Applied Research, together with the international think tank Global Education Futures, conducted a large-scale study using the foresight method. It involved 231 participants — and not only officials and experts, but representatives of all key groups directly affected by AI in schools.

The composition of participants is telling in itself. School principals — 51 people. Methodologists — 47. Students — 45. Teachers — 44. Parents — 32. Representatives of the EdTech market — 12. In other words, at one table — even if a virtual one — were people who usually speak different languages and rarely hear one another: a child who uses a chatbot every day and a principal responsible for an entire school; a developer selling an AI solution and a parent worried about a child’s mental health; a methodologist writing standards and a teacher working with them in the classroom.

The method we used is called Rapid Foresight — a technology for jointly designing the future. Its value lies not in collecting attractive opinions, but in making visible what usually remains hidden: where the positions of groups coincide and where they collide; where conflicts are concealed and where points of agreement exist. Foresight does not average opinions into a colorless middle and does not hide uncertainty. It turns many voices into a structured map that shows both the shared and the contested. This is exactly the kind of map needed to build meaningful policy rather than act blindly.

The value of this particular composition lies in the completeness of the optics. Decisions about education are usually made from above, while those affected by them are asked last — or not asked at all. Here, for the first time, all six perspectives came together in one field, and it became possible to see what is lost in closed-door discussion: what is an obvious benefit for a student may be a threat for a parent; what a principal sees as a management task may be experienced by a teacher as a personal fear; what a developer sees as a breakthrough may appear to a methodologist as a risk to the standard. Without such a complete picture, any strategy risks serving the interests of one group and running into resistance from the others.

Let us begin with the little on which absolutely all groups agreed — from a fifth-grader to a principal with thirty years of experience.

The arrival of AI in education is inevitable. This is not a subject of debate, but the starting point accepted by everyone. AI is already in classrooms; it is too late to argue about that. The only question is how to live with it.

From this shared point follows a second agreement: school will have to change, and change deeply. Participants agreed that the content and logic of educational programs must be updated — a new technological reality cannot be squeezed into a curriculum written for another era. They also agreed that the traditional assessment system is doomed: the future lies with personal trajectories, portfolios, projects, and competence passports, rather than a one-off exam whose answer can now be generated by a machine. And on one more issue the consensus was unexpectedly firm: AI should be a teacher’s assistant, a “second educator,” but not a replacement. Participants from all groups consider the boundary between assistance and substitution fundamental.

But this is where agreement largely ends — and the most interesting part begins.

Before turning to disagreements, it is important to show the common trajectory that participants saw. They described a path of roughly ten years — from the current state to maturity. Each stage has its own face.

The first stage is today and the immediate future, 2025–2026. Its honest name is “loss of focus and chaos.” AI has already entered schools and even kindergartens, but it is used in isolated ways, without a system and without rules. Students place excessive trust in AI — often more than in parents and teachers. Teachers lack knowledge, and attempts to force the transition create a real risk of burnout. The material base is not ready: weak internet, a shortage of devices. Existing AI solutions are poorly localized, and there is an acute lack of Kazakh-language content. This is not a picture of failure; it is a picture of a beginning, when everything is moving but no one is managing the movement.

The second stage is the next three to five years, up to 2030. It can be called “assembly and scaling.” AI agents capable of supporting both teachers and school leaders enter the scene at scale. Unified transparent platforms and a new logic of management based on data rather than intuition become possible. But this is precisely where risks intensify: monopolization of the market by ineffective solutions, growing dependence of children on ready-made answers, laziness, and negative psychological consequences that will have to be prevented in advance.

The third stage is long-term, beyond the 2030 horizon. Participants marked it with a question: “AI awareness?” This is a state in which AI becomes a normal component of any learning and working environment — the “second educator” as a norm rather than an exotic tool — and requires a completely different logic of school organization. But on this horizon two risks that cannot be solved technically become especially acute: the risk of losing humanity — and therefore the need for special attention to values, empathy, communication, and friendship; and the risk of losing national identity and culture, which participants believe must be addressed proactively already now.

Ten years — from chaos to awareness. But, as we will see, this path can be traveled in very different ways, and there is no agreement on exactly how.

When participants spoke about what worried them, their concerns formed three clear groups. It is important to distinguish them, because each requires its own response.

The first group is cognitive and psychological threats. This is fear for children’s very ability to think. Erosion of thinking skills, when a child shifts mental work onto an algorithm. Dependence on AI and digital addiction. Negative influence on emotional life. Growth of unchecked information and, as a result, a crisis of trust — in sources, in school, and in one another. This is the deepest group of risks, because it concerns what makes a human being human.

The second group is social and ethical problems. Here the issue is school as an institution. Loss of authority of the teacher and the school. Degradation of the teaching profession. Violation of academic integrity. Erosion of what might be called the school’s “social glue” — the fabric of relationships for which children, in essence, gather together. Growth of digital and social inequality between those who have access and devices and those who do not. And, separately, cultural homogenization: the threat that global AI solutions trained on other material will erase Kazakhstani traditions and identity.

The third group is institutional and managerial risks. This concerns the system itself: cyber threats and data security; unreliability and opacity of AI technologies — we do not always understand how and why an algorithm made a decision; monopolization of solution markets; formalization and bureaucratization of implementation, when a living task turns into reporting; and regulatory delay — rules lag behind technology and remain fragmented.

Three families of risks — cognitive, social, and managerial. Responsibility for them lies at different levels: the first cannot be solved by an order, and the third cannot be solved in the classroom.

Now to the main reason foresight was needed. Participant groups agree that AI is inevitable, but they diverge sharply on the conditions under which it should be allowed in. These disagreements are systematic and run along several lines.

The first line is the boundaries of application. Students and representatives of the EdTech market tend to expand the use of AI: give us more new solutions and more opportunities. Parents and principals, by contrast, demand strict boundaries in order to preserve humanity and safety. Methodologists insist on standards and restrictions. The same tool is seen by one group as an opportunity and by another as a threat.

The second line is the speed of implementation. EdTech and students want to move faster: the world is changing, and falling behind is dangerous. Methodologists and parents demand caution and control: the price of error when working with children is too high. This is not a dispute about direction — everyone agrees where to go — but a dispute about pace, and it is no less acute.

The third line is the focus of attention — what should be considered the main issue. For students, the center is technology and practical benefit. For parents, values and child safety. For methodologists, the balance between human beings and AI. Each group looks at the same reform through its own optics and sees its own picture.

These disagreements also appear in more specific issues. On upbringing and psychology, everyone agrees that values and mental safety are no less important than technology — but parents emphasize control, students emphasize interest and balance, methodologists emphasize identity, and principals emphasize responsibility and honesty. On teacher training, everyone agrees that retraining is an absolute priority — but teachers primarily want grassroots exchange of practices among themselves, methodologists and principals insist on systemic support institutions, parents think about quality control, and EdTech thinks about partnership and standards. On infrastructure, everyone agrees that without it any talk about AI will remain declarative — but EdTech wants an open market, principals and methodologists want state standards, and parents want filters and safety.

The conclusion from this map is fundamental for all AI policy in education: these disagreements cannot be eliminated by an administrative decision. No order can reconcile a student dreaming of freedom with a parent demanding boundaries. Such conflicts can only be spoken through, coordinated step by step, and kept in a productive channel. Governance here is not a choice of the “right” side, but the maintenance of balance among all sides.

Despite all the disagreements, the foresight process revealed an unexpectedly strong zone of consensus — a set of ideas that almost all groups were ready to support.

Beyond what has already been named — the inevitability of AI, the need to revise programs, the transition to portfolios, and the role of AI as an assistant rather than a replacement — participants agreed on several more points. Teacher retraining is a key condition for everything else, but it should be conducted not only “from the top down” but with active involvement of teachers themselves. Without basic infrastructure and clear “rules of the game” for national and international providers, any initiatives will remain on paper. And perhaps most important: values, human qualities, and psychological safety must become the school’s main area of attention — in partnership with society, not instead of it.

From this agreement came an idea that sounds paradoxical at first in a conversation about digitalization: if AI will be everywhere, schools need “AI-free zones.” Spaces where live interaction, not mediated by an algorithm, is deliberately preserved — so that children do not lose the ability to think over long horizons, read long texts, argue face to face, and form friendships. This is not Luddism and not a rejection of technology. It is a conscious decision to protect the human element that technology can displace unnoticed.

It is worth showing honestly that this idea itself is also a subject of live debate. When it was brought to professional discussion, a strong objection was voiced: will an “AI-free territory” become a convenient pretext for those who do not want to change anything at all — a way to hide from change under a beautiful slogan and leave everything as it was? This is a legitimate concern. It shows how delicate the matter is: even where everyone agrees on a value, the way to protect it immediately becomes contested. That is exactly why such questions should not be resolved behind closed doors; they need public discussion with all sides.

Participants also emphasized one more point of consensus: existing AI systems ignore Kazakhstan’s cultural traditions and values. This is a risk that, in the common view, must be addressed not “someday,” but now.

Foresight was only the first of three studies in the cycle. It was followed by an expert session with industry specialists — representatives of the regulator, innovative schools, civil and international experts, and the EdTech community. It added an important dimension to the picture.

Experts placed possible AI projects along two axes. The first: whether the project helps the teacher or, conversely, replaces the teacher. The second: how “sensitive” the project is, meaning how likely it is to provoke resistance and debate. A pattern emerged that everyone planning implementation should keep in mind.

Projects in which AI helps the teacher and does not encroach on the teacher’s role fall into the low-sensitivity zone — they are accepted calmly. AI for reducing administrative workload, AI as a methodological assistant for teachers, chatbots for typical organizational questions, AI diagnostics of knowledge gaps without automatic grading — all this is perceived as reasonable support. But the closer a project comes to replacing the teacher and intervening in sensitive areas, the stronger the anxiety becomes: automatic assessment of knowledge, AI as the main instructor, and hidden monitoring of students’ behavior and cognitive patterns all fall into the red zone.

The conclusion is sobering: precisely those projects that look most impressive technologically — replace, automate, track — are also the most socially explosive. This does not mean they cannot be done. It means they cannot be done in a rush, without special pilot regimes, explanation, and protection. Technological attractiveness and social safety often pull in opposite directions here.

If we step back and look at the foresight results as a whole, a picture emerges that explains a great deal.

AI is changing not the tools of school, but its architecture. Content, methods, roles, and values are being transformed simultaneously — and this cannot be reduced to “digitalization.” The risks arise not from the technology itself, but from misalignment: when technologies move ahead of methodologies, assessment fails to keep pace with learning, upbringing falls outside the digital logic, and teachers are overloaded without support. Each group honestly holds its own focus — methodologists worry about methods and architecture, principals about governability and regulations, teachers about practice, workload, and fear of replacement, parents about safety, mental health, and values. None of these focuses is unnecessary.

The main value of the foresight process is that it made this multiplicity of voices visible and structured. It became clear that movement is inevitably going in several directions at once: from a single program to personal trajectories, from exams to portfolios and projects, from the teacher as a source of knowledge to the teacher as a navigator and facilitator, from control to support and analytics. It also became clear that the main task is not to choose one “correct” position out of six, but to assemble them into a working whole.

But Kazakhstani voices are only half of the conversation. To avoid stepping on other people’s rakes, it is important to understand what other countries have already tried — and why even the leaders have not yet fully succeeded. This is the subject of the next material in the series.

The material was prepared based on the results of foresight sessions held in autumn 2025 and winter 2026.