Reimagining education as ethics meet AI

In the evolving architecture of education, we find ourselves at a philosophical and technological crossroads—a place where the calculus of optimal decisions meets the ethics of transformation. Artificial Intelligence (AI), once a speculative possibility, is now an active agent reshaping the educational landscape. But as AI embeds itself deeper into classrooms and curriculum—automating assessment, personalising content, and optimising learning pathways—it brings with it a cascade of complex questions. Chief among them: not just whether it works, but whether it works for us—for learners, for communities, and for the society we hope to build.

This article explores the ethical dimensions of AI in education—not as an abstract academic concern, but as a tangible, urgent challenge with real consequences. If education is about cultivating the ability to think, adapt, and act in a rapidly changing world, then we must scrutinise how AI is influencing that purpose. The promise of AI-enhanced learning—efficiency, personalisation, access—obscures deeper risks: bias, exclusion, opacity, and perhaps most dangerously, the erosion of our educational imagination.

The algorithmic classroom: between promise and peril

AI systems are no longer experimental add-ons—they are becoming infrastructure. From adaptive learning platforms to school assignment algorithms, these tools increasingly guide decisions about student potential and resource allocation. Yet their inner workings often remain opaque, turning education into a black box of automation. The result is a system where students become data points, and decisions about their futures are governed not by educators, but by code.

This shift presents significant equity concerns. UNESCO and the OECD have documented how AI systems, trained on historical or non-inclusive data, tend to replicate and even amplify existing social inequalities. Minority students, for instance, may be flagged as “at-risk” by predictive models, shaping how teachers interact with them or what opportunities they receive. Grading algorithms may penalise students for non-standard grammar, disproportionately affecting learners from diverse linguistic or cultural backgrounds.

The danger is not only in error, but in authority. As decisions become more automated, they also become harder to challenge. As Ben Williamson cautions, the datafication of education threatens to strip learning of its human depth, turning it into a managerial exercise in optimisation rather than growth.

What is the purpose of education in an AI world?

But even if AI could be made fair and transparent, a deeper question looms around what are we educating for? In a world where automation is redefining work and life, it is no longer enough to ask whether students can master content. We must ask whether they can adapt. The classic dilemma from decision science—the optimal stopping problem—asks: when do you stop exploring and start acting? In education, this is not just a mathematical challenge but a metaphor for survival.

Proactivity, collaboration, and agility—these are no longer optional traits for students or institutions. They are imperatives. The educational systems of the future must cultivate not only knowledge, but the capacity to navigate uncertainty. That means prioritising emotional intelligence, social efficacy, critical thinking, and creative exploration. It means building learners who can evaluate and question technology—not just consume it.

Between bias and possibility: designing for ethical intelligence

Bias in AI is not a glitch; it is a mirror. When an algorithm recommends fewer resources to marginalised schools, or ranks students based on flawed proxies, it reflects the inequities of the world in which it was built. Yet this very fact also offers a pathway forward: if bias is learned, it can also be unlearned.

Concrete cases remind us that the risks of AI in education are not hypothetical—they are already unfolding. Automated essay scoring systems, for instance, have been found to penalise students whose language use deviates from dominant norms, particularly those from minority or multilingual backgrounds. Here, AI doesn’t simply grade; it encodes a cultural gatekeeping mechanism, reinforcing narrow standards of expression. Elsewhere, predictive analytics meant to identify “at-risk” students have disproportionately flagged learners from marginalised communities—not because they are less capable, but because the data used to train these models reflects histories of structural disadvantage. And in the construction of AI-curated curricula, western epistemologies often dominate, crowding out local or non-western knowledge systems. These examples are not glitches in the system—they are revelations of the system’s underlying logic. They ask us not only how AI works, but for whom it works—and at what cost.

But technical solutions alone are insufficient. We must also ask: are we teaching students how to live ethically with technology? Are we giving them the tools to critique, resist, and reimagine? AI literacy is more than knowing how to use a tool—it’s understanding its power, its limits, and its values.

A framework for the future: preconditions, literacies, skills

In response to these challenges, a new framework for future education is emerging—one grounded not only in efficiency, but in meaning. It unfolds in two phases: first, understanding the current terrain – data infrastructures, curricular gaps, and institutional readiness; second, designing actionable models for transformation. At the heart of this model is a triptych of educational virtues:

· Proactivity: the capacity to act with foresight and initiative.

· Collaboration: the ability to work across boundaries—social, disciplinary, and technological.

· Agility: the willingness to learn, unlearn, and adapt.

These virtues are embedded in measurable competencies, from digital and data literacy to emotional intelligence. They are tracked through six domains—from the accessibility of tools to the transparency of platforms. But more than metrics, this model is a philosophical stance: that the goal of education is not simply knowledge transmission, but the cultivation of informed, ethical, and flexible agency.

When ethics becomes a learning problem

Ultimately, this article argues for a shift in how we think about ethics in AI—not as a static checklist of dos and don'ts, but as a learning problem. Just as students need supportive environments to thrive, AI needs carefully designed contexts to evolve in ways that align with human values. Ethics, in this view, is less about controlling machines and more about educating them—and ourselves.

What makes learning impossible? Too much noise. Not enough trust. Environments that punish curiosity. These same obstacles apply to AI. If we want intelligent systems to behave ethically, they must be trained in conditions that mirror good pedagogy: transparency, diversity, reflection, and space to question.

And perhaps the most hopeful idea of all: AI that is always trying to learn might, paradoxically, be more ethical than any perfectly programmed machine. Because the will to learn implies the will to change—and the will to change is, in many ways, the soul of education.

We stand at a pivotal moment. AI can be a catalyst for justice—or a machine of exclusion. The difference lies not in the code, but in the culture around it. To make AI truly educational, we must not only improve its design—we must expand our own moral imagination. And that begins by refusing to settle for knowing. We must commit to learning.

And so, in the spirit of both humility and hope, we close with this aspiration:

We would like to see AI that really is trying to learn, always trying to learn.

Our faith is that such an AI, shaped by the principles of education—not domination—would be more ethical than any machine built only to execute.

By converting the problem of ethics into a problem of learning, we tap into the deepest wisdom of pedagogy: that what we nurture is what we become.

Views are personal. Anželika Berķe Berga, Associate Professor, Riga Stradiņš University; Chitro Majumdar, Sovereign Risk Advisor, RsRL Board Member, and co-theorist of the Delbaen–Majumdar Theory filtering AI bias.