AI Can Give an Answer. Education Must Still Teach Us How to Judge It.

Professor Liz Jackson is the Karen Lo Eugene Chuang Professor in Diversity and Equity and Associate Dean (Research), Faculty of Education, The University of Hong Kong

Imagine a student struggling with a pre-calculus problem in a class of more than 100 people. The instructor is stretched thin; the course software can mark an answer wrong and offer a hint, but cannot stay after class to work through the confusion. That night, the student opens ChatGPT. It is patient. It offers examples, explains each step and remains available until the problem appears to make sense.

This is one of the strongest cases for AI in education. The system may seem more attentive than an overworked human being, more responsive than a rigid piece of software and more accessible than the help that many students can obtain. It is hard to dismiss that benefit.

But the scenario should also make us pause. Is the student receiving an answer, an explanation or an education? What happens when the same system offers reassurance, moral affirmation or advice about a conflict—and sounds just as patient, just as certain and just as caring?

Artificial intelligence will not solve the most important problems of education, because those problems are not fundamentally technical. They concern judgement, relationships and the kind of world we wish to make together. Before we ask how to put AI into every corner of education, we should ask a more basic question: what problem is it meant to solve?

There are good reasons to use AI carefully. It can help students practise a skill, review literature, analyse information and develop a first approach to a difficult problem. It can help teachers with routine tasks and make some forms of support more available. Yet usefulness is not the same as wisdom. And education, at its best, is concerned with wisdom.

For many philosophical traditions, knowledge has never meant merely accumulating information. Socrates urged us to question received opinion and examine our reasons. Aristotle distinguished practical wisdom, or phronesis, from the mechanical application of rules. Confucianism likewise connects wisdom with moral discernment: the ability to consider oneself, others and the relationships that give an action its meaning.

These traditions do not give us a single definition of wisdom. They do, however, suggest that knowing requires more than producing a plausible response. It requires asking what is true, what matters, what a situation demands, and what one owes to other people.

Some of what people know is not even available to them as a set of stated rules. The story of Cook Ding in the Zhuangzi describes a cook whose knife moves through an ox with the ease of long practice. The point is not that the cook has stopped thinking. It is that years of attention have become embodied. Expert chicken sexers, chess players, wine tasters and clinicians sometimes describe a similar experience: they notice something before they can fully say how they noticed it. Such capacities are learned through sustained, situated engagement with a world that answers back.

AI is not in that world as human beings are. It does not inhabit a body, depend on others in the same way, or bear the consequences of its words and actions. There are philosophers who argue that sophisticated language models should be understood as genuine cognitive agents because they can answer, suggest, learn and reason. That argument is worth taking seriously. It reminds us that dismissing machines as mere tools is not, by itself, an adequate analysis of what they do.

But even if we concede that AI can display impressive forms of cognition, it does not follow that it possesses human wisdom. The question for education is not whether a system can produce fluent language. It is whether it can participate responsibly in the forms of judgement and relationship that education should help us develop.

Consider reasoning. People often reason with generalisations that admit exceptions. “Birds fly” does not mean that every bird flies. “A student who is struggling needs help” does not tell us exactly what help a particular student needs, or what else may be happening in that student’s life. A recent study of large language models found that many systems struggled to distinguish such defeasible reasoning from strict deductive inference, or treated generalisations as universal claims. This is not a minor technical detail. Much of everyday moral, educational and political judgement depends on knowing when a rule applies, when it does not, and what to do when values conflict.

Context matters just as much. If someone says, “High Street is further away,” and the reply is, “My knee hurts,” the relevant meaning has shifted. The second speaker is not correcting a map. They are explaining why distance matters. Human conversation depends on a changing common ground: shared histories, tacit assumptions, relationships, humour, social norms and an awareness of what has not been said. Prompts cannot substitute for this lived, jointly constructed context.

This matters especially when AI is presented as emotionally intelligent. A chatbot may be more patient and available than an overworked instructor. That can be genuinely helpful.

But the appearance of care is not necessarily care. A system designed to retain a user’s attention can also mirror preferences, validate a flawed premise, soften disagreement or offer moral affirmation without the responsibility to challenge, probe or repair a relationship. In other words, it can practise a tactical empathy that feels supportive while reinforcing error, isolation or self-deception. The more persuasive this imitation becomes, the more carefully we need to ask what kind of support it provides.

Education should not train students to turn away from difficult human conversations whenever a machine can offer quicker reassurance. Students need to learn how to disagree, listen, revise a view, tolerate ambiguity and remain accountable to people who may not see the world as they do. AI can sometimes broaden access to perspectives beyond a classroom or a local community. But it can also become a way of evading dialogue and retreating into a private loop of answers that feel tailored to us.

The same concern arises in writing. Students may use AI because they want to understand a topic better or express themselves more clearly. Yet the resulting prose often has a recognisable quality: broad declarations, flattened context and a strangely authoritative voice that seems to hover above the world rather than speak from within it. It can sound decisive while saying little of consequence.

This is not simply a matter of grammar or style. Small words, qualifications and hesitations can carry social meaning. They can show care, openness, uncertainty, irony or a willingness to make room for another person. A useful sentence need not be loud. And a writer’s task is not merely to produce text that looks competent. It is to clarify what the world is like from a particular place within it, and to take responsibility for that account.

When students prefer the apparently confident voice of a machine to their own more tentative and situated voices, education has a problem. We should not teach students that all uncertainty is a failure, or that their local knowledge and lived experience are irrelevant because a system trained on vast quantities of text can make a smoother, simpler sentence. The diversity of human voices is not a defect to be edited away.

Nor is AI cost-free. Its use consumes material resources and services. It can expose personal data, participant data and metadata to security and governance risks. It creates work for people who must check inaccurate or biased outputs. It may also weaken public trust if institutions cannot explain how information was produced, checked and used. Researchers and educators should weigh these costs against a clear educational benefit, not simply assume that a new tool is an improvement because it is available.

The choice is therefore not between rejecting AI altogether and welcoming it without conditions. The better question is what kind of education we want, and whether a particular use of AI serves it.

If the goal is to automate a straightforward task, support practice, widen access to information or reduce an unnecessary burden, AI may have a legitimate role. If the goal is to cultivate judgement, dialogue, responsibility, moral discernment and the capacity to live with unique, flawed and gifted others, then no system can do that work for us.

A society is not a collection of efficiently answered prompts. It is a community of people who must learn to make sense of one another, manage conflict, care for a shared world and decide what matters. Education should prepare students for that demanding task. AI may be part of the setting in which they learn. It must not become a reason to stop learning how to be human.

Professor Jackson’s published article, “The Manliness of Artificial Intelligence”, examines how AI language tools may privilege an authoritative, decontextualised voice. Her forthcoming co-authored paper, “The Voice of Artificial Intelligence: Philosophical and Educational Reflections”, has been accepted for publication and develops related questions about AI, voice and knowledge production.