Much of the current debate about artificial intelligence treats different human capacities as if they belonged to a single scale: more data, faster processing, better outputs. From that perspective, systems that outperform us in some tasks appear to be gradually approaching, or even surpassing, human judgement. But the activities usually grouped together under that heading are not the same kind of thing. To predict, to decide, to deliberate, and to judge are related capacities, but they are not interchangeable.
Prediction: Estimating What Is Likely
Prediction concerns likelihood. It identifies patterns in past data in order to estimate what is likely to happen next. Weather forecasts, fraud detection systems, recommendation engines, and language models all operate substantially in this way. They estimate probable continuations, probable behaviours, or probable outcomes.
Predictions may inform action, sometimes decisively, but they do not settle it. A model may predict that a borrower from a particular postcode is more likely to default. That may be statistically useful. It does not by itself determine whether granting the loan would be fair, prudent, or socially desirable.
Decision-Taking: Selection Within Given Criteria
Decision-taking means selecting among alternatives according to criteria that have already been set. A navigation system chooses the fastest route. A hiring filter ranks candidates based on specified criteria. A school admissions algorithm allocates places according to published priorities.
This can be highly effective because the system is not determining the worth of the criteria themselves. It is operating within them. Choosing the fastest route is not deciding where one ought to go. Ranking candidates by chosen metrics does not decide whether those metrics deserve priority.
Decision in this sense is powerful, but it remains bounded by a prior human act: someone has already decided what counts.
Deliberation: Weighing What Cannot Be Fully Measured
Deliberation begins when no single rule or metric resolves the matter. It involves weighing competing goods, considering consequences, interpreting circumstances, and recognising that some values cannot be reduced to one scale.
Should a hospital prioritise urgency, age, prognosis, equality, or waiting time? Should a school prioritise examination performance, inclusion, character formation, or social mobility? Should a company protect jobs, maximise profit, or preserve long-term trust?
These are not merely technical questions. They involve rival goods that cannot be fully compared. Deliberation does not process considerations toward an optimum. It weighs what cannot be fully measured and reaches a position that remains contestable.
Judgement: Commitment Under Responsibility
Judgement is deliberation that has issued in commitment. It includes elements of prediction, decision, and deliberation, but it cannot be reduced to any of them, because what distinguishes it is not the quality of the reasoning alone, but the nature of the commitment it produces.
Judgement occurs when someone determines what ought to be done here and now, under conditions where certainty is unavailable, and consequences matter. A machine may recommend, rank, or simulate an argument. But it cannot bear responsibility in the human sense.
A system can be audited, corrected, regulated, or even blamed instrumentally. That is not the same as answering for a determination. Responsibility requires a subject capable of recognising fault, offering justification, and bearing the consequences of being wrong.
Systems may be part of the chain of action, but they do not stand within it as accountable agents.
A Simple Example
Imagine a university awarding a scholarship.
A predictive model estimates which applicants are most likely to complete the degree with high marks. That is a prediction. An algorithm then ranks candidates based on grades, income levels, and predefined weightings. That is decision-taking. A human committee notices that one student’s lower grades followed a year of serious family difficulty. Members discuss whether resilience, disadvantage, or future potential should matter more than raw attainment. That is deliberation. The committee then chooses and publicly stands behind the decision. That is judgement.
The acts are connected, but they are not the same. Each requires something different of the human being involved.
Why This Matters for Education
If education is understood mainly as the production of correct outputs, much of its domain becomes vulnerable to automation. Systems can already predict, classify, rank, retrieve, summarise, and generate competent responses at speed. If these are treated as the highest aims of learning, machines will increasingly outperform students in the very tasks schools reward.
But education has a deeper purpose.
Students need to learn what predictions can and cannot tell them. They need to understand how decisions are shaped by criteria chosen in advance. They need practice in weighing goods that cannot be reduced to a formula. Above all, they need opportunities to commit themselves to conclusions they can justify and defend.
That kind of formation does not arise through exposure to information alone. It develops through use, through practice that requires one to take a position and answer for it.
Conclusion
Judgement begins where no metric fully determines what ought to matter, where someone must decide, and where that decision must be defended and owned.
The more we distribute execution to intelligent systems, the more essential it becomes to form individuals capable of governing what those systems do not and cannot settle.
That capacity, to deliberate under genuine uncertainty and to answer for the determination reached, is what education must still form. No system can form it on our behalf.