Judgment

Judgment is the ability to assess a situation and decide, even though no clear rule dictates what is right. One does not simply apply a regulation, but weighs what is appropriate in this particular case. A teacher who wavers between a grade of 2 and 3 on an essay needs judgment. A doctor who, when choosing between two possible treatments, also considers the patient’s life needs it just as much. The term originates in philosophy, but has for some years now constantly appeared in texts about artificial intelligence. There, it means what humans are supposed to contribute once a computer program has already made a suggestion.

What machines calculate and what they do not decide

Modern AI systems are very good at finding patterns in large amounts of data. They can predict which loan is likely to default or which text fits a given question. What they do not provide is the decision as to whether this suggestion is actually defensible in the concrete case. It is precisely this gap that judgment is meant to fill.

The difference can be illustrated with an example. A system evaluates applications and screens out a person because their résumé has a gap. Statistically, that may be a meaningful signal. But whether one is allowed to reject a person because of it is not a computational task, but a question of fairness and responsibility.

That is why laws such as the EU’s regulatory framework for artificial intelligence require that a human retain oversight in high-risk applications. The technical term for this is “human oversight.” This oversight is only worth something if the human understands the matter and can genuinely object. Anyone who merely rubber-stamps a decision is not exercising judgment, but merely supplying a signature.

Experience, context, and the courage to object

Judgment does not arise from knowledge alone. Three things are needed: experience with similar cases, understanding of the context, and the willingness to stand behind the decision. Knowledge tells you what options exist. Judgment tells you which of them fits here.

The philosopher Immanuel Kant described judgment as the ability to subsume an individual case under a general rule. This ability, he argued, cannot be replaced by further rules. For every rule, one would again need a rule for when to apply it. At some point, a human simply has to decide.

With AI systems, an additional difficulty arises. Many models phrase their answers very convincingly, even when they are wrong. Anyone reading a fluently written answer is more inclined to believe it. Here, judgment also means distrusting the confident tone and checking it.

From the classroom to the boardroom

In school, this becomes apparent in dealing with chatbots. A program delivers a ready-made essay, but whether the arguments actually fit the assigned topic is something the student must judge for themselves. Anyone who cannot do this fails to notice when the text misses the point of the task. That is precisely why curricula now require students to check AI results rather than simply hand them in.

In companies, the term appears in reports on personnel and automation. Banks have software pre-sort loan applications, while final approval remains with employees. Doctors receive hints from analysis programs about suspicious areas on an X-ray, but they make the diagnosis themselves. In job postings, judgment is increasingly cited as a qualification in its own right.

A common misconception is that judgment is simply gut feeling. That is not the case. A gut feeling need not be justified, but a judgment must be. Anyone who makes a judgment can explain which reasons they weighed and why they decided as they did. It is precisely this justification that an AI system cannot, as yet, provide on its own.