AI does not replace human judgment. It amplifies it.

A new headline shows up every week: “AI will replace teachers”, “AI will replace programmers”, “AI will replace lawyers”. And every week, the conversation asks the wrong question.

The question is not who it replaces. It is who it amplifies. Because a tool that multiplies capability does not erase the differences between people: it widens them. And that distinction — which looks like a nuance — changes everything: what is worth learning, what is worth teaching, and how it is worth working.

The mirage of replacement

The history of technology is, in large part, the history of this repeated misunderstanding. When the calculator arrived, nobody stopped needing to understand mathematics. Quite the opposite: it freed mathematicians from tedious arithmetic so they could take on the genuinely hard part — reasoning, modelling, interpreting. The calculator did not replace the mathematician. It replaced a task.

The spreadsheet went the same way: the end of accountants was predicted, and what happened was the reverse — there were more financial analysts than ever, because suddenly it was cheap to ask questions that used to cost weeks of manual work. The tool made execution cheaper, and in doing so it drove up demand for people who knew what to ask with it.

AI does not compete for your job. It competes against your critical thinking.

The same is happening now, but at a scale that is frightening. Artificial intelligence executes at a speed no human can match: it writes, codes, summarises, translates, proposes. But it executes what you ask of it. And that is where the whole game sits: knowing what to ask is judgment, knowing how to ask is a skill, and neither judgment nor skill can be downloaded — they are built.

There is one important difference from earlier technologies, and it is worth stating plainly: the calculator never appeared to think. AI does. It produces fluent, confident, well-structured text — whether it is right or not. That appearance of intelligence is precisely what makes the user's judgment more valuable, not less. Faced with a tool that always answers with confidence, the question “is this actually true?” stops being academic and becomes a professional survival skill.

The amplification gap

Here comes the uncomfortable consequence of the argument, the one almost nobody wants to face: if AI amplifies what each person brings, then it also amplifies the inequalities they start from.

Two students with the same tool do not get the same result. The one who arrives with vocabulary, with mental structure, with the habit of verifying, turns every interaction into learning: they interrogate, contrast, reformulate. The one who arrives without that floor accepts the first answer and moves on. One accumulates advantage with every use; the other accumulates dependence. The gap between them does not close with access to technology — it widens faster.

This has a direct implication for institutions and governments: handing out tools is not education policy. Handing out tablets or licences without building the judgment that directs them is handing out crutches. The real public policy of this moment is not about access, but about forming judgment, about building critical thinking. And that is a task no platform solves on its own, because judgment cannot be installed: it is cultivated, with teachers, with practice and with time.

What changes in education

If AI can generate a mediocre essay in seconds, then teaching people to produce mediocre essays stops making sense. What gains value is precisely what AI cannot do alone: framing the right question, judging whether an answer makes sense, connecting ideas across different fields.

This forces us to revisit something schools have avoided for decades: assessment. As long as we keep measuring the final product — the essay handed in, the exam completed — we will increasingly be measuring a student's ability to operate a tool, and less and less their thinking. The alternative exists and is well known: assess the process. Ask students to defend orally what they wrote. To show their drafts, their sources, their decisions. To explain why they discarded one path and chose another. AI can write the essay; it cannot hold the conversation about the essay.

This is not a threat to education. It is an invitation to do it better. To stop rewarding memorisation and start rewarding judgment.

Three skills that become critical

• Asking good questions. Whoever asks well, directs. Whoever only answers, obeys. Asking well demands knowing the subject: ignorance does not formulate good questions — good questions are born of prior knowledge. That is why content does not die; it changes function: it stops being the end and becomes the fuel of judgment.
• Evaluating evidence. AI hallucinates with confidence. Telling the true from the plausible is a central competence — and it can be trained: contrast sources, ask for the reasoning behind a claim, look for the data that would contradict it.
• Thinking across disciplines. The best ideas live at the crossings: where science touches politics, where technology touches the human. AI is encyclopaedic but literal; the lateral leap between fields remains a profoundly human gesture.

The new pact with work

Something similar happens at work. AI agents can take on the repetitive. But someone has to decide what gets automated, by what criteria, toward what end. That someone — for now, and for a good while — is you.

Consider a concrete case. When an organisation automates its communication with customers, the technical part — connecting systems, firing off messages, logging replies — is the easy part. The hard part comes before: deciding what deserves an automated message and what demands a human call, where efficiency improves the relationship and where it degrades it, what gets measured and what gets protected. The agent does not make those decisions. Whoever directs it does. And when nobody makes them with judgment, automation does not reduce chaos: it industrialises it.

That is why the professional profile that gains value is not the one who executes faster, but the one who defines better. Define the problem, define the quality standard, define when the result is good enough to go out into the world. Execution gets cheaper; definition gets more expensive.

The difference between using AI to free up your time and using it to make yourself dispensable lies entirely in how you direct it. And directing requires what no machine gives you: a view of where you want to go.