Training the teachers before the pupils

Idiap and HEP Valais announced it in early June: a partnership to prepare future teachers to understand and use artificial intelligence, before the question is put to them by their own classes. The news would fit in a few lines were it not so clearly at odds with the prevailing intuition on the subject — that of the digital native, the adolescent who would by nature know how to handle a tool that the adult before him is discovering only belatedly.

This intuition is false, or at least ill-posed. It confuses ease with competence. A pupil of fifteen handles an interface with more speed than a teacher of fifty; granted. But speed of use is not what a classroom lacks. What it lacks is the capacity to judge whether a generated answer is sound, to spot where it flatters the pupil instead of correcting him, to tell a piece of homework that was thought through from one that was merely produced. A mathematics teacher who has been marking papers for twenty years recognises in a single line the reasoning of a pupil who is stuck and that of a pupil who has delegated the calculation without understanding the operation. This capacity for judgement is not acquired at fifteen. It is built over years in the classroom, over errors seen and seen again, over the different ways of explaining the same notion to children who do not understand it in the same way.

I make the same argument elsewhere about qualified work in general: artificial intelligence does not level competences, it multiplies the most experienced ones. A notary of thirty years' practice draws more from it than a trainee does, because he already knows what to ask of it and where to distrust the answer. A teacher who has marked fifteen generations of essays already knows which mistakes recur, which explanations fall flat, which pupils disengage in silence rather than by raising their hand. This experience the tool can amplify, by generating differentiated exercises, by picking out in the results of an entire class regularities that a human eye takes weeks to see. It cannot produce it in a teacher who is only beginning, whatever his talent.

A young teacher fresh from HEP Valais nonetheless brings something that seniority does not guarantee: an ease in experimenting without fear of losing face in front of a class, a curiosity for uses that the previous generation would not have spontaneously considered. Judgement is learned with time. Openness, for its part, is sometimes lost along the way. Nothing compels us to choose: initial training must preserve this openness while accelerating the acquisition of judgement, and this acceleration did not occur of its own accord before the question became urgent.

To train the teachers before the pupils is therefore neither excessive caution nor an outdated hierarchical reflex. It is to acknowledge that the orchestration of a class — which decides when to slow down, when to push, when to let a pupil make a mistake in earnest — remains a senior's trade, even when the one who practises it is twenty-five and has only just come out of HEP Valais. It is this judgement that initial training must install before the technical gesture, and that is the right sequence.

To let classrooms equip themselves as they go, without having trained those who will have to rule on each use, would cost more than the delay we believe we are avoiding by rushing. The gesture is learned quickly. Judgement is not.