‘Train your own neural network’: an astrophysicist sees AI as an existential crisis for education

According to Douglas Scott, debates on AI in science overlook the key point: the immediate threat lies in the education of students, who must continue to learn to think for themselves.

Opinion pieces on the effect of major language models and agents on research are multiplying, including in astronomy. In a short text published on arXiv, astrophysicist Douglas Scott believes that they miss the most pressing problem for teacher-researchers: if the future of research is a matter of concern, the impact on teaching is an existential crisis, and it is already here.

Preserving the core of training

The author argues to defend what is at the core of a curriculum: to train graduates capable of thinking independently, "like an astrophysicist". AI tools must be used to support learning, not to replace it.

To get this message across to students, he offers an image borrowed from the vocabulary of AI itself: it is their own neural network, their brain, that they must train. Delegating exercises to a model is tantamount to depriving that network of the data it needs to learn.