Paper short abstract:

Starting from the observation that biological realism is sidelined in the training of future AI specialists, I examine the ambiguous status of "Artificial Neural Networks" in the teaching, learning, description and promotion of ordinary Machine Learning Algorithms.

Paper long abstract:

Drawing on a study of the development of French education in artificial intelligence, based on ethnographic observations of three university courses, I examine the way in which political, economic and industrial considerations are helping to rebuild the AI project, and consequently the place allocated to biological realism in the ordinary practice of teaching, learning, developing, perfecting and implementing contemporary AI algorithms.

I will show that biological realism and the heuristic ambition to learn more about the mind from AI experiments, despite the existence of ambitious but rare research programs in this field, are being sidelined in the training of future AI specialists in favour of an engineering conception of AI, centered on the search for instrumental efficiency.

As a result, even though ordinary Machine Learning and Deep Learning algorithms, which currently dominate the AI scene, can indeed be defined as a repertoire of mathematical and computational techniques historically inspired by the workings of the biological brain, I will argue that the neurobiological metaphor of "Artificial Neural Networks" through which these algorithms are currently described serves essentially didactic and promotional purposes, in a context where biological realism of algorithms is of secondary importance, if at all. Indeed, if current ANNs can be considered as "views of the mind", it is perhaps first of all in the sense of what Latour calls in French "vues de l'esprit" (1985), i.e. as a system of inscriptions through which current Machine Learning algorithms are described, taught, learned and promoted.