Automatism yesterday and today: reflexions on the limits of artificial intelligence from Simondon

Abstract

With the growing popularity of artificial intelligence techniques, questions about the capabilities and limitations of these technologies gain notoriety. This work proposes that, as contemporary forms of automatism, these technologies share similarities with those analyzed by Gilbert Simondon’s philosophy of technique. It also proposes that therefore his thought is of great value to reflect on them. Developing mainly from “Complementary note on the consequences of the notion of individuation”, and taking artificial neural network as an example algorithm, the article presents three moments when artificial intelligence would express limitations in relation to its structure that was fixed by its inventor. These limitations are in the input data, in the training of the model and in the final representation of the knowledge allegedly learned. The article concludes with a brief analysis on the consequences of extensive use of AI technologies in the light of the limitations presented.