Le bruit et le pastiche : le surréalisme et le pop art comme précurseurs du contexte culturel de l’induction statistique
Le surréalisme constitue une première lignée de la modernité qui a valorisé le pouvoir créateur, l’inconscient, le rêve et les états d’hallucination provoquée, tout en s’intéressant paradoxalement aux machines et en proposant de mettre en rapport l’automatisation de la technique et l’automatisation de la psyché (Picabia, Duchamp).
Toutefois, pour que cette répétition ne soit pas à l’identique, on introduit du bruit de plusieurs façons. Il peut y avoir du bruit en mélangeant différentes catégories des données, c’est qui permet de créer des formes métamorphiques. Le bruit peut aussi venir d’une accentuation de certains paramètres logiciels pouvant aller jusqu’à la pareidolie. Enfin, il y a un bruit plus classique qui est produit informatiquement pour troubler partiellement la répétition et créer un équilibre entre le reconnaissable et ce qui n’existe pas déjà dans le dataset.
Surrealism constitutes a first lineage of modernity which valued creative power, the unconscious, dreams and states of provoked hallucination, while paradoxically taking an interest in machines and proposing to link the automation of technology and the automation of the psyche (Picabia, Duchamp).
Pop art constitutes a second lineage that placed at its heart what seemed to be on the periphery of art, i.e. the mass media and their industrialized production. Cubism had already initiated this position. With pop art, the background of our perception is always already constituted by this large-scale production, which overdetermines the aesthetic framework in which the work of art is received.
The first lineage can be interpreted as the strange junction between the anthropological and the technological, because the deepest part of the human being is anonymity and a form of automatism that has some proximity to the technical. The second lineage removes from the work of art its nature of event and interruption (one would have to see its exact provenance to know whether it has never existed historically or whether it is an afterthought reconstitution) in favour of a pastiche on the borderline of an almost unlimited production, that of the industrialized media. The work of art would thus be situated on the fringe of media production, always exceeding it quantitatively and overwhelming it like a flow.
However, when observing the results produced by artificial imagination, i.e. by the use of the technologies of the said artificial intelligence to produce cultural media, these two components are clearly visible. On the one hand, the pastiche of mass culture, which aesthetically reminds us of pop art. On the other hand, a dreamlike aspect whose visual resemblance to surrealism is striking. This double resemblance is rooted, it seems to me, in the way these images are produced, that is to say in their technogenesis.
Thus, recursive neural networks feed on one side of a large stock of information called Dataset, some of which has been accumulated thanks to the participation of Internet users since the appearance of Web 2.0 and which in this sense are the results of mass culture. On these stocks they will perform statistical calculations to define the proximity between one unit and another, for example a pixel in X and Y or an RGB colour. Thanks to this induction, they will be able to produce similar results, i.e. to continue the series that may belong to the stock. Thus they will be placed on the side of the pastiche, i.e. of a repetitive reference. The repetition is here the goal of the learning.
However, to ensure that this repetition is not identical, noise is introduced in several ways. Noise can be introduced by mixing different categories of the data, which is what makes it possible to create metamorphic shapes. Noise can also come from an accentuation of certain software parameters that can go as far as pareidolia. Finally, there is a more classical noise that is produced computer generated to partially disturb the repetition and create a balance between the recognizable and what does not already exist in the dataset.
To sum up, pop art finds an echo in the fact that statistical induction software feeds on pre-existing data and therefore feeds on our society of industrial production by trying to mimic them. Surrealism has an echo in these softwares through the introduction of a noise that blurs the categories and gives a metastable balance between the known and the unknown producing a Unheimlichkeit effect as our cognitive system tries to bring the latter back to the former. This double echo is not a historical or technical causal link, but allows us to envisage a cultural context in which our apprehension of images is bathed. Moreover, a thorough cognitive study could perhaps demonstrate that this echo is based on prior cognitive categories, as in the work opened by Andy Clark in Surfing Uncertainty: Prediction, Action, and the Embodied Mind.
By comparing the resemblance of the artificial imagination with certain movements of the avant-gardes and the technical modes of operation of these software programs, it is possible to draw the conclusion that most of the criticisms of the artificial imagination are based on an erroneous conception of the image, which is conceived as the appearance of a pure singularity unrelated to a context of mass culture. This conception found in Badiou’s work is idealistic in so far as it is unrelated to the material production of images and perhaps has its origin in that most singular moment of Romanticism (hypothesis to be verified Maurice Elie, Lumières, couleurs et nature) when certain writers and philosophers made the artist a fictional character, that is to say, an image, their image.