An Alternative to Cognitivism: Computational Phenomenology for Deep Learning

Abstract

We propose a non-representationalist framework for deep learning relying on a novel method computational phenomenology, a dialogue between the first-person perspective (relying on phenomenology) and the mechanisms of computational models. We thereby propose an alternative to the modern cognitivist interpretation of deep learning, according to which artificial neural networks encode representations of external entities. This interpretation mainly relies on neuro-representationalism, a position that combines a strong ontological commitment towards scientific theoretical entities and the idea that the brain operates on symbolic representations of these entities. We proceed as follows: after offering a review of cognitivism and neuro-representationalism in the field of deep learning, we first elaborate a phenomenological critique of these positions; we then sketch out computational phenomenology and distinguish it from existing alternatives; finally we apply this new method to deep learning models trained on specific tasks, in order to formulate a conceptual framework of deep-learning, that allows one to think of artificial neural networks’ mechanisms in terms of lived experience.

Editorial: Bio A.I. - from embodied cognition to enactive robotics

“…From a radically embodied perspective, one might argue that the entire field of cognitivist deep learning is on shaky foundations by virtue of needlessly appealing to the literal sense of the mind-machine metaphor, i.e. to minds as literal information processors (van Gelder, 1990 ; Van Gelder, 1995 ; Hutto and Hipólito, 2021 ; Beckmann et al, 2023 ). In their view, because computation and information processes cannot be found “in the wild” independent of human (scientific) practices, the literal sense of the analogy pushes toward a rudimentary view of natural intelligence (even if operationally useful in some circumstances).…”

“…. (van Gelder, 1990;Van Gelder, 1995;Hutto and Hipólito, 2021;Beckmann et al, 2023). In their view, because computation and information processes cannot be found "in the wild" independent of human (scientific) practices, the literal sense of the analogy pushes toward a rudimentary view of natural intelligence (even if operationally useful in some circumstances).…”

Principled Limitations on Self-Representation for Generic Physical Systems

“…Theorem 1 restricts any ANN from representing to itself whether it is in training or processing mode, though its outputs can make this difference evident to an external observer. Replacing the notion of representation with a first-person notion of a phenomenological “in the world—lived experience” [ 75 ] does not change this conclusion; experiences input streams, not its own processing, while only experiences a one-bit state change. Making the “supervisor” a component of the system, as in a Generative Adversarial Network (GAN), also does not change the conclusion; neither the supervisor component nor the combined system has access to either the overall architecture or any of the computed functions.…”