Brain representation of perceptual stimuli at different levels of awareness

TL;DR

This paper challenges the assumption that some brain representations are permanently unconscious, proposing that neural architectures inspired by primate cortex can generate metarepresentations of nonconscious knowledge, enabling awareness through adaptive resonance.

Contribution

It introduces neural network models that explain how unconscious knowledge can become conscious via context-sensitive adaptive learning and resonance mechanisms.

Findings

Subliminal targets can become conscious in appropriate contexts

Neural architectures can generate metarepresentations of nonconscious knowledge

Adaptive resonance facilitates awareness of nonconscious signals

Abstract

This article questions the widespread assumption that there are brain representations that will always remain unconscious in the sense of being inaccessible to individual awareness under any circumstances. This implies that some part of the knowledge generated by the brain is once and for always excluded from consciousness and, therefore, from being communicated to the outside world. This standpoint neglects the possibility that the human brain might have a capacity for generating metarepresentations of nonconscious knowledge contents at a given moment in time through context sensitive adaptive learning, and is somewhat difficult to reconcile with experimental findings showing that initially subliminal targets can be made available to awareness, or break through to supraliminal levels of processing, when they are embedded in an appropriate perceptual object context (relevance condition). Specific properties of neural network architectures, inspired by the functional organization of the primate cortex, are able to explain how a human brain could generate this kind of perceptual learning. Signals or knowledge processed outside awareness could be made available to awareness through adaptive resonance of bottom-up and top-down signal exchanges in massively parallel neural network architectures; in other words, on the basis of statistically significant signal matches in the domain of time and in the domain of memory content.