Advances Brain Research and Neuroscience
ISSN: 3071-2750 | DOI: 10.64030/3071-2750
Rozita Aboutorabi
This review explores conscious processing and metacognition through the framework of Stanislas Dehaene’s Global Neuronal Workspace (GNW) theory, as articulated in his book Consciousness and the Brain. While Dehaene provides a compelling model of how thoughts are generated, maintained, and encoded in the brain, this paper proposes a hierarchical extension that incorporates metacognition as a distinct and dynamic component of cognitive architecture. The model comprises three interacting layers: (1) a Sensory Integration Layer in the parietal and temporal association cortices that unconsciously processes multisensory input and forms semantic associations, functioning like the input layer of an autoencoder; (2) an Intermediate Encoding Layer in deeper association areas or prefrontal cortex that transmits abstracted concepts and beliefs, akin to the bottleneck of a UNet-like structure; and (3) a Metacognitive Layer, located in the frontopolar and dorsolateral prefrontal cortex, which actively evaluates, modulates, and reconfigures lower-level processing, influencing cognitive strategies and behavioral outcomes. So, the model consists of two interconnected autoencoders. The base UNet encodes multi-sensory inputs into latent representations and reconstructs them through a decoder, capturing predictive coding and feature abstraction. The top-level metacognitive autoencoder receives the base UNet bottleneck outputs, encoding them into higher-level latent representations that capture ethical and reflective features. Its decoder refines and corrects the outputs, performing self-assessment, error correction, and ethical guidance. This hierarchical design enables the network to simulate metacognitive reasoning, dynamically evaluating and adapting its outputs in response to novel situations, mirroring human-like reflective and ethical cognition.