Baars has registered at least 25 years of active pursuit of “conscious contents” as “coherent, global information” in brain-like behavior. Global workspace theory (GWT) postulates that the role of consciousness is to evoke widespread brain activation accompanied by considerable unconscious working memory activity involving verbal rehearsal, visual semantics, and executive functioning. The paper’s first figure offers empirical evidence for widespread activation, while a second figure characterizes identified components’ complex interplay.
Franklin adds an agent point of view (intelligent distribution agent (IDA)), along with additional consciousness and cognition points of view. IDA works through cognitive cycles, presented in a figure and described in the text (a nine-point elucidation). IDA is now being extended to LIDA, with the L signifying learning.
Over the past few years, collaboration between these authors has effected a fortuitous meshing of GWT thinking and IDA/LIDA (computer) implementation. The goal sought is “fleshing out” GWT relative to perceptual, episodic, procedural, and attentional learning.
Overall, the paper emerges principally as a review and suggests future promise. The authors are obviously in agreement, commendably so, with a postulate I forwarded in a work over 30 years ago, to wit, that complex schematics require implementation to be “convincing.” Somewhat disappointing here, then, is the lack of description of extant (neural net implementation) details. The authors’ contention that neural network (NN) models have “drawbacks at a more integrative, architectural level” thus appears insufficiently supported; at the same time, contrary views, perhaps also agent based, may have to undergo evaluation by these authors as they continue expanding their own insights.