In 1998, Jean-Pierre Changeux and I presented a synthesis of these four ideas in the form of a plausible "global neuronal workspace"( GNW) brain architecture. Our theory distinguishes two computational spaces: specialized, modular, encapsulated and automatic processors, and the global space formed by long-axon neurons capable of disseminating information to distant brain regions. Access to consciousness would correspond to the sudden ignition of these long-axon neurons, distributed throughout the cerebral cortex, but particularly dense in the prefrontal and parietal cortices. According to the theory, during conscious perception, the GNW stabilizes, for a minimal duration, in a meta-stable reverberant state, which keeps the information "on line" for an arbitrary length of time. Thus, we call "conscious" any information that, having reached this level of central and global processing, has become accessible to all processors and thus, in particular, to the verbal processors that enable a verbal report.
Note that, according to the model, the same processor can be involved in both conscious and non-conscious processing, with the conscious mode characterized by top-down amplification and synchronization with GNW neurons. Furthermore, GNW neurons are assumed to be the seat of permanent spontaneous activity modulated by (1) bottom-up arousal signals; (2) selective reinforcement signals. The resulting temporal series of metastable states would correspond to the "stream of consciousness" described by William James. These states of activity have been simulated in semi-realistic neural network models (Dehaene & Changeux, 2005; Dehaene, Kerszberg, & Changeux, 1998; Dehaene, Sergent, & Changeux, 2003).