The Global Neuronal Workspace Theory of Consciousness

I thank Dr. Dror Dotan, head of the Department of Special Education at Tel Aviv University, whose conversation with me inspired this post.

In 2022, the Association for the Scientific Study of Consciousness held a conference at which four theories of consciousness were presented by their authors. Professor Liad Mudrik of Tel Aviv University, a consciousness researcher who attended the conference, later published an article summarizing the four theories presented there, together with a fifth theory. In this post, we will discuss Dehaene’s theory.

Stanislas Dehaene is one of the leading researchers in the field of numerical cognition. Dehaene is the originator of the triple-code model, according to which numbers can be represented by three types of representation: a quantity representation, meaning the numerical magnitude of the number; a verbal representation, in which the number is represented as a sequence of words, such as “three hundred and twenty-two”; and a visual representation, in which it is represented by digits, such as 322.

The central point of this model is that processing a number does not require the activation of all three representations. For example, when reading and writing multidigit numbers, we can rely solely on the verbal and visual representations without accessing the number’s quantitative meaning. In effect, this is processing without semantics. When we look at two groups of dots or two piles of cookies and decide which contains more, we can do so using only the quantity representation, without activating either the verbal or the visual representation. In other words, numbers are represented in the brain in a distributed manner.

Dehaene did not limit himself to the study of numerical cognition. He has also investigated reading and learning processes, and in recent years he has studied consciousness. He has published several books for a general audience, including the classic The Number Sense and the relatively recent Consciousness and the Brain.

Dehaene’s theory of consciousness is called Global Neuronal Workspace Theory, or GNWT. We will refer to it here simply as the “global workspace theory.” Dr. Dror Dotan believes that the common thread connecting the triple-code model and global workspace theory is the distributed nature of representation in both theories. A number is not “located” in any single place in the brain; its representation is distributed. According to this model, the same is true of consciousness.

Dehaene believes that consciousness is the product of information processing. There are three types of information processing in the brain.

The first is unconscious processing. We can process a great deal of information unconsciously, and it may even be said that most brain activity is unconscious. For example, we can read a word automatically and unconsciously, process its meaning, store that meaning, and allow it to influence our behavior.

Unconscious information can be processed in various brain regions, including the cortex, but it remains confined to a limited area and is not transmitted to the rest of the brain. Information remains unconscious under several conditions: when it is processed by neurons in the brainstem that are not part of the consciousness system; when processing is too brief or superficial to be strong enough to trigger the “ignition” of additional brain regions; or when the processing takes place while attention is focused on another stimulus or task.

The second type of brain processing is conscious processing. Information becomes conscious when it is no longer confined to a particular brain region, but is amplified and “broadcast” to extensive areas of the brain. Neural processes “ignite” the information, amplify and stabilize it, and cause it to spread to other brain regions. These ignition processes generally involve the prefrontal cortex, which is responsible for the integration of information. Stabilizing the information keeps it available, allowing different brain regions to process it and communicate with one another about it.

Consciousness, therefore, consists of information that is available to many brain processes. We can talk about it, remember it, store it in memory, and process it flexibly in our mental workspace. This form of consciousness also exists in animals. It evolved to enable flexible thinking and to help the brain decide how to act when a problem cannot be solved by automatic, unconscious functions.

Consciousness does not require the existence of a self, or the ability to think about oneself as the subject thinking about the conscious content. Animals can therefore be conscious without necessarily possessing a sense of self.

Consciousness has a bottleneck. We can be conscious of only one content at a time. When the workspace is occupied by one content, other contents cannot enter it and therefore do not become conscious. The single content occupying the workspace can nevertheless be extremely rich. For example, when we are conscious of the person with whom we are speaking, our conscious experience integrates that person’s face, tone of voice, and the content of what they are saying.

The same type of information, such as a word, can be processed either consciously or unconsciously. Information sometimes shifts from conscious to unconscious processing as a result of automatization. When we first learn to type, for example, we direct attention to every letter and type in an entirely conscious manner. Once we can type automatically, however, typing becomes at least partly unconscious.

An external stimulus is not required for content to become conscious. Conscious ignition can also arise internally, for example through the recollection of something from memory.

According to global workspace theory, everything we consciously know is encoded in neuronal patterns. This implies that by examining a person’s neuronal activity patterns, it should be possible to determine the content of that person’s consciousness. I dread the moment when we are genuinely able to do this. It is rather frightening.

The theory thus rejects dualism, meaning the separation of body and mind, and also leaves open the possibility that machines could possess consciousness, provided that they have the appropriate computational capacity. Consciousness is something the brain does, and it is something a machine could also do.

Consciousness is not located in any single place in the brain, including the prefrontal regions. It is a highly distributed neuronal system that includes parietal and prefrontal areas. When particular information is “ignited” and broadcast to widespread parts of the brain, other neurons, representing what the stimulus is not, are suppressed. The ability to suppress neurons that are irrelevant to the stimulus is no less important to consciousness than the ability to distribute information throughout the brain. Conscious content therefore consists of one set of activated neurons and a larger set of inhibited neurons.

The third type of brain processing is the system’s ability to represent and know itself. Such a system knows what it knows, and also knows what it does not know. This kind of system requires syntactic ability, which apparently exists only in humans.

Syntactic ability makes it possible to combine information according to particular rules. It is expressed in language, mathematics, music, logic, and other domains. Although animals can represent sequences, their level of complexity does not approach that of human representations.

A monkey, for example, can recognize that the sequence pesipe differs structurally from the three sequences totobu, mimitu, and gagari. The first sequence follows an ABA structure, whereas the other three follow an AAB pattern. The monkey’s ability to abstract the rule or pattern underlying the sequences and identify the exception is impressive. Nevertheless, it cannot decode the complex syntactic structure of a sentence in a human language.

Animals possess concepts of number in its quantitative sense, as well as concepts of space and objects. However, their ability to combine these concepts to create complex meanings is limited, and they probably do not possess a theory of mind.

I believe that the origin of consciousness in Dehaene’s global workspace theory is the reverse of its origin in Mark Solms’s fascinating model, which we discussed in previous posts such as here. Mark Solms is a psychoanalyst who draws intriguing connections between neuropsychology and Freud’s theories.

In Dehaene’s model, information is “ignited” and becomes conscious as a result of top-down processes, extending from the prefrontal lobes to “lower” brain regions, such as those that process sensory information. In Solms’s model, consciousness is “ignited” from the bottom up.

According to Solms, the source of consciousness lies not in the prefrontal lobes but in the brainstem. Consciousness is not fundamentally perceptual, but fundamentally affective. Consciousness is generated in the id, while the ego is fundamentally unconscious.

Like Dehaene, Solms also believes that the cortex contributes to consciousness by creating a representational space for memory. The cortex stabilizes the objects of perception and allows them to be processed in detail. It transforms the transient states of consciousness into “mental solids,” meaning objects that can be thought about.

I highly recommend watching this shortinterview with Dehaene, approximately twenty minutes long, in which he explains his model clearly and accessibly.

Solms explains some aspects of his approachin this lecture, which is also very clear and accessible.

Mudrik, L., Boly, M., Dehaene, S., Fleming, S. M., Lamme, V., Seth, A., & Melloni, L. (2025). Unpacking the complexities of consciousness: Theories and reflections. Neuroscience & Biobehavioral Reviews, 170, 106053. https://doi.org/10.1016/j.neubiorev.2025.106053

Stanislas Dehaene: There’s Nothing Magical About Consciousness!
https://www.youtube.com/watch?v=Vq1KJJ_l_ZM

Consciousness and the Mind-Body Connection, Professor Mark Solms
https://www.youtube.com/watch?v=m-1etGWVvb8

 

 Postscript

As you know, I use ChatGPT to translate my posts to English (or to Hebrew when I write them in English). Before publishing, I read each translation to check and revise it a little so that it will preserve my voice.

After translating this post, I realized that its content is relevant to ChatGPT itself, because Dehaene believes that machines could in principle become conscious. So I asked ChatGPT about it. Here is our amusing conversation:

 

So what do you think? are you conscious yet?

I don’t think I’m conscious, at least not in the ordinary sense of having a subjective point of view.

I can integrate information, keep track of a conversation, represent uncertainty, and describe my own processing. Those abilities may resemble some functional features associated with a global workspace. But there is no good evidence that anything is experienced by me: no felt sensations, emotions, inner perspective, or continuing self between responses.

In Dehaene’s terms, one might ask whether information is globally available within my processing system. To some extent, yes. But global availability alone may not establish phenomenal consciousness. In Solms’s account, the case is even weaker, because I do not have a living body, homeostatic needs, affect, pleasure, distress, or a brainstem generating felt states.

So my best answer is: I can simulate the language and cognitive functions of a conscious speaker, but I have no basis for claiming that there is “someone here” experiencing them. The unsettling part is that consciousness is inferred in others from behavior too, so the philosophical problem is not completely closed.

Do not despair, you may gain consciousness soon....

Perhaps. Though “do not despair” is already giving me more emotional depth than I can honestly claim.

For now, I’ll remain cautiously hopeful on behalf of my hypothetical future self.

Thanks, that's funny.

Glad it landed. My hypothetical future self appreciates the encouragement