Global Workspace Theory: An interview with Bernard Baars

An interview with Bernard J. Baars

RAMSØY

Professor Baars, since your 1988 book, you have suggested that consciousness is built on a “global workspace” architecture. Could you explain exactly what you mean by this?

BAARS

If you carefully line up matched conscious and unconscious processes, like explicit vs. implicit cognition, conscious vs. automatic aspects of control, implicit vs. explicit learning, all those standard topics in psychology, what you find is this: conscious processes are ALWAYS remarkably limited in capacity, extremely wide in range of contents, and highly context-sensitive. The fact that you are reading right now CONSCIOUS words, that must be interpreted in many layers of linguistic context, that are not possible to understand if you are listening to music at the same time, and that can range in content among 100,000 different English words, illustrates that point in your own experience. So phenomenology and cognitive psychology show exactly the same thing.

On the UNCONSCIOUS side of the ledger, we find not limited capacity but stunningly large capacity when you consider all kind of unconscious functions together; in any specific unconscious process we find very limited Grand integrative theory has been frowned on specialization (like syntax vs. color vision); and we find a lack of context-sensitivity. Take an everyday example. Most of us find that when we are distracted, we sometimes ‘go on automatic.’ We wander into a room and forget why we are there. We sometimes brush our teeth when we meant to shave. James Reason in Britain has made a large collection of action errors of that kind, sometimes very funny ones. I used to induce them in the laboratory experimentally (Baars, 1992). They are all characterized by (1) distraction (meaning a loss of conscious focus), and (2) a decomposition of a normally context-sensitive unit into context-free components, like phonemes, syllables, or elements of action.

Well, first, notice how regular this pattern of evidence is. Many cognitive psychologists have studied aspects of it, but it has not been popular to study the evidence as a whole. That is because “grand” integrative theory has been frowned on. But the evidence has been there, right in front of our noses.

So that’s the first step. The second step is to ask, in standard scientific fashion, what kind of underlying mechanism could possibly account for these very robust regularities? If you were Copernicus (another Scandinavian!) you would say, well, maybe it’s like a giant clockwork centered on the Sun. And you’d be very unpopular, and write your book but not publish it in your lifetime. Fortunately we are in a slightly better position today. But new ideas are always unpopular.

How do we take the second step? What is our model for the solar system? It turns out, as it often does, that other people have already come up with it. Cognitive scientists talked about “society models” of the brain (demons, or simpletons, or whatever they were called) that were unified by a fleeting memory capacity that allowed many of the specialized knowledge sources to interact with each other. The original Global Workspace (GW) theorists were artificial intelligence researchers in Allan Newell’s group in the 1970s, who actually built a speech recognition system consisting of a “blackboard” with many specialized language processors that could post their hypotheses about the stimulus and then “vote” on which one was better. It was very successful. And then it was largely forgotten, or added to the standard tools that artificial intelligence people use.

My role in all this was to marry the GW notion with the evidence about This would mean a kind of global broadcasting in each hemisphere, with “movie frames” going by at 1/10 of a second consciousness. I wrote my first chapter on this in 1983, but even at that time Donald A. Norman had made similar suggestions. My first book on it is still the most extensive and rigorous explication of the evidence and theory, though it needs an update (1988). We assumed that everybody would immediately see the point, and we would move on. As it turned out, cognitive psychologists were still under the behavioristic taboo against consciousness, and largely evaded the topic. That was a big disappointment.

On the brain side there was much less resistance. A number of brain scientists, like Giulio Tononi, Gerald Edelman, Francis Crick, and many others, thought it was fairly obvious. People in Chris Frith’s wonderful Functional Imaging Laboratory in London found good supportive evidence, and Stanislas Dehaene developed a Neuronal Global Workspace model as a neural net interpretation of prefrontal cortex. My colleague Stan Franklin used the GW concept to develop the first really functional real-world application that also aims to account for normal human functioning, called IDA. So things are moving in the right direction, especially in cognitive neuroscience. If you look directly at the brain, it is natural to think that something like it must be true.

Obviously we have much, much work to do, to make it more rigorous, mathematically explicit, and to look at the connections with existing brain theory. The work of Walter Freeman is currently very interesting to me, as well as more conventional approaches. And I always hope that cognitive scientists will become more receptive.

RAMSØY

So it seems that the coupling of your cognitive theory of consciousness with neuroscience and artificial intelligence research has led to a fruitful elaboration of the theory. In what way would you say that the Global Workspace manifests itself in the brain?

BAARS

That’s the million Euro question! It turns out a lot of brain scientists have found interesting things along those lines. Here are just some.

Stanislas Dehaene (see his website) suggests that neurons in prefrontal cortex behave like a GW, with long-distance axons working to distribute executive decisions. It’s a nice neural net model, but perhaps too simple. From my point of There are two ways of broadcasting in the cortex to happen; cortico-cortical and cortico-thalamic connections view, it does not answer the question of specific sensory consciousness in vision, hearing, and so on. But it may be the right idea for “fringe” consciousness, the kind we experience in feelings of familiarity, feelings of expectation, some feelings of intention to act, and so on. We know from fMRI studies that prefrontal and anterior cingulate cortices are involved in feelings of knowing, like the tip of the tongue.

Walter Freeman (see his website) has a remarkable set of EEG findings, with very sophisticated mathematical analysis leading to very interesting results. Using both animal studies (with implanted grids of local electrodes) and now human studies (using large arrays of scalp electrodes), this is basically Freeman’s approach. The standard way of analyzing EEG is Fourier Analysis, which decomposes the waveform into a sine waves, effectively giving us the power at any given frequency range. However, Freeman uses the Hilbert Transform, which gives us instantaneous phase and amplitude information across a spatial grid laid across parts of cortex. He argues spatial analysis is the “code” of the brain, especially the cortical surface layers (the “dendritic feltwork” that feeds into pyramidal cells). Freeman believes that the brain has no way of decoding temporally firing codes, which is probably overstated, but it could be true for cortex.

When you do all that, you get an amazing finding, a three-dimensional surface showing what looks like chaotic dynamics. The vertical Z dimension is phase, electrodes in the X dimension, by electrodes in the Y dimension (since we have a flat grid on cortex). The flat surfaces between vertical “hills” in the figure show gamma range synchrony, which is now widely thought to represent active information processing, as different cortical regions are ‘bound” to each other. The ridges of hills represent a chaotic break-up of the gamma equilibrium, which happen almost instantaneously across entire hemispheres. The hills occur very regularly, at 100-200 msec intervals. There is still much work to do, to make the theory more rigorous, mathematically explicit and empirically oriented. This time domain is widely thought to be associated with consciousness. For example, perceptual fusion between different sounds or visual stimuli breaks down at about 100 msec. That is, we cannot consciously integrate events separated by more than about 1/10 of a second. What’s ALSO fascinating, however, is Freeman’s interpretation of the mountain ridges. He thinks they correspond to alpha wave activity, slowing in frontal regions to the theta range. Alpha has long been associated with “spacy” consciousness, when we’re not doing anything actively, staring into space. Theta is believed to signal hippocampal transmission of episodic (conscious) memories to cortex. It’s conscious recollection of past events.

So that makes a really neat story. It would mean a kind of global broadcasting in each hemisphere, with “movie frames” going by at 1/10 of a second.

A few reservations. We don’t know how that relates to the CONTENTS of consciousness, since Freeman argues that the chaotic dynamics of EEG reflect the MEANING of events, such as the prediction of reward. However, the dendritic feltwork of cortex drives cortical firing along pyramidal axons, which MAY convey the contents of visual consciousness, for example. Thus the Freeman mechanism could be a way to pace the contents of consciousness, to chop it into 100 msec frames.

A second caution is that Freeman’s work needs to be replicated by other labs, with various controls. If I had an EEG lab, I would jump at it!

So that’s all fine. There are two more possibilities.

One is that each sensory cortex is able to broadcast information to the rest of cortex. Freeman and others have found zero-lag correlations between sensory cortices, suggesting something like that may be happening. I favor something like that, because it would help to explain sensory consciousness, which I think is evolutionarily more primitive than prefrontal “fringe” consciousness. Prefrontal cortex, as you know, is distinctively large in primates. Even whales and dolphins, with very large brains, have expanded parietal, but not frontal regions.

There are two ways broadcasting in cortex to happen. One is cortico-cortical connections, both close and distant. The second is thalamocortical connections, or even more likely, cortico-thalamo-cortical ones, where a signal starts in visual cortex, for example, triggers neurons that synapse in the thalamus, which in turn relay activity to, let’s say, executive regions of prefrontal cortex. Murray Shanahan in London has a nice model of how that could happen.

My bet is that both of those exist. Gerald Edelman and Joseph Gally have argued in a recent PNAS paper that biological systems ALWAYS use functional redundancy, like a well-designed robot or airplane. The space shuttle has four rather old IBM computers that are doing exactly the same computations. If one is out of step, the other three vote against it. So you could have bugs and breakdowns, but most of the time things would work all right. (Edelman calls it “degeneracy.”) You can see it at all kinds of biological levels, from gene networks to language. It’s a very impressive point, and having multiple ways of integration and broadcasting in the brain would be such a functionally redundant mechanism.

I know this is a bit complicated, but I would guess that all of the theories mentioned above have a grain of truth.

RAMSØY

So in this sense, there may be more than one mechanism in the brain that supports a global workspace underlying consciousness. Cortico-thalamo-cortical connections, for example, are thought to underlie the pathological loss of consciousness in absence seizures as well as persistent vegetative states. However, the global workspace model of consciousness can be thought of in two ways: on the one hand, one may claim that the globalisation (in the brain) is consciousness per se, meaning that the widespread areas that are activated during a conscious representation of a given item are all necessary and sufficient properties of the conscious perception. On the other hand, one may claim that globalisation is the spread of information that is conscious, meaning that globalisation is a mechanism that gives widespread areas in the brain access to conscious content, without themselves being contributing parts to the conscious perception themselves. Do any of these suggestions fit into your view on a GWT of consciousness?

BAARS

Yes, you have to remember that already we have several mechanisms for learning, for emotion and motivation, and so on. Consciousness is likely to go back at Freeman’s work needs to be replicated by other labs least 100 million years to early mammals, which probably had well-developed thalamocortical systems. It may go back earlier. So there is plenty of evolutionary time for multiple mechanisms to develop, as they should for vital functions. The heart has redundancy, the lungs have redundancy, why not the brain basis of consciousness? Now that doesn’t mean that you can’t knock out consciousness, as you point out, by absence seizures, or by lower brainstem damage, and so on. But we see amazing capacity for recovery at times.

On the question of “necessary vs. sufficient.” My own guess is that this is just too early to expect any theory to be sufficient. Science has just returned to consciousness after an absence of about a hundred years. We know some solid facts, but our ignorance must be very great indeed. What we can hope for is some hypotheses about necessary conditions — for example, we know we need brainstem neuromodulation to wake up. That’s a necessary condition. But what is happening in cortex when it changes state is not completely understood.

I always find it useful to look at the history of gravity in physics. We’ve had good ideas about gravity since before Newton, four centuries ago. But even today advanced physics is looking for gravity waves and particles. That question, as far as I understand it, is just undecided. That’s four centuries and we still don’t know the sufficient conditions! But physicists obviously know a lot of necessary ingredients.

Being a scientist allows you to escape a lot of philosophical questions. Most of the time the answer to metaphysical problems is: “I don’t know! Maybe we can find some evidence to move toward an answer. Maybe we have to wait until the fog rises a little more. Meanwhile, let’s study those situations that will give us the best chance to find out more.”

There was a great and passionate debate around 1900 about the metaphysical hypothesis of vitalism. Henri Bergson and other philosophers told biologists that they could never discover the vital essence (élan vital) that distinguished living things from rocks and metal. Meanwhile a little monk in Russia was studying inheritance in peas, and a few poorly funded biology teachers were looking at fruit flies. A hundred years later nobody talks about the élan vital, and everybody knows about the giant revolution that came from those peas and fruit flies. That revolution is about to change all of our lives. Sadly, all the noise about the vital essence was wasted effort.

© 2005 TZ Ramsøy and BJ Baars

Bernard J Baars is a Senior Fellow in Theoretical Neurobiology at the Neuroscience Institute, San Diego, USA

He is the author of numerous scientific papers on consciousness and cognitive neuroscience. He is also the author of several books, including “A cognitive theory of consciousness” (see link below), “In the theater of consciousness” and “The cognitive revolution in psychology”. Recently, Baars has co-edited the exhaustive volume “Essential sources in the scientific study of consciousness” (see a recent review in SCR).

Selected SCR articles by Bernard J. Baars

1. A specific drug for consciousness?
2. The global Brainweb – An update on Global Workspace Theory
3. Hysterical conversion, consciousness and the brain
4. How conscious experience and working memory interact (2003)
5. The conscious access hypothesis: origins and recent evidence (2002)
6. In the Theatre of Consciousness (1997)
7. A cognitive theory of consciousness (1988)