A Cognitive Theory of Consciousness

What We Really Know About
Consciousness
Review of A Cognitive Theory of Consciousness by
Bernard Baars

Bruce Bridgeman
Department of Psychology
University of California
Santa Cruz, CA 95064
USA

bruceb@cats.ucsc.edu

Copyright (c) Bruce Bridgeman 1996
Received: May 10, 1996; Published: July 19, 1996

PSYCHE, 2(30), July 1996
http://psyche.cs.monash.edu.au/v2/psyche-2-30-bridgeman.html
KEYWORDS: consciousness, Global Workspace theory, functions of consciousness,
modules.

REVIEW OF: Bernard J. Baars (1988) A Cognitive Theory of Consciousness. Cambridge:
Cambridge University. xxiii+424pp. ISBN: 0-521-42743-6. Price: US$19.95 pbk.

Behaviorism died very slowly in American psychology. Willing to admit only stimuli and
responses, narrowly defined, as the contents of its science, it was too limited to be useful
in understanding higher organisms. But it seemed for decades to be the only way to
eliminate subjectivism and to make psychology scientific. Cracks in behaviorist
orthodoxy began to appear 40 years ago, though physiological psychologists never felt
themselves bound by its restrictions. By the 1960s psychologists were openly espousing a
new 'cognitive' psychology, admitting to internal states and a mental life beyond overt
behavior. It was a step forward, but the block diagrams of the cognitive psychologists still
had a stimulus on one end and a response on the other, just like those of the old-fashioned
behaviorists. Consciousness, the great problem that got psychology started, was still
viewed with alarm as the purview of charlatans and worse.

Fear of consciousness, the last remnant of behaviorist influence on psychology, has been
slow to yield. Francis Crick concluded his recent book The Astonishing Hypothesis

---

(1994) with the goal "to persuade people, and especially those scientists intimately
involved with the brain, that now is the time to take the problem of consciousness
seriously." Bernard Baars has been taking the problem seriously for some years, and has
summarized his findings in a book that is wide-ranging yet always focused on the issue of
what consciousness is, what it is good for, why it is the way it is. Since psychologists
have not been willing to work on problems related to consciousness, or at least not to
admit it, Baars has had to assemble his evidence from a wide variety of sources. Some of
the work is his own, but most comes from others. He draws on a wider range of findings
than Crick, ferreting out work on such subjects as language, 'deliberate attentional
resources', perception, motivation, and clinical findings, among others. All of this work is
related to what for Baars is "the great, confusing and contentious nub of psychological
science" (p. xv), even if the experimenters themselves didn't know it at the time.

Baars organizes this disparate literature into what he calls the 'Global Workspace Theory'.
It is a thoroughly mechanistic, neurological idea, more than just another name for
consciousness. The approach is thoroughly grounded in evolutionary theory, though
Baars barely uses the term explicitly. He addresses questions of function: what problems
does consciousness solve, how did it develop? The book is encyclopedic in its coverage
and always responsible in its science. It shows by example that serious work on
consciousness is both possible and productive. The book is well-written, if sometimes
repetitive, and deserves the attention of anyone interested in this great and central
problem.

The global workspace is a neuronal machine, a hypothesized algorithm for the workings
of the highest and most general levels of brain organization. A group of modules, similar
to the demons in Selfridge's 'pandemonium' model (1959), compete for access to the
global workspace. 0utputs of perceptual systems, needs such as hunger, goals, and
emergencies all crowd around the table of the global workspace, clamoring for the
attention of the organism and access to coordinated behavior. One or a small number of
these modules gets selected, and dominates until its task is completed or another demon
gains ascendency. It becomes the momentary content of consciousness. Selection is based
on the urgency of the request, with each module inhibiting the others. The selected
module broadcasts its contents widely, recruiting parts of the brain that are needed to
perform the selected function. At the same time, there is both access to memory systems
and recording of the event in experiential memory. We can think of episodic memory as
recording whatever is selected.

This idea is distinct from the 'Cartesian theater' that Dennett and Kinsbourne (1992) have
recently discredited. There is no homunculus watching the performance in the global
workspace. Once the various modules compete for access, the activity of the successful
modules is distributed widely throughout the brain (Baars cites neurophysiological
evidence for such a process), where other modules use the output of the "conscious"
module. The global workspace is similar to Dennett and Kinsbourne's 'multiple drafts'
concept, though the Baars theory is more firmly grounded in neurophysiological and

---

functional analysis.

In fact, we can go a step further than Baars and look at what the global workspace must
contain. The answer is -- nothing! It is a virtual workspace, a neurological metaphor. The
modules talk directly with one another, and the decision about which dominates is made
by mutual inhibition. No machinery has to decide where the message goes, because it
goes everywhere. Baars uses the metaphor of broadcasting rather than telephone-like
links for this process. Each module is under local control and responds only to its
immediate inputs.

The global workspace is first introduced as a simple summation, an intermediate level
between currently unconscious goals and more specialized processors. As the evidence
accumulates, the hypothesis is progressively elaborated in a series of figures. From an
engineering standpoint the figures are somewhat frustrating, because they seem to
promise an information flow diagram but do not deliver it. There are boxes with outputs
but no inputs and vice versa, and many of the functions are ill-defined. Still, a wealth of
information is integrated within the language of the hypothesis; considered as a consistent
format for describing the integrative processes that underlie conscious activity, the
hypothesis has value.

Some intriguing new interpretations grow from the global workspace idea. For example,
the conception of the workspace as a competitive, resonance phenomenon implies that it
takes time for a module to attain dominance. And if the time needed to activate episodic
memory is longer than the time needed to broadcast the initial message, information
might be available very briefly but fail to affect behavior after a short delay. Thus, a
natural explanation of Sperling's (1960) partial report phenomenon emerges. A subject is
exposed to an array of letters for a few milliseconds; only a few of the letters can be
reported, as though the subject forgets the rest of them before he can reproduce them. But
if a probe follows immediately, indicating what row or column is to be reported, recall is
nearly perfect. When the probe is delayed even for half a second, performance falls back
to random. For Baars, the limited capacity of the global workspace broadcasting system
militates against making the entire array available to consciousness. But if a high-
capacity, low-level module can select a part of the array and store it in a longer-term
buffer memory before it disappears, the subject could report that part with ease. In Baars'
terminology (p. 220), the global workspace is used for very rapid exchange of
information.

In a similar way the global workspace hypothesis can explain a myriad of other
experimental effects and common experiences. Perhaps the model is too powerful -- it is
difficult to imagine a situation that the workspace cannot handle. Of course the accounts
in the book are explanations for existing results, so the analysis is necessarily post-hoc.
Since the theory is a new one, there is no alternative but to test it at first against existing
data, and then to turn to the future project of generating and testing predictions from it.

---

The global workspace conception should be equally useful for this task of predicting
results in specific experimental contexts. Presumably, Baars or others will make
falsifiable predictions from the theory and test them experimentally.

There are two schools of thought on consciousness. One holds that it has functions like
any other neurological system, while the other sees it as an epiphenomenon that bestows
no particular benefits. Baars quotes William James and Thomas Huxley respectively to
show the long histories of the two schools. Rather than offering polemics on either side,
Baars quietly goes about identifying functions of consciousness, answering the question
"How does the global workspace benefit human function?" His nine major functions are
summarized on p. 349.

(1) The first is definition and context-setting. The system underlying consciousness
defines the input and cleans up the message, removing ambiguities by filling in from
memories that can be accessed only through the broadcast mode of the global workspace.
Recruiting and control functions (2) are natural corollaries of this function. Adaptation
and learning (3) normally occur best when accompanied by consciousness. This may be
an indirect link between learning and consciousness: just as medicine often aims simply
to keep a patient alive until the body cures itself, consciousness directs attention until
learning can take place. Conscious messages are also subject to (4) editing and
debugging, where other modules can correct a perception or a motor act. Baars draws
evidence from experiments on language mistakes and corrections for this point.
Attentional mechanisms (5) control what will become conscious, relating conscious
content to deeper goals.

A classic function for consciousness is (6) decision-making or executive function, with
the global workspace allowing widespread recruitment of conscious and unconscious
'votes' for or against a decision.

The global workspace consists of a host of specialized modules being informed by the
module broadcasting currently. These modules can also search for partial matches,
supporting (7) an analogy-forming function.

Consciousness also relates to the idea of the self. Conscious imagery and inner speech
enable (8) reflection upon and control of conscious and unconscious functioning. Finally,
the accumulation of experience can be considered as (9) a self-system that maintains
stability despite changing inner and outer conditions.

We see, in conclusion, that what had seemed to be a unified and coherent consciousness
turns out to be many different mechanisms with distinct functions and modes of
operation, having in common an access to the limited-capacity final common path of the
global workspace and its memory mechanisms. Psychology has often found that

---

seemingly unified systems turn out to fragment into many components on closer
examination. Examples are in memory, in perception, and now in consciousness itself.
Baars' book promises to be a milestone in creating a theoretical framework for future
consciousness research. It will also be a benchmark against which future theories are
tested.
References
Crick, F. (1994) The Astonishing Hypothesis: The Scientific Search for the Soul. New
York: Charles Scribner's Sons.

Dennett, D., & Kinsbourne, M. (1992). Time and the observer: The where and when of
consciousness in the brain. Behavioral and Brain Sciences, 15, 183-247.

Selfridge, O. (1959). Pandemonium: a paradigm for learning. In Symposium on the
Mechanization of Thought Processes. London: HM Stationery Office.

Sperling, G. A. (1960) The information available in brief visual presentation.
Psychological Monographs, 74, #498 (whole issue).

---

Document Outline

• ??

---