Stage Effects in the Cartesian Theater:
A review of Daniel Dennett's Consciousness Explained
Kevin B. Korb
School of Computer Science and Software Engineering
Monash University
Clayton, Victoria 3168
Australia
korb@bruce.cs.monash.edu.au
Copyright (c) Kevin B. Korb 1993
PSYCHE, 1(4), December 1993
Previously Held: http://psyche.cs.monash.edu.au/v1/psyche-1-4-korb.html
Keywords: philosophy of mind, qualia, functionalism, multiple
drafts model, Cartesianism, folk psychology
1. Dennett's Multiple Drafts Model of Consciousness
1.1 `We're all zombies. Nobody is conscious' (Dennett 1991,
p. 406) is an assertion Dennett is actually brought to make
in his attempt to maximally provoke his readers; his title
serves a similarly provocative role.
<1>
1.2 Judging by such expressions, Dennett undoubtedly would
like his work to be even more provocative than it actually
turns out to be. The main thrust of Consciousness
Explained is
to apply a widely accepted thesis about the
relation between mind and brain---non-homuncular
functionalism---in a program of philosophical therapy to rub
away a variety of puzzles raised by both philosophers and
experimentalists about consciousness. I believe that the
central thesis will be relatively uncontentious for most
cognitive scientists, but that its use as a cleaning solvent
for messy puzzles will be viewed less happily in most
quarters. I will start by briefly presenting Dennett's
views on functionalism and method and then present some
concerns about Dennett's sanitation projects.
1.3 Functionalism is the theory that evolved out of
mind-brain identity theory, the thesis that mental states or
processes are just identical to some particular brain states
or processes. That theory, introduced by J. J. C. Smart and
U. T. Place in the 1950s, suffers from the objection that
mental states cannot be shared since the physical goo of our
brains cannot be shared. Most obviously, Martians cannot
have pain states if they do not have brain states (i.e.,
their `brains' may be made out of something other than
neural goo; cf. D. Lewis, 1980); but even more radically,
you and I cannot have the same pain states since you and I
cannot share brain states. Theorists, in response to the
pressure of such arguments, have tended toward
functionalism, or the token-identity theory: mental states
are to brain states as types are to tokens---as say having
one dollar is to having a particular dollar bill (or to
having a Susan B. Anthony coin). Although mental states are
not identical to brain states then, they supervene upon
brain states; that is, pain (or whatever) is realizable via
some kinds of brain state. This does not rule out other
possible varieties of realization; for example, Martians may
realize a pain state by way of some fluid hydraulic system,
rather than a neural system. What identifies it as a pain
state is not the physical token state that constitutes it,
but the functional (more generally, causal) relations which
that token state has to other token hydraulic states and to
the system's overall behavior. Dennett, together with most
cognitive scientists (but certainly not all), is a believer
in some flavor of functionalism.
1.4 He is also opposed to homuncular theories, again
together with most commentators. A homuncular theory is one
that proceeds to explain some cognitive (conscious) ability
or process by first producing some functional analysis that
appears to make progress for a time, but then, when things
get messy, simply invents an internal agent---a
homunculus---which turns out to have all the abilities that
needed explaining in the first place. Since the homunculus
in its turn will need to be explained, there is an immediate
threat of infinite regress. The most extreme variety of
homuncularism is the dualist ghost in the machine: no matter
how good an explanation of mental function you may produce
in terms of brain processes, a committed dualist will
complain that it is incomplete because it fails to explain
that aspect of mentality which is non-physical. If someone
offers a functional account of natural language
understanding, for example, a homuncular dualist will insist
that in addition to all of that (whatever it may be),
there must be a creature in the middle of it all that has
the real understanding, namely the Mind. Neither dualism
nor homuncular functionalism are sufficiently appealing to
have many adherents nowadays: where writers are moved to
issue an explicit opinion, they are very largely negative
opinions.
1.5 Despite this, Dennett shows that the homuncular concept
retains a powerful grip on the imaginations of many, perhaps
most, cognitive scientists. While explicit dualism and
homuncularism are (no doubt properly) `endangered' theses, a
great many theories and judgments advanced by cognitive
scientists rely at some point upon there being a magical
place in the head where everything comes together---in what
Dennett calls the Cartesian Theater. This concept is
pernicious in a variety of ways. For one thing, it leads to
lazy analysis: if we can rely upon some arbitrarily complex
central process to clean up our functional loose ends, we
needn't be very careful about specifying whatever functional
processes we do provide. But worse, this Cartesian
Materialism (functionalism with the Theater at the center)
again leads to infinite regress: if there is a theater where
consciousness is `projected', then there must be an observer
viewing the projection (else why bother with the theater?).
As before, we will find it difficult to understand this
observer: if the theater and its audience are needed to
understand conscious processes, then an `inner' theater and
`inner' audience will be needed to understand the observer,
and so on. But if the theater and its observer are not
needed to understand conscious processes, then why introduce
them in the first place? As Dennett notes, the best place
to stop an infinite regress is usually at the beginning.
1.6 It is not so difficult to imagine that the brain---and
consciousness---might be analyzable into a set of processes
with no dominant center. Certainly it is not so difficult
to build computer programs which behave in complex ways,
including learning and adapting to a complex environment,
without there being a `central authority' which decides on
every action: Oliver Selfridge provided a model for this in
his program Pandemonium in the 1950s (Selfridge, 1959). In
similar spirit, Dennett proposes in opposition to Cartesian
Materialism his Multiple-Drafts model of consciousness: the
idea that consciousness is more or less a continuing (but
not continuous) flow of drafts of text; the texts are
constantly being edited and re-edited, passed from one
process to another. Sometimes they initiate speech,
sometimes they are stored in long-term memory, other times
they are entirely forgotten.
1.7 What counts as the content of consciousness on this
model is by and large up for grabs. But even such vagueness
may be appropriate when we reflect that we do not have any
clear idea of what to count as the content of consciousness
independent of our theories of consciousness. There is one
criterion for such content that has been widely accepted
among investigators: what a subject reports expresses some
conscious content. This is, as Dennett remarks, what we
normally (but not always) regard as sufficient evidence
for a human to be, or have been, conscious of the content
expressed. This gives us a counterfactual criterion of
consciousness, which we may call the `probe criterion': if
an agent were to inquire of the subject what she or he was
aware of (or whether she or he was aware of X), then if the
subject would have reported being aware of X then the
subject was conscious of X. This criterion, however
incomplete, is useful. Indeed, Dennett uses it to point out
that an example very often employed by philosophers to argue
that certain complex cognitive tasks are performed
unconsciously (e.g., by Dennett himself, 1969; and by me in
Korb, 1991) is unsuccessful: in `automatic driving' people
appear to have been unaware of their `actions' for a few
minutes while driving and simultaneously day-dreaming. But
Dennett notes that if someone were to have inquired about
say the bridge just crossed during the reveree, the driver
would very likely not respond `What bridge?' Automatic
driving is better explained by rapid forgetting than by
unconscious behavior.
1.8 The adoption of this criterion does not mean that
Dennett is (or should be) passively acceptive of the
content of subjective reports about the contents of
consciousness. Introspective reports are not random
mumblings: they express some conscious content; but at the
same time they may report a quite different content. The
most obvious example is a lie about one's own beliefs: such
a lie reports one content (what you would have your
interlocutor believe) while expressing another (your
internal belief state). The larger problem with
introspective data has been the ease with which people
confabulate mental events that apparently `aren't there' at
all. But this and related difficulties do not imply that
introspective reports have no scientific value, as some
extreme behaviorists would have it. The behavior of making
such reports is itself as factual as any behavior scientists
are otherwise called upon to explain. The content of
introspective reports typically describes---in what Dennett
calls the subject's heterophenomenology---a `mental world'
that is more or less coherent, bears certain relations to
what is known about neuroscience, and has some pattern of
similarities and dissimilarities with the introspective
reports of others. What cognitive science should be aiming
at is producing an explanation of these facts that is
concise, that covers as much of the data as possible, and
that stands up well to futher tests. This is just, of
course, what we demand of any scientific theory. And we
should expect that the resultant theory will end up
denying the face-value interpretation of some kinds of
introspective reports, even when those reports are repeated
by all comers. This is again strictly in parallel with
other sciences: almost all of our theories describe some
cases where appearances are simply deceptive (e.g., almost
every measurement of a star's location will be misleading
within some small angular distance from the sun).
1.9 I have summarized the main elements of Dennett's
Multiple Drafts model of consciousness and the methodology
he proposes that the cognitive sciences adopt for its study.
While not uncontroversial nor unoriginal, I expect thus far
the story is congenial to most functionalists. Less so will
be Dennett's applications of the theory to various puzzles
about consciousness: despite suggestions to the contrary,
they do not simply fall out as corollaries to the Multiple
Drafts model. A few examples follow.
2. The Phi Phenomenon
2.1 The `phi phenomenon' refers to the apparent motion
created by rapidly presenting a series of still pictures
each of which shows a slight displacement of some image---the
basis of motion pictures. This is an interesting
psychological effect made even more interesting in an
experiment proposed by Nelson Goodman and conducted by
Kolers and von Gruenau (1976): when lights are successively
illumined in a series (as in some neon signs) we seem to see
a single light in motion; what would happen if two
successively illuminated spots were of different colors, say
green and red? What would happen to the color of `the' spot
as it moved from the first to the second position? The
experimental answer (p. 114): `The first spot seemed to
begin moving and then change color abruptly
in the middle
of its illusory passage toward the second location,' which
it reached 150 msec later (when the red spot was
illuminated). This raises a few strange questions: how can
the apparent colored spot move in the right direction,
toward the second location,
before we have any idea
(input) of where the second location is? And, how can the
green spot change to red in the
middle of this traversal
before we have any idea (input) that red is what it is
supposed to become?
2.2 Dennett discusses two possible models accounting for
this phenomenon: Orwellian and Stalinesque revisionism. The
Stalinesque model supposes that there is some pre-conscious
editing of the input going on. In particular, the green
image first received is held up in the editing room until
the red image becomes available. The pre-conscious editor
then rapidly paints some appropriate intermediate images,
splices it all together and passes the result on to the
Cartesian Theater for presentation to Consciousness. The
Mind only sees what the Stalinesque editors want It to see.
The Orwellian model operates in reverse: Consciousness
receives the green image; then It receives neutral
background images; subsequently Consciousness receives the
red image. The Orwellian editor, realizing the discrepancy,
quickly revises the operative history of the episode so that
there are appropriate intermediate states. When
Consciousness needs to review history, say to answer some
question, It finds that the green spot changed to red during
an intermediate traversal of the two locations, and reports
that as fact. Now, of course, the Orwellian history isn't
written down; it just seems to Consciousness that such a
color change was actually seen. And what happened to the
initial intermediate images that showed no such traversal or
color change? They have simply been lost, forgotten in the
shuffle.
2.3 So which account sits better with the Multiple Drafts
theory? To make a fair assessment we have to eliminate the
references to `Consciousness' and `Mind'; but neither
account strictly requires them anyway. Dennett claims that
there is then nothing to choose between the two models: when
we take a sufficiently low-level view of what's going on in
the brain, the where and when of consciousness becomes
blurry, and so there is nothing to choose between a piece of
retroactive Orwellian editing and proactive Stalinesque
editing. As there is no `finish line' (Cartesian Theater)
against which to measure whether the editing happens before
or after Consciousness, there can be no fact of the matter
about which of the Orwellian or Stalinesque model is the
true one: `there is really only a verbal difference between
the two theories' (p. 125). Indeed, there is no fact of the
matter about when anything becomes conscious.
2.4 Here I think Dennett is simply wrong. Recalling the
probe criterion of consciousness, we can ask what the
subject would report were a query inserted between the green
and red stimuli. Would anyone report that the spot was
in-between the two end points? Such a response would
require more than retroactive editing, it would require
retroactive causation! That is out of the question, but
it doesn't follow that the green flash, or neutral
background, would be reported: perhaps they would have been
held up in an editor's room. Of course, we cannot in any
case squeeze either an inquiry or a response inside a
150msec window, factually or counterfactually. But the
verbal probe was always understood to be a crude device: we
can surely suppose that our low-speed probes and their
low-speed responses may come to be known to be highly
correlated with some specific brain processes which are
themselves high-speed processes. Let's simplify and suppose
that low-speed reports of green sensations are highly
correlated with some high-speed brain process Q. If Q is
then observed during these tests before 150msec has elapsed,
then this is clear evidence in favor of the Orwellian model
and against the Stalinesque model in the same way that the
probe criterion provides evidence against the idea that
automatic driving is an unconscious process. It is simply
false that the two models are merely `verbally different'
from one another: the confusion of an inability to test one
theory against another now with an in principle
inability to test between them is a fundamental confusion.
It is the same as confusing operationalism <2> with the
demand that new theories must make some possible
difference in our observations, which is prerequisite to
there being any possible experimental science deciding
between those theories. (Dennett is similarly confused
about verificationism; cf. p. 403.)
2.5 The idea that there is no `when' to consciousness is
also wrong. The blurriness of boundaries is no good
argument for the non-existence of boundaries. There is no
clear boundary between baldness and hairiness, but that is
no consolation to those who are bald and don't like it. So
long as there is no Consciousness Cell---some specific neuron
that fires when and only when a subject is conscious (and
who could believe that?)---then there will be some
vagueness about where and when conscious processes are
occuring. But if conscious processes indeed supervene on
physical processes, there is nonetheless a where and when to
consciousness: all physical processes occur in
space-time.<3>
3. Folk Psychology
3.1 Dennett has previously argued that consciousness,
whatever it may be, is so different from our ordinary
conception of it that we ought to abandon the concept
(Dennett, 1979). Here, while not repeating that argument,
he claims that the ordinary notions of belief, desire,
etc.---the mental predicates of `folk psychology'---ought to
be abandoned: `we may use the oversimplified model of folk
psychology as a sort of crutch for the imagination when we
try to understand self-monitoring systems, but when we use
it, we risk lapsing into Cartesian materialism' (p. 320).
The eliminative materialists (e.g., Churchland 1981) have
similarly argued that `folk' concepts of psychology will
someday be replaced by some as-yet-unconceived clean,
scientific concepts of neuroscience. Churchland however,
had no specific objections to `believe', `want', and so on
other than his prejudicial
conviction (yes: belief) that
these terms would not show up in the not-yet-conceived laws
of future neuroscience. Dennett, on the other hand, has a
specific, but mistaken, objection to `belief'.
3.2 Dennett presents, sympathetically, David Rosenthal's
analysis of consciousness in terms of belief (Rosenthal,
1986, 1990). We may express our beliefs quite
unconsciously, without our knowing that we do so, indeed
without our knowing that we have such beliefs (such as some
racists who sincerely believe that they are not); but we
cannot report our beliefs (sincerely) without also being
aware of those beliefs. If we call occurent beliefs (those
we are aware of now) thoughts, in order to report a belief
we must be thinking about it; that is, we must have an
occurent, second-order belief about the belief being
reported. And now (p. 307):
Since a hallmark of states of human consciousness is that
they can be reported (barring aphasia, paralysis, or
being bound and gagged, for instance), it follows, on
Rosenthal's analysis, that ``conscious states must be
accompanied by suitable higher-order thoughts, and
nonconscious mental states cannot be thus accompanied''
(1990, p. 16). The higher-order thought in question
must of course be about the state it accompanies; it must
be the thought that one is in the lower-order state (or
has just been in it---time marches on). This looks as if
it is about to generate an infinite regress of
higher-order conscious states or thoughts, but Rosenthal
argues that folk psychology permits a striking inversion:
The second-order thought does not itself have to be
conscious in order for its first-order object to be
conscious. You can express a thought without being
conscious of it, so you can express a second-order
thought without being conscious of it---all you need be
conscious of is its object, the first-order thought you
report.
3.3 The difficulty Dennett has with this analysis is that an
explosion of higher-order beliefs may occur nevertheless.
On the way to expressing the second-order thought there
might creep in some error, so that what is reported is not
the first-order belief. That is a relatively mundane kind
of error. But why can't there also be an error that
intervenes between the experience and the belief about it;
can't one's belief about one's current conscious state be
mistaken? It seems that we can cut through such
complicating possibilities by asserting that reflexive
second-order belief implies first-order belief: I believe
that I believe P logically implies I believe P.
But this merger will not quite do the work that needs to
be done.... Even if it is intuitively plausible that you
cannot be mistaken about how it is with you right
now, it is not at all intuitively plausible that you
cannot be mistaken about how it was with you back
then.... The logical possibility of misremembering is
opened up no matter how short the time interval between
actual experience and subsequent recall---this is what
gave Orwellian theories their license. But as we saw in
chapter 5, the error that creeps into subsequent belief
thanks to Orwelllian memory-tampering is
indistinguishable---both from the outside and the
inside---from the error that creeps into original
experience thanks to Stalinesque illusion-construction.
(pp. 318-9)
3.4 Of course, we saw no such thing. Nor does the
simplifying principle of my believing that I believe that P
implying I believe that P require adoption of the
demonstrably false: I believe that I believed that P
implies that I believe that P. Time is truth. And it is
not a difficulty of `folk' psychology that we can
meaningfully talk of higher- and still higher-order beliefs
that do not immediately collapse into first-order belief:
that is rather an aspect of the expressiveness of ordinary
psychology. Children will soon tire of extending the
sentence `I know that you know that I know that you know
that ... you want the last cookie', but that's hardly
because anyone believes each extension has the same meaning
as the sentence it embeds. Rosenthal's analysis has content
and merit that is neither exhausted nor explained by
Dennett's Multiple Drafts model.
4. Zombies and Robots
4.1 For many philosophers the crux of the matter of
consciousness comes down to what to make of qualia---the
phenomenal qualities of the things of which we are
conscious, the raw feels and sensations that make up much of
our conscious lives. One way to get at the issue is to ask,
could there be a (philosophical) zombie? That is, could
there be something which is physically indistinguishable
from ordinary humans but which fails to be conscious? The
idea that there
could be a zombie appears to arise from a
few plausible considerations. If the functional role of
pain, for example, is to inform us of injury and to motivate
us to avoid such injury in the future, there seems to be no
reason to disbelieve that such a role could be occupied by
some non-conscious processes. If we had an incredibly
complex program for a robot which provided for every
human-like cognitive function except pain, then surely we
could go on to provide damage-detection sensors and a
complex of goals and goal-oriented programming to avoid
activating such sensors. The same must be true of all other
kinds of qualia. But if we can in principle program such
functions, then we can in principle reconstruct that program
in neural matter. In that case we have a zombie: for by
stipulation we have implemented the function of every
conscious phenomenon without the consciousness.
4.2 Dennett does not go along with this; nor can any
functionalist. One method he uses to break down this kind
of reasoning is based upon the Multiple Drafts model of
consciousness. If consciousness is constituted of some
ever-changing flux of cognitive processes, some of them
operating on the contents of others, then we have only a
very fuzzy border between the presence and absence of
consciousness. If we think of any one function or small
clump of functions in isolation---say, pain functions---then
it is easy to imagine implementing those functions in a
system without consciousness. However, when we pile one
function on top of another on top of another into some
extremely complex whole, our intuitions about such matters
dissolve. Consider that no single neuron is conscious; nor
will it become conscious when connected to one other
neuron. Nevertheless, if we repeat the process billions of
times we must end up with consciousness: we know in
advance that something like human brains have consciousness.
Surely this argument is right: if consciousness is, loosely,
a property of the brain as a whole, then it is just
ill-conceived to look for it also within the constituent
parts of the whole---whether the brain is looked at as a
complex of neurons or as a complex of functions.
4.3 Since robots, on the initial argument above, have the
same functions as their neural doppelgaengers, they must
also have the same functional properties, such as
consciousness. So Dennett in fact applies just the same
argument to robots, claiming that adding a piece of
computational function here to one there may in fact get you
to robot consciousness in the end (pp. 438f). The argument
here is a bit weaker than that for humans. Since we know in
advance that brains are conscious, we can be confident that
adding neural functional complexes together can in principle
give us something conscious. We know nothing in advance
about robot consciousness, however. And certainly without
knowing in advance that the goal is achievable the argument
goes nowhere: I can always jump just a tiny bit higher than
my last jump; this does not mean that after some
unimaginably large number of jumps I can reach the moon.
Nor is doubt here necessarily just an expression of
prejudice against robots: although there are good reasons to
believe that computers can capture the full range of
computational abilities of the human brain, this does not
imply that such computer systems will also (necessarily)
implement the full range of brain functions---what has not
been shown is that all important brain functions are
strictly reducible to computations.
4.4 Qualia, on Dennett's view, are neither more nor less
than dispositional properties of cognition. The idea that
there is something mysteriously ineffable about
consciousness is particularly his object of attack. If
there is some content of consciousness that is entirely
immune to functional analysis, then it must be a content
which has no causal effects, which is causally ineffective
(disregarding non-functional causal influences). This must
be a curious content indeed, since anything which is
entirely causally ineffective must also be entirely
unremarkable in the literal sense; it will also be literally
unmemorable. Indeed it is a strange thought that
philosophers can coherently talk about such content at all.
The use of the concept of qualia to argue for such ineffable
objects appears to be a generally confused one: such
`qualophiles' want to argue both that conscious content is
effective (we drink the water because we are conscious of
thirst) and that the same conscious content is ineffective
(the `precise' feel of thirst cannot be expressed or
analyzed into any functional relations). It is in this
refinedly mysterious sense of consciousness that Dennett is
compelled to announce that we are none of us conscious. I
automatically assent.
5. Conclusion
5.1 I hope to have given some impression of the range of
topics, without pretending to have surveyed them. As I have
made clear, there is much in this book that is disputable.
And Dennett is at times aggravatingly smug and confident
about the merits of his arguments (comparing his
`revelations' about consciousness to a magician's revealing
the operation of stage tricks, for example; p. 434). All in
all Dennett's book is annoying, frustrating, insightful,
provocative and above all annoying. Unfortunately---in this
age of academic overproduction---I must conclude that for now
Consciousness Explained is unavoidable reading for those
who intend to think seriously about the problems of
consciousness.
Notes
<1> To be sure, the context of the quote is packed with
qualifiers and a footnote complains about my quoting him out
of context. But honestly how could I resist?}
<2> Operationalism claims that the content of our concepts is
exhausted by the operations we use to measure or judge their
application. This idea presupposes that we cannot find new
means to measure or judge the presence of previously defined
concepts, and is refuted, for example, by the history of the
concept of temperature.
<3> In the replies to commentaries section of Dennett and
Kinsbourne (1992) my point here is evidently acknowledged.
Presumably, Dennett's target is those who insist on asking
precisely when or exactly where consciousness begins and
ends---and trying to mean by that a degree of precision that
is unobtainable in principle. But while such requests are
misleading, and puzzles based upon them pointless, it is
also misleading in another way to insist flatly that there
is no fact of the matter as to where and when consciousness
occurs. (Incidently, readers of Consciousness Explained
will find Dennett and Kinsbourne (1992) of
interest---especially, perhaps, the numerous peer
commentaries on Dennett and Kinsbourne's description of the
Multiple Drafts model and their replies thereto.)
References
Churchland, P.M. (1981). Eliminative materialism and
propositional attitudes.
Journal of Philosophy, 78, 67-90.
Dennett, D. (1969). Content and consciousness. London:
Routledge & Kegan Paul.
Dennett, D. (1979). On the absence of phenomenology. In D.
Gustafson & B. Tapscott (Eds.) Body, mind and method:
Essays in honor of Virgil Aldrich. Dordrecht Reidel.
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.
Kolers, P.A., & von Gruenau, M. (1979). Shape and color in
apparent motion. Vision Research, 16, 329-335.
Korb, K. (1991). Searle's AI program. Journal of
Experimental and Theoretical AI, 3, 283-296.
Lewis, D. (1980). Mad pain and Martian pain. In N. Block
(Ed.) Readings in philosophy of psychology: Vol. 1 (pp.
216-222).
Rosenthal, D. (1986). Two concepts of consciousness.
Philosophical Studies, 49, 329-359.
Rosenthal, D. (1990). A theory of consciousness (ZIF
Report No. 40). Zentrum fuer Interdisziplinaere Forschung,
Bielefeld, Germany.
Selfridge, O. (1959). Pandemonium: a paradigm for learning.
In Symposium on the Mechanization of Thought Processes.
London: HM Stationery Office.
