Mandik, P. (in press). An epistemological theory of consciousness? In Alessio Plebe, ed. Philosophy in the Neuroscience Era, Special issue of the Journal of the Department of Cognitive Science, Univ. of Messina.
PETE
MANDIK, Department of Philosophy William Paterson University of New Jersey.
ABSTRACT:
This article tackles problems concerning the reduction of phenomenal
consciousness to brain processes that arise in consideration of specifically
epistemological properties that have been attributed to conscious experiences.
In particular, various defenders of dualism and epiphenomenalism have argued
for their positions by assuming special epistemic access to phenomenal
consciousness. Many physicalists have reacted to such arguments by denying the
epistemological premises. My aim in this paper is to take a different approach
in opposing dualism and argue that when we correctly examine both the
phenomenology and neural correlates of phenomenal consciousness we will see
that granting the epistemological premises of special access are the best hope
for a scientific study of consciousness. I argue that essential features of
consciousness involve both their knowability by the subject of experience as
well as their egocentricity, that is, their knowability by the subject as
belonging to the subject. I articulate a neuroscientifically informed theory of
phenomenal consciousness—the Allocentric-Egocentric Interface theory of
consciousness—whereby states of recurrent cortical networks satisfy criteria
for an epistemological theory of consciousness. The resultant theory shows both
how the epistemological assumptions made by dualists are sound but lead to a
reductive account of phenomenal consciousness.
"…if I were a reductionist, I would be this sort of
reductionist…"
David Chalmers 1996, p. 189
This article takes up the issue of the plausibility of epistemological theories of consciousness: solutions to the so-called “hard problem” of phenomenal consciousness (Chalmers 1996) that are rooted in physicalistic explanations of what we know and how we know it. Such accounts elaborate how physical systems come to (perceptually) know their physical environments and show how perceivers may come to find themselves positing and puzzling over the phenomenal aspects of experience: the qualia that have given physicalists so many headaches over the years. The main point of an epistemological theory of consciousness is to solve (or dissolve) apparent metaphysical issues concerning qualia in favor of epistemological explanations of why minds in the business of knowing about the physical world would ever come to think that there were qualia in the first place. In its most extreme form, for instance, as articulated by Daniel Dennett (1991), the point of an epistemological theory of consciousness is to show that once it is explained why we think that there are qualia, there is no further metaphysical work to be done in explaining qualia themselves. In other words, once the structure of our justified (and not so justified) beliefs concerning qualia is laid bare, no further work—only damage and confusion—is done by supposing that these beliefs are true. A (perhaps derisive) description of such a project is as forgoing any explanation of qualia in favor of explaining them away. In less extreme forms of epistemological theories, as in the case of recent work by Andy Clark (2000a, 2000b), it is not denied that there are metaphysical facts about qualia. Instead, the point of less extreme epistemological theories is that certain facts about our epistemologies entail certain metaphysical facts about qualia. Even Chalmers himself flirts with such an epistemological account (1996, pp. 287-292) but as Clark notes, Chalmers ultimately does not pursue it (2000a, p. 31). My aim in this paper is to pursue such a theory, and further, to flesh out how both neuroscience and philosophy converge to support such a view. I begin by focusing on Clark's work. I argue that Clark’s account fails, but that its failure instructively illuminates the path that any successful epistemological theory of conscious must follow. In later sections I flesh out what sort of beast one finds when one follows the path to its end.
Clark’s argument hinges on an oft-cited distinction in work
on consciousness: Block’s distinction between phenomenal consciousness and access consciousness (Block, 1995). Phenomenal consciousness is
what you have when there is something it is like to be you. You are subject to phenomenal consciousness when you
are subject to mental states that have qualia. Access consciousness is a more
boring form of consciousness. You are subject to access consciousness only
insofar as you are sensitive to information that you may subsequently use in
various ways, including verbal report. Access and phenomenal consciousness are
supposed to be distinct notions. The most notorious cases in which these
notions come apart are in hypothesized zombie cases—cases involving subjects devoid of phenomenal consciousness.
Zombies, in spite of being phenomenally vacant, may exhibit similar outward
behaviors and internal physiologies to us and thus are able to pick up and
report on all the information about their physical environments and bodies'
interiors that we are. However, one need not buy into the preposterous
imaginability of zombies to appreciate the possibility of access without
phenomenology. Human blindsight patients are offered as real world examples in
which access to visual information (e.g. that a red chair is present) is had
without the phenomenal raw feels that make it the case that there is something
it is like to see a red chair (Weiskrantz, 1996). Whether phenomenal and access
consciousness are entirely distinct is questionable and oft questioned (see,
for example, Dennett, 1995). Recently Clark has argued for a case in which a
certain kind of access consciousness entails a certain kind of phenomenal
consciousness. More specifically, according to Clark, if one has access to
information regarding which of several sensory modalities in virtue of which
one is responsive to the world, then there must be some phenomenal feel that
distinguishes one modality from the next—there must be something it is like to sense the world through one modality rather than the
other. Clark’s claim is especially interesting in the way he unpacks it. The
reason that this entailment is supposed to hold is supposed to provide an
entering wedge for a solution to the notorious “Hard Problem” of consciousness
(Clark 2000a, p. 32).
Before considering Clark’s argument that there is at least one case in which access consciousness entails phenomenal consciousness, it will be useful to consider a case in which phenomenal consciousness follows trivially from access consciousness. It will be useful to have such a case in mind to evaluate whether Clark’s argument is just a version of this trivial inference. Consider first the following minimal characterization of what qualia are: they are the introspectible similarities and differences of sensory experiences (Mandik, 1999, p. 47; Lycan, 1996, p. 69-70). What it is like to see a banana is more like what it is like to see a plantain than it is to smell a lime, and I know this by introspection. I am in cognitive contact, then, with my own qualia. This epistemological intimacy we share with our own qualia is an important plank in many qualiophilic platforms. It is the basis of the prevalent attitude that our qualia are obvious to us, as expressed by Block’s borrowing of Louis Armstrong’s remark on jazz “If you got to ask, you ain't never gonna get to know” (1978, p. 281). Thus, on many accounts of what qualia are, they are among the things that a subject has cognitive access to. Among the things I have access consciousness of are the introspectible similarities and differences between my experiences. I have access consciousness, then, of facts about my phenomenal consciousness. There is thus at least one (trivial) inference from access to phenomena: Access to phenomenal consciousness entails the existence of phenomenal consciousness. (Note that the entailment goes the other way too, since it is built into the notion of phenomenal consciousness that one have accesses consciousness concerning facts about one’s own qualitative states.)
Is Clark’s argument a version of this trivial one? Let us
begin by examining his argument. Clark invites us to imagine an organism or
perhaps even a robot capable of both perceptually discriminating items in its
environment and introspectively discriminating which of several sensory
modalities it employs to achieve its perceptual discriminations. Suppose, then,
that the system in question perceptually judges that bananas are more similar
to plantains than to pomegranates and that the sensory modality by which it
comes upon this information is its visual modality. Suppose further that the
system is able to answer questions about how it made the judgment concerning
its sensory modality. According to Clark, such a system. . .
. . .
must say either:
(a) I have no access to the act by means of which I detect the difference. The answer just comes to me. I perceive nothing when I make my judgments - I simply find myself saying that there are two objects, one red and one yellow, and so on.
Or:
(b) I have access not just to the products of my sensory activity, but also to certain aspects of the sensory activity itself. For example, I am non-inferentially aware that I am using a visual rather than a tactile modality. I am aware that I see, rather than hear or feel, the difference. (2000a, p. 30)
Clark
urges us to conclude that if the system gives the latter answer, then there
must be something it is like for the system to obtain the information in
question. Clark concludes that having access consciousness of which sensory
modalities it used to make the perceptual judgment entails having a mental
state with full-blown phenomenal consciousness. Note that the notion of
entailment operative here is supposed to be stronger than, say, nomological
entailment. For Clark, phenomenal consciousness is supposed to follow conceptually from the kind of access under consideration (2000a, p.
31).
I'll argue that Clark has not supplied the resources to
adequately distinguish his inference from the trivial entailment mentioned
above. Showing this will involve showing how Clark’s position occupies an
unstable territory between two opposing camps: qualia liberals and qualia
conservatives. The liberal sees qualia in lots of places that Clark does not.
In contrast, the conservative sees qualia in far fewer places than Clark does.
For an extreme example of a qualia liberal, consider
someone who holds that the mere fact that a system is able to make perceptual
judgments about features of its environment entails the existence of phenomenal
consciousness. On such a view, if a system is capable of perceptually
discerning bananas from pomegranates, then there must be something it is like
to perceive bananas such that what it is like is different from what it is like
to perceive pomegranates. For this sort of liberal, in order to have qualia,
the creatures need not access information about which modality they employed,
they need only employ the modality. (See, for example, Stone, 2001 and, in
particular, Lycan, 1996, pp. 76-77.) For the qualia liberal, the entailment
Clark describes follows trivially. If a system must have phenomenal
consciousness when it introspectively discerns that the perceptual basis of its
judgments about the fruits was visual as opposed to tactile, then this is
because phenomenal consciousness—what-it-is-like-ness—was already there in
the first place. It was already there to
be introspected, independently of the actual act of introspecting. It does not,
as Clark would have it, arise because
of the introspecting.
In contrast to the qualia liberal, the conservative holds
that adding all the modality-specific introspective access in the world will be
(logically, conceptually) insufficient to give rise to phenomenal
consciousness. For a qualia conservative like Chalmers (1996), the phenomenal
does not logically supervene on the physical and adding the ability to
discriminate one’s own sensory modalities is merely adding more physical stuff
that does not logically entail the addition of qualia. While the liberal says
that Clark’s inference from access to phenomena follows trivially, the
conservative says that the inference doesn’t follow at all. For the qualia
conservative, a system no more needs qualia to detect the difference between
grapefruits and bananas than a thermostat needs qualia to detect the difference
between one temperature and the other. The only things required are mechanisms
sensitive to certain causally efficacious properties, but these mechanisms need
not give rise to any qualia. All may be dark inside, and there is no more
anything it is like to be a thermostat than to be a rock. Similarly, taking a
system with mechanisms capable of detecting external states and adding to it
mechanisms capable of detecting internal states of the previous mechanisms does
nothing to change whether all is dark inside the system. If external states may
be detected without qualia, so may internal states.
Clark’s position constitutes an inadequately defended (by
Clark) middle ground between qualia liberalism and qualia conservatism. For
Clark, the mere presence of the first-order discriminatory states is
insufficient to bestow qualia, but adding certain kinds of higher-order states
does the trick. For the qualia liberal, the system with higher-order states has
qualia, but only because the qualia were already present in the first order
states. For the qualia conservative, there are no qualia necessitated by the
first order states nor are any necessitated by the introduction of any
higher-order states.
Despite
the flaws pointed out above, Clark's project sheds important light on the
future prospects of epistemological theories of consciousness. A key step in
arguing for an epistemological theory of consciousness will be the following:
whatever qualia are, they are such that we can and do have knowledge of them.
Recall the key fact upon which the trivial inference from access to phenomena
turns: that qualia are such that we know them. Call this the “epistemological
criterion”. Might we use something like the epistemological criterion to
construct a transcendental argument for physicalism?
The first steps of the argument will have as a premise the
epistemological criterion—the premise that makes such an argument transcendental.
The links to physicalistic conclusions will be given in portions of the
argument whereby the conditions on being known are linked to physical
conditions, through links relating knowledge to the causal aspects of
perception and introspection. Such an argument involves the following
desideratum for any theory of consciousness, namely, that it be impossible
that, for all you know, you are a zombie.
Many qualiophilic arguments, starting with strong realist
intuitions about qualia, end up with theses about qualia—in particular, qualia
epiphenomenalism (see, for example, Jackson, 1982, Robinson, 1982, and
Chalmers, 1996)—that leave the reader wondering how we can possibly ever know
qualia. If qualia cause nothing, then how can we have justified beliefs about
even our own qualia? The worry arises that qualia, which were supposed to
be essentially phenomenal, thus become quintessentially noumenal. This should
not come as a terrible surprise given the way it recapitulates the history of
so much of classical metaphysics and epistemology. Strong external-world
realism leads quickly to skeptical hand-wringing—how can we know a world that
is so far away? It may seem odd that such a problem can arise for the furniture
of the inner phenomenal world as well, but what is crucial is the way that a
strong, metaphysical, realism cuts the ties between justification and truth
(Putnam, 1981). The historical analogy between external-world realism and
qualia realism serves as well to point out the natural place a transcendental
turn might take on the way to bridge the explanatory gap between qualia and
physiology: the turn depends on the twin suppositions that qualia are knowable
and the knowledge must be physical through-and-through.
Clark argues that certain facts about access, facts I have
construed as epistemological, force subjects into “phenomenal space” and “mark
out a necessarily zombie-free zone”
(2000a, p. 37). I have argued, contra Clark, that the zombie-free zone has not
yet been adequately circumscribed, for it is not clear where and how Clark can
stake out neutral territory between the camps of the qualia liberals and qualia
conservatives. Though Clark’s arguments are flawed, they do point in the
promising direction of an epistemological theory of consciousness palatable to
those unable to stomach Dennett’s qualia nihilism. Clark’s suggestions—and, I
hope, mine—serve to light the way to a zombie free zone where qualia live, and,
importantly, qualia are known.
Before further developing the argument, it is worth noting
the following concerning reductionism and consciousness. Let us call a theory
of consciousness reductive if it entails that my physical doppelganger cannot
be a zombie. A reductive theory of consciousness must have outlines that are
discernible from both the first and third person points of view. The demand on first-person
discernability arises from its being a theory of consciousness. The demand on
third-person discernability arises from its being a reductive theory. A problem
that plagues reductive theories of consciousness is the question of what, if
anything, attaches the outlines discerned from the one point of view to the
other. After the reductive theory has been described the worry arises that the
separate portions may be implemented separately. This worry is oft expressed in
terms zombies: creatures that constitute implementations of the aspects of the
third person portions of a theory without simultaneously constituting
implementations of the first person portions of the theory. Epistemological reductive
theories attempt to bridge the gap from the first-person to the third-person by
discerning epistemic features accessible from the first-person point of view
that necessitate certain elements in the third-person point of view. What
third-person facts must obtain for there to be first person knowledge of
consciousness?
The crucial facts concern the knowability of one's own
conscious states. My conscious states are necessarily knowable as such:
knowable as my conscious states. In order for this to be true, conscious states
must be conceptualized and egocentric. Being conceptualized accounts for
knowability and being egocentric accounts for the knowability of a state as
my own. These features of consciousness
are discernable from the first-person point of view. Alleged counterexamples
depend on attributions from the third-person point of view. This is why
whenever one imagines the existence of a zombie, one imagines someone else
being a zombie. Imagining that one is currently mistaken about whether one is a
zombie is exceedingly difficult .
Now we are in a position to further spell out the
transcendental argument. Spelling this out further, the argument in outline is:
1. I know that I am not a zombie
2. My knowing that I am not a zombie is constituted by the satisfaction of condition K
3. My physical doppelganger satisfies condition K
.: My physical doppelganger is not a zombie
Most of
the heavy lifting will involve spelling out condition K such that 2 and 3 turn
out true. I've already given some indication of what condition K will consist
in. My knowing that I'm not a zombie entails that my conscious states are
necessarily knowable by me as such: knowable by me as my conscious states. In
order for this to be true, conscious states must be conceptualized and
egocentric. In the next two sections I spell out the conceptual and egocentric
components of condition K respectively.
Conscious
states involve conceptualizable contents. There are two related lines of
thought to consider here. The first concerns how the application of a concept
in experience influences what it is like to have that experience. So, for example, wine tastes different
to me now that I have and am able to apply the concept of tannic acid. The
night sky looks a lot different to me now that I have and can apply the concept
of the heavenly bodies being different distances from me. Squeaks in my car
engine sound a lot different to me now that I have and can apply the concept of
a broken valve lifter.
The second consideration takes a bit more time to
elaborate. The elaboration will involve arguing that since phenomenal
experience is necessarily knowable, phenomenal experience is exhaustively
constituted by conceptual content. Putting consciousness aside for a moment,
let us consider some features of knowledge and concepts. Suppose that there is
a rock that is heavy, lumpy and igneous. Suppose that George has the concepts
of lumpiness and heaviness, but no concept of being igneous. Suppose further
that at no point does George acquire the concept of being igneous. What, then,
can George know about the rock? He may know that it is lumpy and heavy, but
barring acquisition of the concept of being igneous bars George from knowing
that the rock is igneous. That the rock is igneous is, relative to George,
un-conceptualized residue. Since idealism about rocks is false, rocks are the
sorts of things that can have lots and lots of un-conceptualized residue. In
worlds with rocks but no knowers, rocks are 100% un-conceptualized residue.
Let’s turn from rocks to phenomenal experiences and ask
whether they can be, in whole or part, unconceptualized residue. One important
thing to note about phenomenal experience is its first-person necessary knowability.
This means that if a person has phenomenal experience then that person is
necessarily able to know that they have phenomenal experience. I take it that
not only am I not a zombie, but I know that I am not a zombie. I may not be
able to know whether or not you are a zombie, but that would simply be a
failure of third-person knowability. If phenomenal experiences are the sorts of
things that might even be beyond the knowability of the persons that have them,
then for all that person knows, they are a zombie, which I take to be absurd.
If a phenomenal experience has any phenomenal quality, Q, that is beyond the
knowability of the person having the experience, then for all that person
knows, they lack experiences with Q. They would be a Q-zombie for all they
know. Again, I take that to be absurd. Since non-zombies can know of themselves
that they are non-zombies, phenomenal experiences can have no phenomenal
qualities that are necessarily unknowable from the first-person point of view.
If something is necessarily knowable by me in that every aspect of it is
necessarily knowable by me, then it can have no aspect that outstrips my
concepts. If there were such an aspect it would inaccessible from the first
person point of view.
So far this seems to show only that there must be a
correlation between phenomenal characteristics and phenomenal concepts. Why
make the further step of identifying phenomenal character with the contents of
phenomenal concepts? Here’s why: If, with regards to the phenomenal, character is
distinct from conceptual content, then it would be possible for me to be in two
different phenomenal states even though I had the same doxastic state. That is,
I could believe that I had an experience with quality Q on both occasions but
in one case the belief would be true and the other case it would be false.
However, if this is possible, then there would be states that are phenomenally
distinct but subjectively indiscernible. I would be unable to know, from the
first person point of view, whether I was in a state with quality Q or not. I
could, for all I know, be a Q-zombie. I take this to be absurd. It follows,
then, that with respect to the phenomenal, character is not distinct from
conceptual content.
We need to consider whether the conditions for conceptual
content would be satisfied by my physical doppelganger. For the purposes of the
current argument I can be relatively open about what counts as a physical
doppelganger. Nothing in the current discussion necessarily hinges on whether
my physical doppelganger has only the same intrinsic physical properties as me
or has, in addition, a similar history of causal interactions with a similar
environment. Suppose we are maximally broad about what physical similarity
entails. We need only be careful that we do not assume at the outset that this
includes phenomenal similarity. One might question, however, why physical
similarity would entail conceptual similarity. A good response to this question
would be another question, namely "why not?" My physical doppelganger
is equally good at sorting objects and having conversations about them as I am.
My physical doppelganger can solve various puzzles you might pose for him, at
least, he will solve all the puzzles that I do. He will get himself out of many
sticky situations, or at least, as many as I will be able to. I'm aware of no
convincing account of concepts that would ascribe concepts only to me but not
to a system capable of all the same sorting, solving, and conversing that I am.
For these reasons my physical doppelganger is my conceptual doppelganger. It is
time now to consider the second component of condition K.
If my conscious experience is such that I can know it as my own, it must not only be knowable, but knowable as my own. This latter feature of consciousness is what I shall call the egocentricity or subjectivity of conscious experience. Being knowable as my own is only one side of the subjectivity of consciousness, the other side is that beings insufficiently similar to me have a difficult time knowing what my conscious experience is like. Discussions of this aspect of the subjectivity of consciousness have been closely associated with the famous knowledge argument of Jackson (1982) which is itself built on certain features of conscious experience remarked upon by Nagel (1974), who famously posed the question of what it is like to be a bat. The answer that Nagel urged is that bat experience must be insufficiently similar to our own for us to know what it is like to be a bat. Jackson's knowledge argument attempts to use the subjectivity of conscious experience as a premise in an argument that certain aspects of conscious experience must be non-physical. In brief, the knowledge argument begins by supposing it possible for someone to know all of the physical facts without ever having had a conscious experience of seeing red objects. It is further supposed that such a person would not, then, know what it is like to see red. Assuming further that such knowledge—knowing what it is like to see red—is factual knowledge, it follows then (from this assumption and the previous suppositions) that this must be knowledge of a non-physical fact. There have been various physicalist responses to the knowledge argument and space does not permit a review of them here. In Mandik (2001) I argue that the requisite notions of subjectivity can be explicated in terms of egocentric representation in certain kinds of neural networks. Below I develop a somewhat different approach, but the notion of egocentric representation will nonetheless play a central role.
Most of the discussion of consciousness in the literature
involves visual consciousness and this paper is no different in this regard. If
we are interested in thinking of visual consciousness in terms of the physical
properties that are responsible for it, it is useful to look at the portions of
the human nervous system responsible for vision in terms of a processing
hierarchy. This processing hierarchy has at its lowest levels the transduction
of information by cells in the eye and at its highest levels neural processing
in the cerebral cortex. There is much else going on between eye and cortex and
we can spell this out in further anatomical detail. The flow of information
begins at the rods and cones in the retina and the retinal ganglia. From there,
information is sent along the optic nerve to the bilateral sub-cortical
structures known as the lateral geniculate nuclei (LGN). Next, information is sent to the first stage
of cortical processing in the back of the brain in the occipital lobes (in area
V1). Next information is sent to further areas in the cortex along two
different routs (Milner and Goodale, 1995). One route is from the occipital
lobes to the parietal lobes. The second is from occipital lobes to temporal
lobes. From there information is sent as far as the frontal lobes (Olson,
Gettner, and Tremblay,1999) and the hippocampus
(Milner and Goodale, 1995). It should be further noted that not only does
information flow from the lowest levels to the highest, but there are also
"back projections" along which information flows from highest levels
back down to lowest (Pascual-Leone and Walsh, 2001).
One particularly important thing to note about the
different levels of the visual processing hierarchy is the different kinds of
representations that occur at the different levels. In brief, the key feature
is that the very lowest levels are highly specific and egocentric (such as
representations in LGN and V1) whereas higher levels are increasingly abstract
and conceptual (such as representations in hippocampus and frontal cortex). To
appreciate this difference with a relatively quick example, consider the
following. Suppose there was a neuron that responded to the presence of a
coffee mug, but only if that coffee mug was presented with a particular
orientation and location in the visual field. This neural activation would be a
representation relatively more egocentric and less abstract than the activation
of a neuron in response to the presence of a coffee mug regardless of its
orientation and position in the visual field. Among representations that we can
regard as egocentric, there are varying degrees of specificity or
egocentricity. A very egocentric representation of a stimulus would respond
only if the stimulus was present in a particular location defined relative to
the retina. This would be a representation of a location in retinocentric
space. Less specific, and thus less egocentric, would be a representation of a
stimulus that doesn't discriminate between stimuli that are presented to
different parts of the retina, but does respond when the stimulus is present in
a location defined relative to the head. While the coffee mug example is
perhaps fanciful, there is little controversy over the claims that (1) stimuli
are represented by cells with retinocentric receptive fields in retinal ganglia
and LGN (Hubel and Wiesel, 2001), (2) there are body-centered representations
that abstract away from retinal locations in areas such as posterior parietal
area 7a (Andersen, 1995), and (3)
there are representations of spatial locations in hippocampus that abstract
away from the orientation of the organism with respect to the larger
environment (Taube, Muller, and Ranck, 1990).
In Mandik (2001) I describe how all kinds of egocentric
representation involve the representation of properties defined relative to the
representing subject and how the different kinds of egocentric representation
(retinocentric, body-centered, etc.) may be distinguished in terms of how much
they leave out or abstract away from the details of the representing subject. I
also spell out how the resultant account of egocentricity need not apply only
to spatial representation but may be generalized to, for example,
representations of times and temperatures. This sketch of the processing
hierarchy gives us several kinds of egocentricity, but which kind is the
egocentricity of consciousness? To answer this, we must say more about
consciousness.
The
question naturally arises of how to relate consciousness to the visual
processing hierarchy and before we address that topic head on, it will be
useful to look at a particular kind of example of conscious visual states. One
particularly useful kind of example to look at concerns the phenomenon of
motion induced blindness (Bonneh,
Cooperman, and Sagi, 2001).
Motion induced blindness doesn't involve any kind of brain damage or malfunction
and is a relatively easy phenomenon to induce in normally sighted subjects. In
a typical experiment, a subject stares at a computer screen that has stationary
yellow dots on a black background. Behind the stationary yellow dots is a
collection of moving blue dots. After a short while, it appears as if one or
more yellow dots disappear. However, this is merely an appearance—in reality
the yellow dots remain on the screen for the entire interval. One hypothesis
that is relatively easily ruled out is that during these moments of blindness
the information about the yellow dots isn't getting into the nervous system at
all. To the contrary, however, the information seems to be making it up pretty
high in the visual processing hierarchy. It gets up as high as cortical areas
including parietal cortex (for further discussion, see Mandik 2005). The key here
is that motion-induced blindness isn't due to a failure of the nervous system
to represent the presence of yellow dots. It is instead due to a failure of the
nervous system to represent the presence of yellow dots consciously. Soon I will say some more about what the relevant kinds
of representation are in the processing hierarchy, but first I must note three
points about how motion-induced blindness relates to relevant notions of
consciousness. First, it is clear in these cases is that at one moment the
subject is conscious of the yellow dot and at another moment she is not.
Second, in conditions in which the subject is conscious of the yellow dot, she
has a representation of a yellow dot that is a conscious representation, and in
conditions in which the subject is not conscious of a yellow dot, the presence
of a yellow dot is still represented in the nervous system, albeit
unconsciously. Third, we must address the most interesting (to philosophers)
aspect of these conscious mental states, namely their qualia, phenomenal
character, or properties in virtue of which there is something it is like to
have them. While much more can be said about this, for now I will simply
suggest that what it is like is like seeing a yellow dot and this can be
accounted for by the representation of the presence of a yellow dot.
Given this quick sketch of the visual processing hierarchy
I want to turn now to where in the processing hierarchy to locate conscious
representations. We can arrive at the view that conscious representations are
to be found at an intermediate level by considering reasons for rejecting
locating conscious representations at either the highest or lowest extremes. To
see why egocentric representations at the lowest levels don't suffice for
consciousness, we might do well enough to consider that no one takes seriously
the suggestion that retinal ganglion activity suffices for conscious
experiences. But we can add that even egocentric representations slightly
higher up in the hierarchy are insufficient for conscious visual states. So,
for example, Milner and Goodale (1995) describe a patient, DF, who suffered
bilateral damage to lateral occipital cortex resulting in visual form
agnosia—an inability to consciously perceive the shapes of objects. DF can
nonetheless respond to visual information about form and orient her hand
appropriately to put a card in a slot despite reporting that she cannot see the
slot or its orientation. Further, she cannot report the slot's orientation. It
seem that in spite of this failure to consciously perceive form, DF is
utilizing egocentric representations to orient her hand appropriately. Just as
it is inappropriate to locate consciousness at the lowest end of the visual
processing hierarchy, so is it inappropriate to locate it at the highest end.
For examples of representations at the highest part of the hierarchy we can
consider instances of relatively abstract categorical knowledge, like your
knowledge that all mammals are warm-blooded. This is a piece of information
that you have known and thus represented for a long time, but it is highly
unlikely that you were doing so consciously the entire time. It is likely that
until reading the previous sentence this is the first time in a while that the
thought occurred to you consciously. What do these various examples establish?
At best, they show that states at the far end of the hierarchy can sometimes be
unconscious, not that they never can be. Nonetheless, when we examine conscious
states we notice something important, namely that they plausibly exist in the
middle of the hierarchy and thus combine aspects of higher- and lower-level
representations. Consider, as a typical example of a conscious visual state,
the visual perception of a coffee mug. You see the coffee mug from a particular
point of view with a particular orientation in your visual field and thus does
your conscious experience involve egocentric representations. However, you
further are able to bring to bear conceptual knowledge in your experience: you
see it as a coffee mug and as a concave item that holds beverages. Seeing it as such involves
representations from higher in the hierarchy.
These remarks about the conceptual and egocentric aspects
of conscious states are observations made "from the inside," as it
were. Viewed from the outside, conscious states consist in mutually interacting
representations positioned at slightly different levels in the middle of the
processing hierarchy. In other words, activations of representations in a
solely feed-forward manner do not suffice for consciousness, but when there is
feedback from the higher to the lower levels, then the representations involved
constitute conscious states. One line of evidence for such a view comes from
experiments involving trans-cranial magnetic stimulation conducted by
Pascual-Leone and Walsh (2001). By utilizing precisely timed magnetic
stimulation of certain cortical regions they were able to allow the flow of
information up from V1 to V5 and prevent the normal feedback of information
from V5 down to V1. V5 is an area associated with the conscious perception of
motion, but results suggest that motion was consciously perceived on occasions
of V5 activation only when feedback to V1 was allowed. Further evidence for the
need of reciprocal interaction between higher and lower levels comes from Lamme
et al. (1998), who suggest that the responses elicited by stimuli in
anesthetized animals constitute merely feed-forward activation of
representations in perceptual networks and lack feed-back activations from representations
higher in the processing hierarchy. The resultant view of the neural basis of
consciousness is what I have referred to elsewhere as the Allocentric
Egocentric-Interface theory of consciousness:
[C]onscious states are hybrid states that involve the
reciprocal interaction between relatively allocentric and relatively egocentric
states: a conscious state is composed of a pair of representations interacting
at the Allocentric-Egocentric Interface. Unconscious mental states are states
that are either too high up or too low down in the hierarchy or are not engaged
in the requisite reciprocal interactions. What a person is conscious of is
determined by what the contributing allocentric and egocentric representations
are representations of. The phenomenal character of these states is identical
to the representational content of the reciprocally interacting egocentric and
allocentric representations.
(pp. 463-464).
For the
present purposes we can regard the allocentric representations as conceptual
representations. The lower of the two states in the hybrid provides the
egocentricity of consciousness. The higher of the two provides the conceptual
content of consciousness. We thus have a sketch of theory that spells out how a
consciousness can reduce to physiological processes in a way that answers to
the demands of the transcendental argument.
It will
be useful to take stock and summarize what has gone on so far. First we
examined Clark's arguments that if a creature was able to non-inferentially
discriminate its own sensory modalities, it must do so in virtue of there being
something it is like for the creature to be in those sensory states. I
criticized Clark for insufficiently building a bridge from what a subject knows
of its own states to there being something it is like to be in those states. I
praised Clark for lighting the way toward a potentially successful project, one
in which one may argue transcendentally for a reductive theory of consciousness.
I sketched how such an argument might go, starting with considerations that
each of us knows that he or she is not a zombie and leading to the
considerations that the structure of conscious experience must be both
conceptual and egocentric. I then spelled out how the conceptual and egocentric
criteria can be satisfied by a neurophysiological theory of consciousness
whereby conscious states are hybrids of mutually interacting representations
activated at different levels of a sensory processing hierarchy. What remains
open is to solve the sort of problem that befell Clark's attempt at an
epistemological theory of consciousness, namely to spell out a clear connection
between the relevant epistemic notions and the notion of what it is like to be
in a conscious state. It is to this last step that I turn. Crucial in what
follows is the equating of what it is like with how things appear with how
things are represented by states with conceptual contents.
To see how the relevant notions of appearance figure into the above account of consciousness, it will be useful to spell out how the above view of consciousness relates to perception and introspection, for perception and introspection are faculties by which things appear to us. There are ways things seem when we perceive them and there are ways our mental states seem when we introspect them. The account of perception and introspection I favor is one developed by Churchland (1979) and one I've elaborated elsewhere (Mandik 2006). The view of perception at play here is that “perception consists in the conceptual exploitation of the natural information contained in our sensations or sensory states” (Churchland , 1979 , p. 7). The view of introspection is analogous: introspection of sensations is the conceptual exploitation of natural information that our sensations contain about themselves. The notion of the conceptual exploitation of information contained in sensations can be spelled out, following Churchland, in terms of two senses in which a sensation may have intentionality, that is, two senses in which a sensation may be a sensation of something. These two senses of "sensation of" are an objective sense and a subjective sense. Adapting Churchland’s formulations (1979, p. 14) yields:
A given (kind of) sensation one has is a sensation of X (in the objective sense of “of”) if and only if under normal conditions, sensations of that kind occur in one only if something in one’s perceptual environment is indeed X.
A given (kind of) sensation one has is a sensation of X (in the subjective sense of “of”) if and only if under normal conditions, one’s characteristic non-inferential response to any sensation of that kind is some judgment to the effect that something or other is X.
These
notions allow for the explication of many important features about perception,
in particular: (1) the distinction between what can be perceived and what actually is perceived and (2) the
distinction between what is perceived and what is inferred. What can be
perceived is determined by the objective intentionality of sensations. My
vision is quite poor without my contact lenses and my inability to see the
small print at the bottom of an eye chart is due to the lack of information
present at my irradiated retina. Wearing my contacts increases the information
my sensory states carry about distal objects and thus increases the number of
things I can perceive. This does not however, alone suffice for what I actually
do perceive. I may fail to perceive some distant object not because my eyesight
is insufficiently acute but because I simply have not noticed it. What I do
perceive depends on what concepts are brought to bear in my non-inferentially
elicited judgments about the causes of my sensations. Thus, if my contacts are
on, I'm looking in the right direction, and I have the concept of an insect,
then I am in a position to actually perceive some small insect flying through
my line of sight.
Regarding the distinction between what is perceived and
what is inferred, consider the following example adapted from Mandik (2006).
Jones and Smith both witness a man in a realistic gorilla suit perform a
realistic imitation of a gorilla. Both suit and performance are convincing to
the untrained eye and Smith, having an untrained eye, is initially convinced.
Jones, in contrast, is a special-effects expert and thus is not fooled. Jones can see that this is a man in a
costume. Suppose that at some later point Jones tells Smith that this is merely
a man in a suit. Smith trusts Jones and believes him. Nonetheless, Smith cannot
shake the impression that this is a real gorilla. What is going on with Jones and Smith? Suppose that they
both have equally acute eyesight and in a sense, then, see the same things
insofar as they have visual sensations with the same objective intentionality.
Jones and Smith differ, however, in that Jones is, in virtue of his expertise
in special effects, able to automatically apply the concept of a man in
response to his sensations of the man in the suit. Smith, in contrast, is able
to apply the concept of a man only as the consequence of an inference and
further is having difficulty overcoming his tendency to automatically apply the
concept of a gorilla.
With this sketch of perception in hand, we can go on to
sketch an account of introspection. Focusing on the introspection of
sensations, introspection is analogous to perception in that each involves the
automatic application of concepts in judgments elicited by sensations.
Introspection differs from perception in the concepts that are brought to bear
and the information thereby exploited. Perception involves concepts of external
world objects and properties for the exploitation of information that
sensations carry about those external world objects and properties. The
introspection of sensation involves concepts about sensations for the
exploitation of information that sensations carry about themselves.
Sensations, despite being brain states seldom seem like brain states, and this is due in large part to the
facts that (1) most people lack the requisite neuroscientific concepts and (2)
fewer still have sufficient training to automatically apply neuroscientific concepts in response to their own
sensations. But as argued in Churchland (1979) and Mandik (2006), the barriers
to introspecting ones own brain states as such are not insurmountable. But this
is beside the crucial point here. More important for present purposes is how
the above accounts for the ways things seem in perception and introspection,
for herein lies an explanation of knowledge of what its like to have conscious
experiences.
The account of perception and introspection supplies a
means for distinguishing both what is perceived and what is introspected from
what is inferred. Combining this account of perception and introspection with the
account of consciousness spelled out earlier provides a means for
distinguishing conscious from unconscious perceptions. With these various
distinctions in place, we are in a position to give an account of appearance
that can achieve what Clark could not: an explanation of why the ability to
know the difference between various states entails that there is something it
is like to be in those states (in a way that isn't simply an instance of the
"trivial inference" mentioned in §1).
The strategy in what follows will be two-fold. First, I
will discus the notion of appearance that needs explaining: a phenomenal notion
of appearance or ways things seem. Second, I will look at the minimal account
of appearance that comes along with the mere existence of conceptual states, an
epistemic notion of appearance. Then I will discuss how, building upon an
epistemic notion of appearance, we are able to recover the requisite notion of
phenomenal appearance. To appreciate the notion of appearance in need of
explaining, consider the following scenario.
The Blue Dog Scenario:
Smith and Jones see a dog that is in fact white but due to a trick of the electric lighting, seems blue. Smith is unaware of the facts about the lighting and so believes that the dog is blue. Jones knows about the lighting situation and so believes the dog is white. Jones would agree, though, that in spite of his believing it to be white, the dog seems blue.
What is going on in the minds of
Smith and Jones that constitute the ways things appear to them? We will return
to this question after the consideration of a different scenario.
The Monty Hall Scenario:
Smith and Jones are playing Let’s Make a Deal with Monty Hall. There are three doors for Smith and three for Jones. Behind one of Smith’s doors is a car. Likewise for Jones. They each pick their door number one. Before door number one is opened, Monty Hall opens door number three and reveals that there is a goat behind it. Monty asks if they’d like to keep door number one or switch to door number two. Smith figures there is a fifty/fifty chance that the car is behind door number one, so he believes door number two to not be a superior choice. Jones knows the explanation of the relevant probabilities and so believes correctly that there is an advantage in switching. Jones admits, though, that while he trusts the explanation, he doesn’t totally understand it, and sympathizes with Smith’s urge to not switch.
What is
going on in the minds of Smith and Jones in the Monty Hall Scenario that
constitutes the ways things seem to them relevant to their decisions in the
game? Here it looks like a purely epistemic notion of appearance can do the work. Things seem to Smith
and Jones to be such-and-such in so far as they apply various concepts in their
judgments that things are such-and-such. We may need, of course, to appeal to
various judgments and further, various dispositions of varying strength toward
distinct judgments, but none of this obviously takes us out of the realm of
epistemic appearance. So, for example, here is a straightforward and
uncontroversial explanation of what is going on. Smith has a disposition to
judge door number two to not be a superior choice and is aware of no overriding
considerations against resisting his disposition. Jones similarly has a
disposition to judge door number two to not be a superior choice, but is aware
of overriding considerations in favor of resisting this disposition, so he
resists. He believes door two to be superior but agrees that it seems not to be
superior. In what does this latter seeming consist? It consists in his
overridden disposition to make a certain judgment.
What, then, should we say of the Blue Dog Scenario? It is
worth noting, first, just how much mileage we can get out of an explanation
constructed to be analogous to the explanation of the Monty Hall Scenario.
Smith has a disposition to judge the dog to be blue and is aware of no
overriding considerations against resisting this disposition. Jones similarly
has a disposition to judge the dog to be blue, but is aware of overriding
considerations in favor of resisting this disposition, so he resists. He
believes the dog to be white but agrees that it seems to be blue. In what does
this latter seeming consist? It consists in his overridden disposition to make a
certain judgment.
One might object at this point that the epistemic
appearances appealed to in the explanation of the Blue Dog Scenario are mere epistemic appearances, that is, phenomenal appearances
have not yet been taken into account. While there is some truth to this, the
problem is not so much with the explanation of the scenario, but with the
scenario itself. Note that nothing going on in the scenario has essentially to
do with consciousness. However, we can modify the scenario slightly to add that
the relevant perceptions are conscious perceptions, that is, that Smith and
Jones are have conscious visual experiences of the blueness of the dog. Adding
consciousness to the scenario requires that we add appeal to a theory of
consciousness in an explanation of the relevant notions of appearance.
So-called phenomenal appearances are reducible to a
sub-class of epistemic appearances. There’s nothing going on in the mind in
these scenarios that can’t be explained in terms of information bearing states
(the sensations) and our conceptual reactions to them (the judgments). So, what
are qualia? They are introspectible properties of conscious states, where what
is introspectible is in part determined by what is there to be introspected and
in part determined by what concepts a subject is able to automatically bring to
bear in the judgments elicited by the sensory states. Conscious states are
hybrid states of mutually causally interacting judgments and sensations.
It might be objected at this point that the application of
concepts is unnecessary, that whatever phenomenal consciousness consists in, it
is the sort of thing that may exist independently of our conceptual states.
However, this objection garners no support from either first-person or
third-person views on consciousness, and we can develop this point along two
lines of thought. On the first line of thought, we look to sensory processing
hierarchies and note, as was noted before, that pre-conceptual states are
insufficient for consciousness. The states at the lowest levels of sensory
processing hierarchies are states that act as detectors of stimuli in
egocentric space, and as such are essentially no more complex than measuring
devices such as thermometers, devices that are themselves devoid of phenomenal
consciousness.
The second line of thought against this objection points out that alleged counter examples to the view that conscious states are conceptual-states would be inaccessible from the first-person point of view. Insofar as what is seen can be known, then visual experience has conceptual content. If visual experiences have distinctive contents, then there is no need to postulate a distinctive attitude to account for what is distinctive about experience. If visual experiences have conceptual contents then what is seen can be believed. If what is seen can be believed and there is such a thing as a counter-example to the claim that conscious states are conceptual states, then you would not be able to tell the difference between you and a being that had all of the same beliefs but differed in what visual experiences it had (or even in whether it had visual experiences). You would not know whether you were a counterexample. But this is absurd.
Let us return to the kind of task that Clark discussed in
his argument that a certain case of access consciousness entails phenomenal
consciousness. Suppose that Jones
knows, by introspection of his perceptual experience of the dog, that he is
arriving at the judgment that the dog is blue by seeing a blue dog (as opposed
to, say, being told that there is a blue dog in the vicinity). In such a case,
Jones is having a conscious visual experience of a dog as being blue insofar as
(1) Jones has a sensory state carrying information that there is a blue dog
(2), Jones judges that there is a blue dog, and (3) there is reciprocal causal
interaction between Jones' sensation and Jones' judgment. The appearance to
Jones that there is a blue dog is not a mere epistemic appearance because the judgment in question is
automatically elicited by the sensation. When Jones introspects, again the
resultant judgment is not a mere
epistemic appearance because his judgment that he is having a sensation of blue
is automatically elicited by the sensation. I close with one final question and
one final answer. Question: How do we know that Jones is not a zombie? Answer:
Because we know that we are not zombies and we know that we are just like
Jones. Thus has an epistemological theory of consciousness lead us to a
zombie-free zone where one and the same theory accounts for what consciousness
is and how consciousness is known.
This work was supported in part by grants from The National Endowment of the Humanities and The James S. McDonnell Foundation’s Project for Philosophy and the Neurosciences. I am grateful to audiences of versions of this material at the McDonnell Project "Neurophilosophy: The State of the Art" conference at the California Institute of Technology; the City University of New York Graduate Center Cognitive Science Symposium and Discussion Group; and the 2006 "Toward a Science of Consciousness" meeting in Tucson, Arizona. I am especially grateful for discussions of this and related material with the following individuals: Jared Blank, David Chalmers, Ron Chrisley, Andy Clark, Tanasije Gjorgoski, Uriah Kriegel, Clayton Littlejohn, Doug Meehan, Phillip Pettit, Jesse Prinz, David Rosenthal, Eric Schwitzgebel, Anders Weinstein, Josh Weisberg, Chase Wrenn, and Tad Zawidzki.
Andersen,
R. (1995). Coordinate transformations and motor planning in posterior parietal
cortex, in M. Gazzaniga (ed.), The Cognitive Neurosciences,
MIT Press, Cambridge, MA, 519-532.
Block, N. (1978). Troubles with functionalism. In Perception and Cognition: Issues in the Foundations of Psychology, ed. C. Savage, Minneapolis: University of Minnesota Press. 261-326
Block, N. (1995). On a confusion about a function of consciousness, Behavioral and Brain Sciences 18 (2), 227-288.
Bonneh,
Y., Cooperman A., and Sagi, D. (2001). Motion induced blindness in normal
observers, Nature, 411(6839), 798-801.
Chalmers, D. (1996). The Conscious Mind, Oxford University Press, New York.
Churchland,
P. (1979). Scientific Realism and the Plasticity of Mind, Cambridge University Press, Cambridge.
Clark, A. (2000a). A case where access implies qualia? Analysis 60: 1: 30-37.
Clark, A. (2000b). Phenomenal
immediacy and the doors of sensation. Journal of Consciousness Studies 7: 4: 21-24.
Dennett, D. (1991). Consciousness Explained. New York: Little Brown & Co.
Dennett, D. (1995). The path not taken. Behavioral and Brain Sciences 18: 2: 252-53.
Hubel, D. and Wiesel, T. (2001) Brain mechanisms of vision in Bechtel, W., Mandik, P., Mundale, J., and Stufflebeam, R. (eds.), Philosophy and the Neurosciences: A Reader, Basil Blackwell, Oxford, pp. 179-198.
Jackson, F. (1982). Epiphenomenal qualia. Philosophical Quarterly 32: 127-36.
Lamme, V. A. F., et al. (1998). Feedforward, horizontal, and feedback processing in the visual cortex. Current Opinion in Neurobiology, 8, 529 – 535.
Lycan, W. (1996). Consciousness and Experience. Cambridge, MA, MIT Press.
Mandik, P. (1999). Qualia, space, and control. Philosophical Psychology 12: 1: 47-60.
Mandik, P. (2001). Mental representation and the subjectivity of consciousness, Philosophical Psychology 14 (2), 179-202.
Mandik, P. (2005). Phenomenal consciousness and the allocentric- egocentric interface. In: R. Buccheri et al. (eds.); Endophysics, Time, Quantum and the Subjective World Scientific Publishing Co.
Mandik, P. (2006). The introspectability of brain states as such. In Brian Keeley, (ed.) Paul M. Churchland: Contemporary Philosophy in Focus. Cambridge: Cambridge University Press.
Milner, A. and Goodale, M. (1995) The Visual Brain in Action. Oxford University Press, New York.
Nagel, T. (1974). What is it like to be a bat? Philosophical Review 83, 435-50.
Olson, C., Gettner, S., and Tremblay, L. (1999). Representation of allocentric space in the monkey frontal lobe, in N. Burgess, K. Jeffery, and J. O’Keefe (eds.) The Hippocampal and Parietal Foundations of Spatial Cognition, Oxford University Press, New York, pp. 359-380.
Pascual-Leone, A. and Walsh, V. (2001). Fast backprojections from the motion to the primary visual area necessary for visual awareness, Science, 292, 510–512.
Robinson. W. (1982). Causation, sensations and knowledge. Mind 91: 364: 524-40.
Stone, J. (2001). What is it like to have an unconscious mental state? Philosophical Studies 104: 179-202.
Taube, J., Muller, R., and Ranck, J. (1990). Head direction cells recorded from the postsubiculum in freely moving rats. Description and qualitative analysis. Journal of Neurosciences, 10, 420-435.
Weiskrantz, L. (1996). Blindsight
revisited, Curr. Opin. Neurobiol, 6(2),
215-220.