The Harder Problem of Consciousness[1]
Ned Block
T.H.Huxley
famously said “How it is that anything so remarkable as a state of
consciousness comes about as a result of irritating nervous tissue, is just as
unaccountable as the appearance of Djin when Aladdin rubbed his lamp”.[2] We do not see how to explain a state of
consciousness in terms of its neurological basis. This is the Hard Problem of
Consciousness.[3]
The
aim of this paper is to present another problem of consciousness. The Harder Problem as I will call it is more epistemological
than the Hard Problem. A second
difference: the Hard Problem could arise for someone who has no conception of
another person, whereas the Harder
Problem is tied closely to the problem of other minds. Finally, the Harder Problem reveals an
epistemic tension or at least discomfort in our ordinary conception of
consciousness that is not suggested by the Hard Problem, and so in one respect it
is harder. Perhaps the Harder Problem
includes the Hard Problem and is best thought of as an epistemic add-on to
it. Or perhaps they are in some other
way facets of a single problem. Then my
point is that this single problem
I
believe that the major ontological disputes about the nature of consciousness
rest on an opposition between two perspectives:
In its most straightforward version, deflationism is
a thesis of a priori conceptual analysis, most prominently analysis of mental
terms in functional terms. As David
Lewis, a well known deflationist noted[4],
this view is the heir of logical behaviorism.
Phenomenal realism rejects these armchair philosophical reductive
analyses. But phenomenal realists have
no brief against scientific reduction of consciousness. Of course, there is no sharp line here, and
since the distinction is epistemic, one and the same metaphysical thesis could
be held both as a philosophical reductionist and as a scientific reductionist
thesis.[5]
I apologize for all the “isms” (deflationism, phenomenal realism and one more to come), but they are unavoidable since the point of this paper is that there is a tension between two of them. The tension is between phenomenal realism (“inflationism”) and (scientific) naturalism, the epistemological perspective according to which the default view is that consciousness has a scientific nature—where this is taken to include the idea that conscious similarities have scientific natures. (A view on a given subject is the default if it is the only one for which background considerations give rational ground for tentative belief.) This paper argues for a conditional in which specifications of phenomenal realism and scientific naturalism (and a few other relatively uncontroversial items—including, notably, a rejection of a skeptical perspective) appear on the left hand side. On the right hand side we have a specification of the epistemic tension which I mentioned. Deflationists who accept the argument may opt for modus tollens, giving them a reason to reject phenomenal realism. Phenomenal realist naturalists may want to weaken their commitment to naturalism or to phenomenal realism. To put the point without explicit “isms”: Many of us are committed to the idea that consciousness is both real and can be assumed to have a scientific nature, but it turns out that these commitments do not fit together comfortably.
Modern phenomenal realism has often been strongly naturalistic (e.g. Levine, Loar, McGinn, Peacocke, Perry, Shoemaker, Searle and myself). Dennett has often accused phenomenal realists of closet dualism. Rey has argued that the concept of consciousness is incoherent.[6] The upshot of this paper is that there is a grain of truth in these accusations.
Before I go on, I must make a
terminological comment. Imagine two
persons both of whom are in pain, but only one of whom is introspecting his
pain state and is in that sense conscious of it. One could say that only one of the two has a conscious pain. This is not the sense of ‘conscious’
used here. In the sense of ‘conscious’
used here, just in virtue of having
pain, both have conscious states.
To avoid verbal disputes, we could call the sense of ‘consciousness’
used here phenomenality. Pains
are intrinsically phenomenal and in that sense are intrinsically
conscious. In that sense—but not in
some other senses—there cannot be an unconscious pain.
The plan of the paper is this: first I will briefly
characterize the Hard Problem, mainly in order to distinguish it from the
Harder Problem. I will argue that the
Hard Problem can be dissolved only to reappear in a somewhat different form,
but that in this different form we can see a glimmer of hope for how a solution
might one day be found. I will then move
on to the Harder Problem, its significance and a comparison between the Hard
and Harder Problems. I will conclude
with some reflections on what options there are for the naturalistic phenomenal
realist.
The Hard Problem is one of
explaining why the neural basis of a phenomenal quality is the neural basis of that
phenomenal quality rather than another phenomenal quality or no phenomenal
quality at all. In other terms, there is
an explanatory gap between the neural basis of a phenomenal quality and the
phenomenal quality itself. Suppose (to
replace the neurologically ridiculous example of c-fibers that is often used by
philosophers with a view proposed as a theory of visual experience by Crick and
Koch[7])
that cortico-thalamic oscillation (of a certain sort) is the neural basis of an
experience with phenomenal quality Q.
Now there is a simple (over-simple) physicalist dissolution to the Hard
Problem that is based on mind-body identity: Phenomenal quality Q = cortico
thalamic oscillation (of a certain sort).
Here’s a statement of the solution:
The Hard Problem is illusory. One might as well ask why H2O is
the chemical basis of water rather than gasoline or nothing at all. Just as water is its chemical basis, so Q just is its neural basis
(cortico-thalamic oscillation), and that shows the original question is
wrongheaded
I
think there is something right about this answer but it is nonetheless
unsatisfactory. What is right about it
is that if Q= cortico-thalamic oscillation, that identity itself, like all
genuine identities, is inexplicable. [8]
What is wrong about it is that we are in a completely different
epistemic position with respect to such a mind-body identity claim than we are
with respect to ‘water =H2O’. The claim that Q is identical to
cortico-thalamic oscillation is just as puzzling—maybe more puzzling—than the
claim that the physical basis of Q is cortico-thalamic oscillation. We have no idea how it could be that one
property could be identical both to Q and cortico-thalamic oscillation. How could one property be both subjective and
objective? Although no one can explain an identity, we
can remove puzzlement by explaining how an identity can be true, most
obviously, how it is that the two concepts involved can pick out the same
thing. This is what we need in the case
of subjective/objective identities such as the putative identity that Q =
cortico-thalamic oscillation.
Joe Levine[9]
argues that there are two kinds of identities, those like ‘water=H2O’ which do
not admit of explanation and those like ‘the sensation of orange = cortico-thalamic
oscillation which are “gappy identities” which do allow explanation. He argues that the “left hand” mode of
presentation of the latter is more substantive than those of the former. The idea is supposed to be that descriptive
modes of presentation are “pointers we aim at our internal states with very
little substantive conception of what sort of thing we are pointing
at—demonstrative arrows shot blindly that refer to whatever they hit”. By contrast, according to Levine, phenomenal
modes of presentation really do give us a substantive idea of what they refer
to, not a “whatever they hit” idea.
However, even if we accept this distinction, it will not serve to
explain the “gappiness” of mind-body identities. Consider that the mode of presentation of a
sensation of a color can be the same as that of the color itself. Consider the identity ‘Orange = yellowish red’. Both modes of
presentation involved in this identity can be as substantive as those in the
putatively “gappy” identity just mentioned, yet this one is not “gappy” even if
some others are. To get an identity in
which only one side is substantive, and is so a better analogy to the mind-body
case, consider an assertion of ‘
The standard arguments against physicalism (most
recently by Jackson, Kripke and Chalmers) make it difficult to understand how
mind-body identity could be true, so explaining how it could be true requires
undermining those arguments. I will not
attempt such a large task here, especially since the role of the discussion of
the Hard Problem in this paper is mainly to contrast it with the Harder Problem
to come. So I will limit my efforts in
this direction to a brief discussion of
The key to what is wrong with
The expressions ‘this sudden involuntary muscle contraction’ and ‘this [experience] thing in my leg’ are two expressions that pick out the cramp I am now having in my leg. (These are versions of examples from Loar, op.cit.) In ‘this [experience] thing in my leg’, attention to an experience of the cramp functions so as to pick out the referent, the cramp. (That is the meaning of the bracket notation. The ‘this’ in ‘this [experience] thing in my leg’ refers to the thing in my leg, not the experience.) The first way of thinking about the cramp is an objective concept of the cramp. The second is a subjective concept of the same thing—subjective in that there is a phenomenal mode of access to the thing picked out. Just as we can have both objective and subjective concepts of a cramp, we can also have objective and subjective concepts of a cramp feeling. Assuming physicalism, we could have an objective neurological concept of a cramp feeling, e.g. ‘the phased locked 40 Hz oscillation that is occurring now’. And we could have a subjective concept of the same thing, ‘this [experience] feeling.’ Importantly, the same experience type could be part of—though function differently—in both subjective concepts, the subjective concept of the cramp and the subjective concept of the cramp feeling. Further, we could have both a subjective and objective concept of a single color. And we could have both a subjective and an objective concept of the experience of that color, and the same experience or mental image could function—albeit differently—in the two subjective concepts, one of the color, the other of the experience of the color.
Deflationists will not like this apparatus, but they should be
interested in the upshot since it may be of use to them in rejecting the
phenomenal realism in the antecedent of the conditional that this paper argues
for.
Concepts in the sense used here are mental
representations. For our purposes, we
may as well suppose a system of representation that includes both
quasi-linguistic elements as well as phenomenal elements such as experiences or
mental images. Stretching terminology,
we could call it a language of thought. [11]
In these terms, then, we can remove one type of
puzzlement that is connected with the Hard Problem as follows: there is no
problem about how a subjective property can be identical to an objective
property. Subjectivity and objectivity
are better seen as properties of concepts rather than properties of properties. The claim that an objective property is
identical to a subjective property would be more revealingly expressed as the
claim that an objective concept of a property picks out the same property as a
subjective concept of that property. So
we can substitute a dualism of concepts for a dualism of properties.
The same distinction helps us to solve the Mary
problem. In the room, Mary knew about
the subjective experience of red via the objective concept cortico-thalamic
oscillation. On leaving the room,
she acquires a subjective concept this [mental image] phenomenal property
of the same subjective experience. In
learning what it is like to see red, she does not learn a new fact. She knew about that fact in the room under an
objective concept and she learns a new concept of that very fact. One can acquire new knowledge about old facts
by acquiring new concepts of those facts.
New knowledge acquired in this way does not show that there are any
facts beyond the physical facts. Of
course it does require that there are concepts that are not physicalistic
concepts, but that is not a form of dualism.
(For purposes of this paper, we can think of physicalistic concepts as
concepts couched in the vocabulary of physics.
A physicalist can allow non-physicalistic vocabulary, e.g. the
vocabulary of economics. Of course,
physicalists say that everything is physical, including vocabulary. But the vocabulary of economics can be
physical in that sense without being physicalististic in the sense of couched
in the vocabulary of physics.)
Where are we?
The Hard Problem in one form was: how can an objective property be
identical to a subjective property? We
now have a dissolution of one aspect of the problem, appealing to the fact that
objectivity and subjectivity are best seen as properties of concepts. But that is no help in getting a sense of
what sorts of objective concepts and subjective concepts could pick out
the same property, and so it brings us no closer to actually getting such
concepts. As Nagel (op.cit.) noted, we
have no idea how there could be causal chains from an objective concept and a
subjective concept leading back to the same phenomenon in the world. We are in something like the position of
pre-Einsteinians who had no way of understanding how a concept of mass and a
concept of energy could pick out the same thing.
Naturalism:
Naturalism
is the view that it is a default that consciousness has a scientific nature (and
that similarities in consciousness have scientific natures). I will assume that the relevant sciences
include physics, chemistry, biology, computational theory, and parts of
psychology that don’t explicitly involve consciousness. (The point of the last condition is to avoid
the trivialization of naturalism that would result if we allowed the scientific
nature of consciousness to be…consciousness.)
I will lump these sciences together under the heading ‘physical’,
thinking of naturalism as the view that it is a default that consciousness is
physical (and that similarities in consciousness are physical). So naturalism = default physicalism, and is
thus a partly epistemic thesis. Naturalism in my sense recognizes that
although the indirect evidence for physicalism is impressive, there is little
direct evidence for it. My naturalist is
not a “die-hard” naturalist, but rather one who takes physicalism as a default,
a default that can be challenged. My
rationale for defining ‘naturalism’ in this way is that this version of the
doctrine is plausible, widely held, and leads to the epistemic tension that I
am expositing. Some other doctrines that
could be called ‘naturalism’ don’t, but this one does. I think that my
naturalism is close to what John Perry calls “antecedent physicalism”. (See his
Knowledge, Possibility and
Consciousness, MIT Press:
Functionalism: Functionalism and physicalism are usually considered
competing theories of mind. However, for the purposes of this paper, the
phenomenal realism/deflationism distinction is more important, and this
distinction cross-cuts the distinction between functionalism and
physicalism. In the terms used earlier,
one type of functionalism is deflationist, the other phenomenal realist. The latter is Psychofunctionalism, the
identification of phenomenality with a role property specified in terms of a
psychological or neuropsychological theory.[12] At the beginning of the paper, I pointed to
the somewhat vague distinction between philosophical and scientific
reduction. Deflationist functionalism is
a philosophical reductionist view whereas phenomenal realist
Psychofunctionalism is a scientific reductionist view.
I will be making use of the notion
of a superficial functional isomorph, a creature that is isomorphic to us with
respect to those causal relations among mental states , inputs and outputs that
are specified by common sense, or if you like, ”folk psychology”. Those who are skeptical about these notions
should note that the point of the paper is that a nexus of standard views leads
to a tension. This conceptual apparatus
may be part of what should be rejected.
Those who would like to see more on functionalism should consult any of
the standard reference works such as the Routledge Encyclopedia of
Philosophy. Or see http://www.nyu.edu/gsas/dept/philo/faculty/block/papers/functionalism.html
As I mentioned at the outset, this
paper argues for a conditional. On the
left side of the conditional are phenomenal realism and naturalism (plus
conceptual apparatus of the sort just mentioned). My current point is that I am including
Psychofunctionalism in the class of phenomenal realist naturalist theories.
Thus one kind of functionalism—the deflationist variety—is excluded by the
antecedent of my conditional, and another—the phenomenal realist variety—is in
the class of open options.
Anti-
Skeptical Perspective: In what follows, I will be adopting a point of view that sets
skepticism aside. “Undoubtedly,
humans are conscious and rocks and laptops are not.” (Further, bats
are undoubtedly conscious.) Of course, the anti-skeptical point of view I will
be adopting is the one appropriate to a naturalist phenomenal realist. Notably, from the naturalist phenomenal
realist perspective, the concept of a functional isomorph of us with no
consciousness is not incoherent and the claim of bare possibility of such a
zombie—so long as it is not alleged to be us--is not a form of skepticism.
Multiple
Realization/Multiple constitution: Putnam, Fodor and Block and Fodor argued that if
functionalism about the mind is true, physicalism is false.[13]
The line of argument assumes that functional organizations are often--maybe
even always--multiply realizable. The
state of adding 2 cannot be identical to an electronic state if a non-electronic
device (e.g. a brain) can add 2.
This “multiple realizability”
argument has become controversial lately[14], for reasons that I cannot go into
here.[15] The argument I will be giving is a version of
the traditional multiple realizability argument (albeit an epistemic version),
so I had better say a bit about what a realization is. One of the many notions of realization that
would do for our purposes is the following. A functional state is a kind of second order
property, a property which consists in having certain first order properties
that have certain causes and effects.[16]) For example, dormitivity in one sense of the
term is the property a pill has of having some (first order) property that
causes sleep. Provocativity is the
property of having some (first order) property or other that makes bulls
angry. We can speak of the first order property
of being a barbiturate as being one realizer of dormitivity, or of red as being
one realizer of provocativity.[17]
If we understand realization, we can
define constitution in terms of it. Suppose
that mental state M has a functional role that is realized by neural state
N. Then N constitutes M—relative to M
playing the M-role. The point of the
last condition is that ersatz M—a state functionally like M but missing
something essential to M as phenomenality is to pain—would also have the
M-role, but N would not constitute ersatz M merely in virtue of constituting
M. So the M-role can be multiply
realized even if mental state M is not multiply constituted.
There is an obvious obscurity in
what counts as multiple realization (or constitution). We can agree that neural property X is
distinct from neural property Y and that both realize a single functional
property without agreeing on whether X and Y are variants of a single property
or two substantially different properties, so we will not agree on whether
there is genuinely multiple realization.
And even if we agree that X and Y are substantially different, we may
still not agree on whether the functional property is multiply realized since
we may not agree on whether there is a single disjunctive realization. These issues will be discussed further in
Section VII.
My strategy will be to start with
the epistemic possibility of multiple realization and use it to argue for the
epistemic possibility of multiple constitution of mentality. I will then argue that the epistemic
possibility of multiple constitution of phenomenal properties is problematic. I will use a science fiction example of a
creature who is functionally the same as us but physically different. Those who hate science fiction should note
that the same issue arises—in more complicated forms—with respect to real
creatures, such as the octopus, which differ from us both physically and
functionally.
C (1): We have no reason to believe that
there is any deep physical property in common to all and only the possible
realizations of our superficial functional organization. Moreover—and this goes
beyond what is needed for (1)—but it does make (1) more vivid: we have no
reason to believe that we cannot find or make a merely superficial isomorph of ourselves. By “merely superficial isomorph”, I mean an
isomorph with respect to folk psychology and whatever is logically or
nomologically entailed by folk psychological isomorphism, but that’s all. For example, the fact that pains cause us to
moan (in circumstances that we have some appreciation of but no one has ever
precisely stated) is known to common sense, but the fact that just-noticeable
differences in stimuli increase with increasing intensity of the stimuli (the
Weber-Fechner Law) is not. So the merely superficial isomorph would be governed
by the former but not necessarily the latter. The TV series Star Trek: The
Next Generation (2/26/89) includes an episode (“The Measure of a Man”) in
which there is a trial in which it is decided whether a human-like android, Lt.
Commander Data, may legally be turned off and taken apart by someone who does
not know whether he can put the parts together again. (The technology which allowed the android to
be built has been lost.)[18]
Let us take Commander Data to be a merely superficial isomorph of us (ignoring
his superior reasoning and inferior emotions).
Then (1) can be taken to be that we have no reason to believe that
Commander Data is not nomologically or otherwise metaphysically possible. Note that I am not making so strong a claim
as made in Block and Fodor (op.cit.)—that there is empirical reason to suppose
that our functional organization is multiply realizable—but only that we have
no reason to doubt it.
The
strategy of the argument, you recall, is to move from the epistemic possibility
of multiple realization to the epistemic possibility of multiple
constitution. (1) is the epistemic
possibility of multiple realization.
C (2): Superficial functional equivalence
to us is a defeasible reason for attributing consciousness. That is, superficial functional equivalence
to us provides a reason for thinking a being is conscious, but that reason can
be disarmed or unmasked, its evidential value cancelled.
(2)
consists of two claims, that superficial functional equivalence to us is a
reason for attributing consciousness and that that reason is defeasible. The first claim is obvious enough. I am not claiming that the warrant is a
priori, just that there is warrant. I
doubt that there will be disagreement with such a minimal claim.
What is controversial about (2) is that the reason
is claimed to be defeasible. Certainly,
deflationary functionalists will deny the defeasibility. Of course, even
deflationary functionalists would allow that evidence for thinking something is functionally equivalent to us
can be defeated. For example, that
something emits English sounds is a reason to attribute consciousness, but if
we find the sound is recorded, the epistemic value of the evidence is
cancelled. However, (2) does not merely
say that functional or behavioral evidence for consciousness can be
defeated. (2) says that even if we know that something is functionally
equivalent to us, there are things we can find out that cancel the reason we
have to ascribe consciousness (without challenging our knowledge of the
functional equivalence). A creature’s consciousness can be unmasked without
unmasking its functional equivalence to us.
Here is a case in which the epistemic value of functional isomorphism is cancelled: The case involves a partial physical overlap between the functional isomorph and humans. Suppose that there are real neurophysiological differences of kind—not just complexity—between our conscious processes and our unconscious—that is, non-phenomenal—processes. Non-phenomenal neural process include, for example, those that regulate body temperature, blood pressure, heart rate and sugar in the blood—brain processes that can operate in people in irreversible vegetative coma. Suppose (but only temporarily—this assumption will be dispensed with later) that we find out that all of the merely superfical isomorph’s brain states are ones that--in us—are the neural bases only of phenomenally unconscious states. For example, the neural basis of the functional analog of pain in the merely superficial isomorph is the neural state that regulates the pituitary gland in us. This would not prove that the isomorph is not phenomenally conscious (for example, since the contexts of the neural realizers are different), but it would cancel or at least weaken the force of the reason for attributing consciousness provided by its functional isomorphism to us.
The role of this case is to motivate a further refining of our characterization of Commander Data and to justify (2) by exhibiting the epistemic role of a defeater.
Let us narrow down Commander Data’s physical specification to rule out the cases just mentioned as defeaters for attribution of consciousness to him. Here is a first shot:
v Commander Data is a superficial isomorph of us.
v
Commander Data is a merely superficial isomorph. So we have no reason to suppose there are any
shared non-heterogeneously-disjunctive physical properties between our
conscious states and Commander Data’s functional analogs of them that could be
the physical basis of any phenomenal overlap between them, since we have no
reason to think that such shared properties are required by the superficial
overlap.
v The physical realizers of Commander Data’s functional analogs of conscious states do not overlap with any of our brain mechanisms in any properties that we do not also share with inorganic entities that are uncontroversially mindless, e.g. toasters. So we can share properties with Commander Data like having molecules. But none of the realizers of Commander Data’s analogs of conscious states are the same as realizers of, for example, our states that regulate our blood sugar—since these are organic.
v Commander Data
does not have any part which itself is a functional isomorph of us and whose
activities are crucial to maintaining the functional organization of the whole. [19]
The point of the last two conditions is to specify
that Commander Data has a realization that cannot be seen to defeat the
attribution of consciousness to him either a priori or on the basis of a theory
of human consciousness. (For example, the last condition rules out a
“homunculi-headed” realization.) It
would help if I could think of all the realizations that have these kinds of
significance. If you tell me about one I
haven’t thought of, I’ll add a condition to rule it out.
Objection: we are entitled to reason
from same effects to same causes. Since
our phenomenal states play a role in causing our behavior, we can infer that
the functionally identical behavioral states of Commander Data are produced in
the same way, that is, phenomenally. To
refuse to accept this inference, the objection continues, is to suppose that
the presence or absence of phenomenality makes no causal difference.
Reply: Consider two computationally identical computers, one that works via electronic mechanisms, the other that works via hydraulic mechanisms. (Suppose that the fluid in one does the same job that the electricity does in the other.) We are not entitled to infer from the causal efficacy of the fluid in the hydraulic machine that the electrical machine also has fluid. One could not conclude that the presence or absence of the fluid makes no difference, just because there is a functional equivalent that has no fluid. One need not be an epiphenomenalist to take seriously the hypothesis that there are alternative realizations of the functional roles of our phenomenal states that are phenomenally blank.
We might suppose just to get an example on the table
that the physical basis of Commander Data’s brain is to be found in etched
silicon chips rather than the organic carbon basis of our brains. [20]
The reader could be forgiven for wondering at this
point whether I have not assembled stipulations that close off the question of
Commander Data’s consciousness. Naturalism
includes the doctrine that it is the default that a conscious overlap requires
a physical basis, and it may seem that I
have in effect stipulated that Commander Data does not have any physical commonality with us that could be the basis
of any shared phenomenality. The objection
ignores the option of a shared disjunctive
basis and certain other shared bases to be discussed below.
C (3): Fundamentally different physical
realization from us per se is not a ground of rational belief in lack
of consciousness . So the fact
that Commander Data’s control mechanisms are fundamentally different is not a
ground of rational belief that he has no phenomenal states. Note that I don’t say that finding out that
Commander Data has a silicon-based brain isn’t a reason for regarding
him as lacking consciousness. Rather I
say that the reason falls below the epistemic level of a ground for rational
belief. .
C (4). We have no conception of a ground
of rational belief to the effect that a realization of our superficial
functional organization that is physically fundamentally different along the lines I have specified for Commander
Data is or is not conscious. To
use a term suggested by Martine Nida-Rümelin in commenting on this paper,
Commander Data’s consciousness is meta-inaccessible. Not only do we lack a ground of belief, but
we lack a conception of any ground of belief.
This meta-inaccessibility is a
premise rather than a lemma or a conclusion because the line of thought I’ve
been presenting leads up to it without anything that I am happy to think of as
an argument for it. My hope is that this
way of leading up to it will allow the reader to see it as obvious.
We can see the rationale for
meta-inaccessibility by considering John Searle’s Chinese Room argument. Searle famously argued that even if we are
computational creatures, we are not either sentient or sapient merely in virtue
of that computational organization. In
reply to his critics[21],
he says repeatedly that a machine that shares our computational organization
and is therefore behaviorally and functionally equivalent to us—and therefore
passes the Turing Test--need not be an intentional system (or a conscious
being). What would make it an intentional system—and for Searle, intentionality
is engendered by and requires consciousness--is not the functional organization
but rather the way that functional organization is implemented in the biology
of the organism. But, to take an example that Searle uses, how would we know
whether something made out of beer cans is sentient or sapient? He says: “It is an empirical question whether any given machine [that
shares our superficial functional organization] has causal powers equivalent to
the brain.” (p. 452) “I think it is
evident that all sorts of substances in the world, like water pipes and toilet
paper, are going to lack those powers, but that is an empirical claim on
my part. On my account it is a testable
empirical claim whether in repairing a damaged brain,” we could duplicate
these causal powers. (p.453) “I offer no
a priori proof that a system of integrated circuit chips could not have
intentionality. That is, as I say
repeatedly, an empirical question. What I do argue is that in order to produce
intentionality the system would have to duplicate the causal powers of the
brain and that simply instantiating a formal program would not be sufficient
for that.” (p. 453; emphasis and bracketed clause added).
I
do not deny that one day the question of whether a creature like Commander
Data is phenomenally conscious may become
a testable empirical question. But it is
obvious that we do not now have any conception of how it could be
tested. Searle has suggested (in
conversation) that the question is an empirical one in that if I were the
device, I would know from the first person point of view if I was conscious. But even if we accept such a counterfactual,
we cannot take it as showing that the claim is testable or empirical in any
ordinary sense of the term.
Though I am
tweaking Searle’s flamboyant way of putting the point, my naturalist phenomenal
realist view is not that different from his.
I agree that whether physically different realizations of human
functional organization are conscious is not an a priori matter and could be
said to depend on whether their brains have “equivalent causal powers” to
ours—in the sense of having the power to be the physical basis of conscious
states. (However, I don’t agree with
Searle’s view that the neural bases of conscious states “cause” the conscious
states in any normal sense of ‘cause’.)
I agree with him that consciousness is a matter of the biology of the
organism, not (just) its information processing. The issue that I am raising here for
naturalist phenomenal realism threatens my view as much as his.
I am not denying that we might some
day come to have the conception we now do not have. (So I am not claiming—as
McGinn does—that this knowledge can be known now to be beyond our ken.)[22] I am merely saying that at this point, we
have no idea of evidence that would ground rational belief, even a hypothetical
or speculative conception. Of course
those who meet Commander Data will reasonably be sure that he is
conscious. But finding out that he is
not human cancels that ground of rational belief.
Perhaps we will discover the nature
of human consciousness and find that it applies to other creatures. E.g. the nature of human consciousness may
involve certain kinds of oscillatory processes that can apply to silicon
creatures as well. But the problem I am
raising will arise in connection with realizations of our functional
organization that lack those oscillatory processes. The root of the epistemic problem is that the
example of a conscious creature on which the science of consciousness is
inevitably based is us (where “us” can be construed to include non-human
creatures which are neurologically similar to humans). But how can science based on us generalize to
creatures that don’t share our physical properties? It would seem that a form of physicalism that
could embrace other creatures would have to be based on them at least in part
in the first place, but that cannot be done unless we already know whether they
are conscious.
I have left a number of aspects of
the story unspecified. What was the aim
of Commander Data’s designer? What is to be included in the “common sense”
facts about the mind that determine the grain of the functional isomorphism?
I keep using the phrase “ground of
rational belief”. What does it mean? I
take this to be an epistemic level that is stronger than “reason for believing”
and weaker than “rational certainty”. I
take it that a ground of rational belief that p allows knowledge that p but
mere reason for believing p does not.
I now move to the conditional that I advertised
earlier. Let us start by supposing, but only temporarily,
that physicalism requires a deep (non-superficial) unitary (non-heterogeneously-disjunctive)
scientific (physical) property shared by all and only conscious beings. This version of physicalism seems at first
glance to be incompatible with Commander Data’s being conscious, and the
corresponding version of naturalism (which says that physicalism is the
default) seems at first glance to be epistemically incompatible with phenomenal
realism. That is, naturalism says the
default is that Commander Data is not conscious but phenomenal realism says
that the issue is open in the sense of no rational ground for belief either
way. This is a first pass at saying what
the Harder Problem is.
If this strong kind of physicalism
really is incompatible with Commander Data’s being conscious, we might wonder whether
the reasons we have for believing physicalism will support this weight. I will pursue a weaker version of physicalism
(and corresponding version of naturalism) that does not rule out consciousness having
a physical basis that is disjunctive according to the standards of
physics. However, as we will see, the
stronger version of physicalism is not
actually incompatible with Commander Data’s being conscious, and the difference
between the stronger and weaker versions makes no important difference with respect
to our epistemic situation concerning Commander Data’s consciousness.
Disjunctivism is a form of
physicalism that allows that consciousness is a physical state that is
disjunctive by the standards of physics.
As applied to the current issue, Disjunctivism allows that if Commander
Data is conscious, the shared phenomenality is constituted by the property of
having Commander Data’s electronic realization of our shared functional state
or our electro-chemical realization.
In footnote 12, I mentioned Kim’s critique of the multiple
realizability argument against physicalism.
He argues that if mental property M is nomically equivalent to a
heterogeneous disjunction N, we should regard M as non-nomic and non- “real” because
N is. He argues that if human thought
can be realized by very different physical mechanisms from, say, Martian or
robot thought, then the real sciences of thought will be the sciences of the
separate realizations of it. To call
them all ‘thought’ is simply to apply a superficial verbal concept to all of
them, but the laws of human thought will be different from the laws of Martian
thought. The real kinds are not at the
level of the application of verbal concepts.[23]
Even those who are
sympathetic to this picture of thought must make an exception for consciousness (in
the sense, as always in this paper, of phenomenality). We can be happy with the view that there is a
science of human thought and another science of machine thought, but no science
of thought per se. But we should not be
happy with the idea that there is a science of human phenomenality, another of
machine phenomenality, etc. For since the overlap of these phenomenalities, phenomenality, is something real and not
merely nominal as in the case of thought, it must have a scientific basis. If a
phenomenal property is nomically coextensive with a heterogeneous neural
disjunction, it would not be at all obvious that we should conclude that the
phenomenal property is non-nomic and non- “real” because the disjunction
is. The phenomenal realist naturalist
point of view would be more friendly to the opposite, that the disjunction is
nomic and “real” because the phenomenal property is.
The real problem with
Disjunctivism is that whether it is true or not, we could have no good reason
to believe it. To see this, we shall
have to have a brief incursion into the epistemology of reductive theoretical
identity.
The Epistemology of Theoretical Identity
Why do we think that water = H2O,
temperature = mean molecular kinetic energy and freezing = lattice formation?[24] The answer begins with the fact that water,
temperature, freezing and other magnitudes form a family of causally
inter-related “macro” properties. This
family corresponds to a family of “micro” properties: H2O, mean
molecular kinetic energy, formation of a lattice of H2O
molecules. And the causal relations
among the macro properties can be explained if we suppose the following
relations between the families: that water = H2O, temperature = mean
molecular kinetic energy and freezing = lattice formation. For example, as water is cooled, it contracts
until about 4 degrees (F) above freezing, at which point it expands. Why?
Why does ice float on water? Here
is a sketch of the explanations: The oxygen atom in the H2O molecule
has two pairs of unmated electrons, which attract the hydrogen atoms on other H2O
molecules. Temperature = mean molecular
kinetic energy. When the temperature
(viz., kinetic energy) is high, the kinetic energy of the molecules is high
enough to break these hydrogen bonds, but as the kinetic energy of the
molecules decreases, each oxygen atom tends to attract two hydrogen atoms on
the ends of two other H2O molecules. When this process is complete,
the result is a lattice in which each oxygen atom is attached to four hydrogen
atoms. Ice is this lattice and freezing
is the formation of such a lattice.
Because of the geometry of the bonds, the lattice has an open, less
dense structure than amorphously structured H2O (viz., liquid
water)--which is why ice (solid water) floats on liquid water. The lattice forms slowly, beginning about 4
degrees above freezing. (The exact
temperature can be calculated on the basis of the numerical values of the
kinetic energies needed to break or prevent the bonds.) The formation of large
open lattice elements is what accounts for the expansion of water on the way to
freezing. (Water contracts in the earlier cooling because decreasing kinetic
energy allows more bonding, and until the bonding reaches a stage in which
there are full lattice elements, the effect of the increased bonding is make
the water more densely packed.)
Suppose we reject the assumption that temperature is
identical to mean molecular kinetic energy
in favor of the assumption that temperature is merely correlated with
mean molecular kinetic energy? And
suppose we reject the claim that freezing is lattice-formation in favor of a
correlation thesis. And likewise for
water/ H2O. Then we would
have an explanation for how something that is correlated with decreasing
temperature causes something that is correlated with frozen water to
float on something correlated with liquid water, which is not all that we
want. Further, if we assume identities,
we can explain why certain macro properties are spatio-temporally coincident
with certain micro-properties. The
reason to think that the identities are true is that assuming them gives us
explanations that we would not otherwise have and does not deprive us of
explanations that we already have or raise explanatory puzzles that would not
otherwise arise. The idea is not that
our reason for thinking these identities are true is that it would be nice if
they were true. Rather, it is that assuming that they are true yields the most
explanatory overall picture. In other words, the epistemology of
theoretical identity is just a special case of inference to the best
explanation.
Some suppose that substance
identities such as ‘water = H2O’ are on a different footing from
“property” identities, and that substance identities can be established on
purely spatio-temporal grounds. (Jaegwon Kim gave a paper at Columbia in
December, 1999 making this suggestion, and Tim Maudlin argued that all
theoretical identities are established on spatio-temporal grounds when I gave
this paper at Rutgers.) But deciding that water and H2O are spatio-temporally
coincident is part of the same package as having decided that they are one
and the same. For example, the air above
a glass of water buzzes with bits of water in constant exchange with water in
the atmosphere, a fact that we can acknowledge only if we are willing to
suppose that those H2O molecules are bits of water. The claim that water is H2O and
that water and H2O are spatio-temporally coincident stand or fall
together as parts of one explanatory package.
And once we conclude that the substance liquid water = amorphous H2O
and that the substance frozen water = lattice-structured H2O, we
would be hard pressed to deny that freezing = lattice formation, since the
difference between liquid and frozen water is that the former has an amorphous
structure and the latter a lattice structure.
Substance identities and property identities often form a single
explanatory package.
Back to Disjunctivism
With the epistemology of identity in
place, we can now ask whether there could be an argument from inference to the
best explanation to the conclusion that consciousness is a heterogeneous
physical disjunction, the disjunction of
our realization of the consciousness role and Commander Data’s corresponding
realization. Of course without a prior decision as to whether Commander Data’s
states are actually conscious, there could be no such argument,. Putting this point aside, let us suppose,
temporarily, that Commander Data is conscious.
Even so, the prospects for an argument from inference to the best
explanation to the identity of a phenomenal property with a disjunctive
physical property are dubious. We can
see this in two ways. First, let us attend
to our explanatory practice. We have an important though vague notion of
“fundamentally different” that governs our willingness to regard some
differences in realization as variants of the same basic type and others as
fundamentally different. When we regard
two realizations as fundamentally different, we prefer two non-disjunctive
identities to one disjunctive identity.
Here is an example: Molten glass hardens into an amorphous solid-like
substance. (If there are absolutely no
impurities, fast continuous cooling of water can make it harden without lattice
formation in a similar manner.) We could
give a disjunctive explanation of solid-like formation that included both
freezing and this kind of continuous hardening.
And if we preferred that disjunctive explanation to two distinct
explanations, we would regard the hardening of glass as a kind of freezing and
glass as a solid. But we do not take the
disjunctive explanation seriously and so we regard glass as (strictly speaking)
a super-cooled liquid rather than a solid. And we do not regard amorphous hardening as
freezing. We prefer two non-disjunctive identities,
freezing = lattice-formation and hardening = formation of an amorphous
super-cooled liquid to one disjunctive identity. Of course, the two processes (freezing and
hardening) are functionally different in all sorts of fine-grained ways. But the functional roles of Commander Data’s
functional analogs of our conscious states are also functionally different from
ours in all sorts of fine-grained ways.
Commander Data is functionally equivalent to us in those functional
roles known to common sense and anything else nomologically or logically required
by that equivalence, but everything else can be presumed to be different. Since we can stipulate that our physical
realizations of our conscious states are fundamentally different from Data’s,
whatever exactly fundamental difference turns out to be, the methodology that
applies to the hardening/freezing case can reasonably be applied to the case at
hand.
Of course, there are cases in which
we accept disjunctive identities, e.g. jade is nephrite or jadeite. But jade
is a merely nominal category, which makes disjunctive identities acceptable
even if not explanatory.
A second factor is that the
disjunctive identity, if accepted, would rule out questions that the phenomenal
realist naturalist does not want to rule out. The
question of why it is that water is correlated with H2O or why it is
that heat is correlated with molecular kinetic energy are bad questions, and
they are ruled out by the identity claims that water = H2O and heat
= molecular kinetic energy. Nor can the
identities themselves be questioned.
(See footnote 8.) If we were
to accept that consciousness is a disjunction of the physical basis of our
conscious states and Commander Data’s realization of the functionally
equivalent states, we would be committing ourselves to the idea that there is
no answer to the question of why we overlap phenomenally with Data in one
respect rather than in another respect or no respect at all. For the phenomenal realist, it is hard to imagine
a ground for rational belief that these questions have no answers. One can imagine finding no other account
remotely plausible, but why should the phenomenal realist accept a physicalist
view that dictates that these questions are illegitimate rather than opt for a
non-physicalist view that holds out some hope for an answer. (Remember that physicalism is only a default
view.) Even if we should come to believe
that dualism is unacceptable as well, our reason for accepting Disjunctive
physicalism would not seem to get up to the level of a ground for rational
belief.
Objection: You say
identities cannot be explained, but then you also say that we can have no
reason to accept a disjunctive physicalistic identity because it is not
explanatory.
Reply: Identities
cannot be explained, but they can contribute to explanations of other
things. My point about the epistemology
of identity is that it is only because of the explanatory power of identities
that we accept them and the disjunctive identity countenanced by Disjunctivism
does not pass muster.
Disjunctivism is one way of making
naturalism compatible with Commander Data being conscious, but there are
others. One is the view that
consciousness is as a matter of empirical fact identical to the superficial
functional organization that we share with Commander Data. We might call this view Superficialism (with
apologies to Georges Rey who has used this term for a somewhat different
doctrine). Recall that the phenomenal
realist/deflationist distinction is an epistemic one, so any ontological view
could in principle be held as having either epistemic status. Superficialism is the phenomenal realist claim that consciousness is identical to the
superficial functional organization that we share with Commander Data—as
distinct from the deflationist version of this claim mentioned earlier.
Note that Superficialism says consciousness is a
role property, not a property that fills or realizes that role. A role property is a kind of dispositional
property. Now there is no problem about
dispositions being caused: spraying my bicycle lock with liquid nitrogen causes
it to become fragile. So if pain is a
superficial functional state, we can perhaps make use of that identification to
explain the occurrence of pain in neural terms.
Whether dispositions are causes—as would be required by this
identity--is a more difficult issue that I shall bypass. (Does a disposition to
say ouch cause one to say ouch?)
The difficulty I want to raise is that even if
identifying pain with a superficial functional role does license explanations
of the superficial causes and effects of being in pain, the
identification cannot in the same way license explanations of the non-superficial
causes and effects of being in pain.
Suppose, for example, that psychologists discover that pain raises the
perceived pitch of sounds. Even if we
take the thesis that pain is a disposition to say ouch to help us to explain why
pain causes saying ouch, it will not explain the change in pitch. The epistemic difficulty I am pointing to is
that there is no good reason why the causal relations known to common sense ought to be explained differently from the
ones not known to common sense. So the
identification raises an explanatory puzzle that would not otherwise arise, and
that puts an epistemic roadblock in the way of the identification. This is perhaps not a conclusive difficulty
with the proposal, but it does put the burden of proof on the advocate of the
identification to come up with explanatory advantages so weighty as to rule out
the explanatory disadvantage just mentioned. [25]
Of course, this objection will not apply to the
phenomenal realist identification of consciousness with its total functional
role as opposed to its superficial functional role. Since the physiology of Commander Data’s
states differs from ours, their total functional roles will differ as
well. So this would be a chauvinist
proposal that would beg the question against Commander Data’s consciousness.
Martine
Nida-Rümelin objected that there are a vast number of properties, maybe
infinitely many, that are entailed nomologically or logically by the
superficial functional equivalence, and each of these is both shared with Data
and is a candidate for the nature of consciousness. Certainly a full treatment would attempt to
categorize these properties and assess their candidacy. Some—e.g., possessing complex inputs and
outputs—can be eliminated because they are also shared with mindless
computers. Of course, there may be
others that are not so easily dismissed.
The Upshot
I said earlier
that it seemed at first glance that a form of physicalism that required that
consciousness be constituted by a unitary physical property dictated that
Commander Data is not conscious. We can
now see that at second glance, this is not the case. Even if we preclude a disjunctive physical
basis to the phenomenal overlap between us and Commander Data (assuming that
there is such an overlap), still the physicalist could allow that Commander
Data is conscious on Superficialist
grounds. And even if we reject
Superficialism, there are other potential meta-inaccessible physical bases of a
phenomenal overlap between us and Commander Data.
The upshot is that
physicalism in neither the stronger
(unitary physical basis) nor weaker (physical basis that may or may not be
unitary) versions mentioned above rules out Commander Data’s being conscious. However, the only epistemically viable
naturalist or physicalist hypothesis—the only naturalist or physicalist
hypothesis we have a conception of a reason for accepting--is a deep unitary
physical or otherwise scientific property in common to all and only conscious
beings, a naturalistic basis that Commander Data does not share. So for the physicalist, Commander Data’s
consciousness is not epistemically viable.
Thus our knowledge
of physicalism is doubly problematic:
we have no conception of a ground of rational belief that Commander Data is or is not conscious,
and we have no way of moving from a conclusion that Commander Data is conscious
to any consequence for the truth of physicalism. And this holds despite the fact that
physicalism is our default view. Physicalism
is the default and also inaccessible and meta-inaccessible. The practical
significance—if we ever make a robot that is functionally equivalent to us—is
that the question of its consciousness and also of physicalism are inaccessible
and meta-inaccessible. But even if we
decide that the robot is conscious, we will have a choice between dualism and
an epistemically non-viable version of physicalism (Disjunctivism or
Superficialism). This is all part of the
Harder Problem. A second part follows.