From: Stevan Harnad (harnad@ecs.soton.ac.uk)
Date: Sat Nov 02 2002 - 21:49:13 GMT
On Mon, 21 Oct 2002 wrote:
Darek, sorry for the delay: to UK, Trieste and back...
> df> "Indirect causation" still must involve a chain of caused events that
> df> stretches between the two things that are in a "causal relationship."
>
>sh> Indeed it does, and so if I learn (booleanly) that "zebra" = "horse" +
>sh> "stripes" (with "horse" and "stripes" already grounded), then I am the
>sh> wiser (and "zebra" is grounded).
>
> You seem to be confirming in the above exchange that the grounding of
> "horse," for example, involves a "chain of caused events" between the
> symbol for "horse" and its referent, that is, a horse. That can only
> mean the physical icon is caused to function as a symbol by an actual
> horse or horses, as a necessary cause of this functional role but not
> the sole cause. If no horses existed then the icon would not begin to
> function as a symbol for "horse," although given the existence of horses
> many other systemic conditions must be present as well for the symbol to
> be created. This seems fairly simple in principle. It is puzzling then
> for you to say below that you cannot understand what it might mean for
> symbols to be caused by their referents.
But I really don't. I also don't know what meaning means, apart from
being able to name and describe an object (and feel something as that
goes on). Setting the feeling (and hence meaning) aside, all that's left
is grounding. A symbol is grounded directly if it occurs in a robot
when the robot encounters (and correctly classifies) a (kind of) object.
Note that the connection is not between an object and a symbol, but
between a KIND of object and a symbol -- a kind of object that the robot
can recognize on the basis of features in its input projection on the
robot's sensory surfaces (and through any further processing of that
"shadow"). The ability to make this connection is often gained through
trial and error learning, with corrective feedback, tuning the robot's
feature detectors.
Call that a causal connection if you like. It means whenever an object
of kind X is seen, the robot identifies it as "X". But the causal
connection is between all the different X's that are able to pass
through that learned filter. I find it more perspicuous to call it feature
detection capacity (for a certain kind of object, hence a certain set,
probably infinite, of potential sensory projections) rather than a causal
connection between an object and a name, but I don't mind.
Next, we have symbolic description: Z = Y + X. The robot has never seen
a Z, nor any object that is both a Y and an X, but it has seen X's and
Y's, and "X" and "Y" are grounded. You want to say the grounding is a
causal connection between an object and a symbol, and "X" token and an
X; and I that it is a causal mechanism able to detect a KIND of object
(X's) from their sensory shadows. Have it either way. But what, then is
the connection between "Z" and Z's? The same: The system can now
recognize a Z from its sensory shadow, if ever it encounters one. But
"Z" is grounded even before it encounters a Z, and even if it never
encounters a Z. In the same way, further discourse about Zs is grounded.
"Indirectly" grounded, because it was not based on direct sensorimotor
acquaintance and learning with Z's, but grounded nonetheless. (Further
discourse about Z's is also grounded.)
I do not find it useful to describe this is a causal connection between
"Z" and Z's either, but if you insist, it's a causal connection of much
the same kind as that between "X" and X's, only one "causal" step
removed.
>sh> I have no idea what symbols being caused by referents might mean! What I
>sh> mean is that there are systems that generate symbols (arbitrary squiggles)
>sh> to label object categories (etc.) that their sensorimotor systems can
>sh> detect, discriminate, identify, manipulate.
>
> An "object category" is not an object/event/state, therefore its
> existence in the system, let alone its manipulation, must be
> subjectively attributed to or "read into" what the system is doing
> by an observer. But I won't take time on that point right now.
What is "subjective attribution"? We've bracketed feelings, and what it
feels-like to mean something (because we have no idea what to make of
feelings) and we are left only with the functional notion of grounding.
A robot's category-names (symbols) are labels it is able to uniquely
assign to certain kinds of inputs. No subjectivity or even
"observation," just a chunk of I/O performance capacity (and its causal
mechanism: the robot's feature-learning powers, a sensorimotor learning
device).
>sh> The "causality" in question
>sh> is the system's capacity to (learn to) detect... etc. and label; and
>sh> that's also the grounding of the symbols. With a repertoire of symbols
>sh> thus grounded in that causal power,
>
> "Causal power" is meaningless without some specification of what it is
> that is caused and what causes it. If a mechanical icon functions in
> the system as a symbol because of an object/event/state or
> combination of same, apart from which it would not so function, then
> it is caused by that object/objects.
The robot has a learning mechanism that can learn, by trial and error,
with corrective feedback, to detect certain kinds of objects, on the
basis of their projections on its sensorimotor surfaces. That
transaction -- across time, as the object learns to detect the features
that will allow it to sort and label correctly -- is the grounding
process (but it need not occur in real time: the detectors could have
been built in: either way, the grounding is that CAPABILITY to sort and
label inputs in a certain way).
>sh> We can talk about it, but we can't explain it, or fit it into the causal
>sh> picture we have of the world
>
> Here, here! A point of rare agreement!
ReminderL we are talking here about feeling, not grounding, and we agree
we have nothing to say about it, so let's set it aside:
> sh> (except as an epiphenomenon with no independent causal power).
>
> But if it is logically impossible for will to lack causal power, then we
> simply have to make room for it. If we have to choose between
> an exception to nomic necessity and an exception to logical
> necessity, there is--literally--only one "logical" choice to be made
> between the two.
But it is not logically impossible for will to lack causal power (it is
merely phenomenologically impossible for it not to FEEL as if it has
causal power)! We no more have to make room for will as the true cause of
our actions than we need to make room for real objects as the true cause
of our hallucinations!
> df > If it were not conceivable, we could not understand what it is that
> df > we are not really able to do in not being able to "will" actions.
>
>sh> Of course we could! We know what it is feels like NOT to do something,
>sh> or not to be able to do it
>
> Sure...if we have a basis for understanding exactly what it is we are
> not doing.
No, it is merely one class of feelings ("feels like I'm doing it on
purpose") and another ("feels like I'm being pushed") that we have to be
able to tell apart. That's all just felt causality. Nothing to do with
real causality (except as an unexplained correlation in some cases).
>sh> I couldn't follow that. I can distinguish what I do voluntarily from
>sh> what I do involuntarily
>
> According to you, you do not "do" anything voluntarily or involun-
> tarily, you merely do certain things accompanied by a feeling that
> you are doing them voluntarily.
Correct.
> There is an objective difference
> between doing something a particular way and merely feeling
> that you are doing it that way.
I agree again, and that's precisely the difference I am drawing to your
attention when I say that not only it ain't necessarily so (i.e., it's
not true what you said was necessarily true earlier) but it it's almost
certainly false, on all available evidence and reasoning: There is no
non-telekinetic way that feelings could be independent causes (rather
than mere correlates of the true causes), and telekinesis is almost
certainly false.
>sh> or to do it involuntarily (pushed, or poked
>sh> on the patella), just as we know what it feels like to be unable to
>sh> see. And willed-doing versus induced-doing is not real?
>
> Coercion is coercion is whether internal or external. If I am having
> a bad dream, nothing external is coercing me to see disturbing
> images, but neither am I seeing them willingly. The bad dream is
> something "happening to" me. But given the closure of the causal,
> everything I think, say and do is something that "happens to" me.
I couldn't follow. Some things happen to me, some don't. Then again some
things feel-like they happen to me (and this feeling is indeed
correlated with things happening to me) and other things feel-like they
are not happening to me, but I am happening to them (i.e., causing
them). I am saying that in all cases that latter (agency) feeling is
false, though it does correlate with distinct kinds of happening-to-me
(the kinds of I/O and brain events that are correlated with the feeling
that I am doing something deliberately rather than that something
is happening to me).
The closure of the causal? That's fine in all cases but one: That
unexplained dangler, the hard one: feelings. We haven't the faintest
causal notion about that -- neither of the causes of feelings nor of the
effects of feelings (apart form what they feel-like). This is because we
have only a foggy idea what causes feelings (surely it's something about
the brain and the world), but, worse, we have no idea whatsoever about
how, or why, i.e., no idea about the causality!
That's about what causes feelings, how and why. But the EFFECTS of
feelings (apart form what they feel-like) are an even worse problem, for,
except on pain of telekinesis, they can have no independent effects at
all: The effects that follow them in real time, and that they feel-like
they are the cause of, are in reality merely correlated with feelings;
their causes are (probably) the same thing that causes the feelings --
and inasmuch as we can dope out those causes, the feelings simply
subtract out of the causal equation; hence they are epiphenomena.
>sh> So I can't feel as if I'm flying? Or as if I am travelling backwards in
>sh> time, or faster than the speed of light? It seems to me the limits on
>sh> feeling are merely the limits of our imagination, which certainly is not
>sh> restricted to just what is real!
>
> There is a difference between the feeling of imagining you are flying
> and the feeling that accompanies flying.
No, I'm talking about the feeling of flying when you are in fact not
flying: the hallucination of flying. We would not, after all, call a
hallucination "the feeling of imagining you are seeing": it is the
feeling of seeing -- except there's nothing there in reality! Ditto for
feeling of flying, and feeling of doing (agency). Only difference is
that sometimes when you feel that you see, you are not hallucinating,
and sometimes when you feel that you fly (e.g., in a plane) you are not
hallucinating either; so far when you feel you are flying with your own
wings you're pretty sure to be hallucinating, but perhaps that's not
true in principle. But when you feel you are DOING (other than the
discrimination between the feels-voluntary and feels-inadvertent) you
are ALWAYS hallucinating -- and, on pain of telekinesis, almost
certainly necessarily so (as awkward as that locution may sound).
> If we are talking about
> something that is possible, such as flying in a plane, there is a
> chance at least that the feeling of imagined flight will be the same
> as that of actual flight. But if we are talking about the impossible,
> that cannot be said.
What cannot be said? That it feels like I'm flying? Why not?
> The feeling of imagining that you are moving
> backward in time cannot correspond to the feeling of someone who
> is actually moving backward in time insofar as there is no such
> thing as actually moving backward in time.
I agree with that, assuming actually moving backward in time is
impossible, which is almost certainly necessarily true. Causes catching
themselves by the tail is surely as sci-fi as telekinesis, probably even
moreso. But nothing follows from this. Some feelings happen to have I/O
correlates, others do not.
> If there is no such thing as "will," then we cannot feel the
> same way as someone who actually wills actions.
If there is no such thing as will, then there is no such someone, and
hence nothing it feels like to be such a someone.
> So the question
> is, when we say that we only feel as if we will actions, do we mean
> that we feel the same way someone would who actually did will
> actions?
No. I rather think that feelings and reality are incommensurable, at
best merely correlated. I wouldn't want to swear that what a smooth
surface feels-like is what a smooth surface IS-like. It's probably
enough to say that what a smooth surface feel-like is simply what that
particular feeling usually happens to be correlated with (when you're
not hallucinating). But the only thing a feeling of willing is (usually)
correlated with is a certain kind of doing; and the causes of that kind
of doing will be found to be exhausted by its nonfeeling substrate (which
is probably also the cause -- somehow -- of the feeling [of willing,
or of any other feeling]).
So, no, feeling as if we are will actions is not somehow "like" REALLY
willing actions, because there is no such thing. It is like what it is
like, correlated with what it is correlated with, and in reality caused
(mysteriously) by whatever it is in reality caused by, and that cause is
probably also the cause of the action that we felt-like we were
willing!
> Not only do I think this is what we mean, I think we are
> incapable of thinking of this feeling any other way, although the
> contrary may be asserted argumentatively. Does our inability to
> think of it any other way prove that "will" actually exists? It gives
> us as sound a basis for believing so as for believing, for example,
> that we (meaning you or I as individuals) are autonomous selves
> (Hume notwithstanding), or that logical axioms are true.
It seems to me those two feelings/beliefs are rather similar (autonomy
being rather like freedom), and similarly wrong.
>sh> I'm lost. Feelings are real.
>
> I believe so, too. The question is, On what basis is our belief in
> their reality justified? See below.
That's easy: Descartes already told us: I feel, therefore I feel.
Otherwise put: feeling is being felt, therefore feeling exists.
>sh> (By the way, I think the notion of a "causal conjunction" of real things
>sh> and the imaginary conjunction of imagined things is somehow being
>sh> conflated here.)
>
> Funny, but that's the charge Raymore (following Levine) makes
> against Chalmers' "Zombie" argument. To quote:
>
> article "On Leaving Out What It's Like" by Joseph Levine 3 illustrates
> why a materialist need not give up hope due to Chalmers' arguments.
> Levine contends that arguments based upon what is conceivable (e.g.,
> the logical possibility of zombies) only establish epistemic conclusions
> rather than metaphysical conclusions. Merely being able to conceive of
> a zombie does not yet prove that a zombie could really exist, only that
> we can imagine that one could exist.
I agree completely. Maybe there cannot be Zombies. But if so, I want to
know how/why not. Indeed, I want to know how/why I am not a Zombie. I
don't for a minute believe there could be Zombies; and I of course
cannot believe I am a Zombie, because I am not (cogito, or rather
sentio). So Levine's point about Chalmer's thought experiment is quite
right. This is all an epistemic matter, not an ontic one. It is an
epistemic fact that no one can explain how/why we are NOT Zombies. The
hard problem (the feeling/function problem) is an epistemic, not an
ontic problem. It is an explanatory gap. That there cannot be Zombies
(for some mysterious reasons), even though true, does not help one bit!
>sh> be true). The relevant assumptions here are the causal and conservation
>sh> laws of current physics.
>
> ...laws, our understanding of which is only as sound as those logical
> axioms on which that understanding rests.
Well, yes, but irrelevant. There's nothing about feelings and cognition
that would make me ready to cast doubt on current physical theory. There
could be fairies too, but I wouldn't worry.
>sh> doing Y is what happens when you do Y, knowing Q is what happens when
>sh> you know Q and thinking R is what happens when you think R (the latter
>sh> two tending to be grounded symbolic states rather than just direct
>sh> sensorimotor states).
>
> "What happens" I take to refer to a sequence of physical events
> caused by other physical events. OK, let's briefly analyze this with
> an example. I stub my toe in the dark, as a result of which I yell,
> grab my toe and hop around the room on one foot. The same
> neural firing sequence that causes the external reactions causes a
> feeling of pain. Collaterally, it causes any knowledge or belief that
> is occurring to me about the incident, that is, it causes me to believe
> that I am in fact in pain. Let's diagram it as follows, labeling the
> initial neural event sequence "P1" (and please forgive my informal
> notation):
>
> P1 causes P2a (external actions: screaming, etc.)
>
> P1 causes M1 (feeling of pain)
>
> P1 causes P2b (belief that M1 is occurring)
>
> Each of these takes place in parallel, of course, so that M1 does
> not cause P2b. What is interesting is that M1 is contained within
> P2b. The question then is, why need M1 occur anywhere else
> except in P2b? If P2b is just a causal physical sequence, after all--
> a set of electrochemical reactions--then anything that
> causes P2b to occur will result in my belief that I am in pain.
I'm getting lost in the phenomenological hall of mirrors. By my lights,
M "need" not occur at all, not once, not twice, not ever. That it does
occur is the mystery, and the diagram just needlessly multiples the
mysteries rather than clearing them up.
> There is no need to propose something non-physical and
> "inexplicable" like the actual occurrence of a "feeling of pain." (I
> can hear Dennett breaking out the champagne.)
I can hear Descartes groan. The feeling is undeniable. The absence of
the how/why explanation of it in physical (functional, causal) terms is
almost equally undeniable (almost, because I could hallucinate a [false]
explanation, like William James's man who knew the secret of the universe
what he smelled nitrous oxide; perhaps Dan's in that state!).
> It would be a fond
> hope to believe that there is something in the nature of P2b that
> prevents its occurrence apart from the occurrence of M1, but since
> M1 is non-physical and P2b is physical and must be physically
> caused, there seems little to sustain such a hope. Looking back-
> ward at the causal chain, P2b as a purely physical event
> sequence is entirely opaque with respect to what lies behind
> it.
I'm afraid I have not learned anything from this symbolization. Still
stuck with the reality and inexplicability of feeling (except via
telekinesis).
> Chalmers' solution, as I understand, is to propose that there
> is a non-causal dependence or "acquaintance" relation between
> M1 and P2b, but this seems ad hoc. If P2b is really just a physical
> sequence, then any purely physical causes sufficient to produce
> it will result in my belief that I am in pain.
Chalmers' will remain famous for having given the "hard problem" its
proper name, but certainly not for making any inroads on it. The above
is just spinning wheels.
> There is still a slender thread that prevents us from falling into the
> clutches of Dennett, Churchland, et. al., and that is our old friend
> conceivability.
A Trojan Horse and false prophet: Beware!
> Is it conceivable that a systematic delusion that we
> have feelings could exist apart from the actual existence of
> feelings?
It certainly is (but that's neither here not there).
> I can by a mental stretch contemplate a psychotic
> delusion by which a someone believes himself to be in pain when
> he is not, but only one that is parasitic on the actual existence of
> pain and other feelings.
To wean this from semiotics: I can imagine a psychotic who is rewired to
feel pleasure from I/O that causes me pain. I could imagine a psychotic
who feels something like what I feel when I feel pain, but something
else too, something that makes the composite experience one he finds
gratifying, pleasurable. (This is still is not Dennett's Mr. Chase and
Mr Sanborn, feeling the same feeling but feeling differently about it --
which makes no sense, or rather tries to make a distinction between
identicals.) I can imagine a psychotic whose verbal report is
chaotically dissociated from his feelings. I can imagine a wimp, who
finds believes something to be excruciating that I would find mildly
irritating, and this could be because he is exaggerating or because his
pain-gain really is higher than mine (so again, there is no
Chase/Sanborn indeterminacy here: we are NOT feeling the same thing).
Bottom line: These exercises show nothing. The content of feelings,
differences between feelings, commensurability of feelings -- none of
this touch the hard problem. It is the FACT of feeling (any feeling,
even if there was only one to be had) that is the hard problem -- and
neither Chalmers, nor Dennett, nor conceivability arguments can cast any
light on it.
> But this amounts to an admission that the
> closure of the causal is inconceivable. Either M1 does indeed
> cause the physical sequence P2b, in which case closure is
> violated immediately, or else P2b itself is not entirely physical, in
> which case it is violated when beliefs or knowledge become the
> causes of our physical actions (assuming, of course, that they do).
Alas, I could not follow those conclusions, the premises having already
been so extravagant and irrelevant...
> In other words, Chalmers' "acquaintance" relation required for the
> P2b to be truthful rather than delusional is just logical causation as
> opposed to physical causation.
Maybe, but the structure of this labyrinth is so obviously unsound form
the outside that I can't even bring myself to enter...
And there is no logical causation. There is physical causation (enforced
by forces and their constraints) and there is logical necessity (on pain
of contradiction).
> As Lewis said (C.S., not David)
> an act of knowing must by nature be determined by the object of
> knowledge--that's what "knowing" something means.
To me, knowing is just a feeling that mysteriously correlates with
certain functional I/O capacities and states...
> Whatever
> happens when I "know" that I am in pain, it has to be something
> that could not happen in the absence of my actually being in pain
> for it to be genuine "knowing."
Too many words: The Cartesian "sentio ergo sentitur" already said it
all (without explaining anything, or solving any hard problems).
> Since a purely physical sequence
> is caused, not by what it is assumed to be "of" or "about" but by
> the preceding physical state, a purely physical sequence cannot
> be a genuine act of "knowing. Simple.
Simple, but unhelpful. Epistemic states are merely sentient states, with
certain correlates. The mystery is not that that the robot "knows" that
this is an X: that's just a functional capacity it has. The mystery is
that it feels-like something to know that this is X.
Chrs, Stevan
This archive was generated by hypermail 2.1.4 : Fri Nov 15 2002 - 22:34:54 GMT