Re: Imagery Debate

From: Harnad, Stevan (harnad@cogsci.soton.ac.uk)
Date: Sun Feb 04 1996 - 15:59:17 GMT


> From: "Lucas, Melody" <MFL93PY@psy.soton.ac.uk>
> Date: Sun, 4 Feb 1996 16:06:48 GMT
>
> Upon introspection, people claim to have images in their mind which
> illustrate their thoughts. As loose layman vocabulary this is fine.
> However, as psychologists, we cannot adhere to this idea as if I
> see something and have an image in my head of it, who is looking at
> the image? A 'little man'? So who sees his mental images in his
> head? It can go on ad infinitum. The imagery debate is concerned
> with describing what 'thing' is in our 'mind' when we think and how
> it occurs in the first place. The alternatives are images as
> propositions, as data structures, and a system of rules.

The imagery debate is partly about whether there are images in the mind,
though I think everyone concedes there is something they sort of "see"
when they're doing certain kinds of thinking. Few people would want to
deny that that experience is real. But the real debate is about the
CAUSAL ROLE of those images, or of any internal analogue structure at
all. Propositionalists challenge the causal role of images as
homuncular, and that's fine, but the same does not work for internal
analogues in general (they could be there and could deliver results,
without the need of any homunculus). So UNCONSCIOUS images would be no
problem at all; it's only the causal role of conscious images that is
problematic. But, as we know, the causal role of ANYTHING conscious is
problematic -- except if it floats without any independent causal power
of its own on top of something else that is unconscious and doing the
real work. So have the anti-imagists simply conflated the problem of
internal causation with the problem of conscious causation? And have the
pro-imagists done the same?

> Pylyshyn talked about representations and claimed that they were
> propositions (assertions) organised hierarchically. When a concept is
> active or if concepts are closely related, they go to the top of the
> hierarchy. He suggested that the representation is constructed from
> an interpretation of pictures and words, not a snapshot of them.

What is a "representation"? There's no problem with talking about
internal structures and processes, be they symbolic or analogue;
machines can have them; no reason brains can't have them too. But
"representation" is already getting homuncular: Representation TO WHOM?

So separate the problems: There are some kinds of tasks -- such as
doing arithmetic, logical reasoning, or other verbal tasks -- that seem
to be best done by machines that use symbol manipulation. No reason our
brains shouldn't do them the same way. Then there are other tasks --
 such as matching 2D images of rotated 3D objects -- that can be
performed by machines both by doing symbol manipulation and by doing
analogue rotation. Neither requires a homunculus; one might be more
efficient for some tasks than others (but there's no guarantee the brain
does everything in the most efficient way). The imagery debate can be
seen as being about the relative merits of the two ways of processing
information. But note that this dimension of the debate is indifferent
to whether the internal processing is conscious or not.

The issue of the role of consciousness is a separate one, and should
probably be separated from the imagery debate. And yet in some cases, we
really do feel as if we accomplish something by consciously manipulating
or otherwise consulting mental images, just as we might have done it by
manipulating external physical images. Even here, however, we have to
bracket the special case of the causal role of consciousness, and speak
merely about the internal structure/process.

> Kosslyn suggested that images have physical characteristics. He
> believes that images are produced by the same mechanisms as visual
> perception, and that effective visual processing is dependent on the
> imaging process. Supporting this, subjects have taken the same
> amount of time to, say, imagine scanning an object as it does to
> actually do so. Also, patients with hemifield neglect only have
> half a mental image. Shepard also provided evidence for this by
> showing how the reaction time taken to identify shapes is a function
> of the degree to which it is rotated from the vertical axis.

This all sounds like evidence that internal analogues may be more likely
than internal descriptions, but it is not definitive; the
propositionalists can give an account that would result in the same
timing. You need to say which you find more plausible, and why.

> Kosslyn was criticised because it is possible that the data acquired
> by the experiments mentioned earlier could have been influenced by
> the subjects acting how the researcher expected them to.
> To support Kosslyn's account, he created a computer model with the
> intention that it could do exactly what a human can. By doing so he
> conformed to implementation - dependence, and we all know that
> cognition is implementation - independent ( as is Pylyshyn's model).

I'm not sure why you made this implementational point here; it doesn't
seem relevant. If Kosslyn's model can do what people can do then it is a
model of how people do it. But in what sense was his model imagistic?

> Also, critics claimed that the visual buffer is primitive, and
> thus cannot be affected by expectations and intentions.

So, what is the significance of that?

> The timing of imagery may be congruent with the real thing
> as an effect of previous experience (or not?).

You are leaving your kid-bro far behind here: What point are you making?

> Pylyshyn may be criticised because she claims that cognition is
> computation. This has been disproved by the symbol grounding
> problem. Also, the nature of the hierarchical structure of imagery
> is not congruent with the PDP models which are becoming more popular
> for mind modeling.

This is unfortunately very garbled. Pylyshyn (a "he" by the way) has
claimed that cognition is computation. There is no proof except in
mathematics, so one can only speak of evidence for or against this
theory, not of proof. The symbol grounding problem, however, is neither
proof nor evidence: it is a problem: how to ground symbols without
relying on an external interpreter. You probably had the Chinese Room
Argument in mind, which is likewise not a proof, but an argument against
computationalism. It is unclear what the "hierarchical structure of
imagery" is (Pylyshyn spoke of the hierarchical structure of
propositions), nor is it clear whether or not PDP models (you have not
explained them) are congruent with the structure you have in mind, or
what they have to do with the imagery debate. This will have to be sorted
out carefully.

> P.S. Am I supposed to know the ANSWER to this (I suppose if there
> were an answer, it wouldn't be an imagery DEBATE).

Not the answer, but you should be very clear -- kid-bro clear -- on what
the issues are and on what the evidence and arguments on both sides are.
Then you should say which side you favour, and why.



This archive was generated by hypermail 2b30 : Tue Feb 13 2001 - 16:23:58 GMT