From: Richard Loosemore (rpwl@lightlink.com)
Date: Thu Jan 19 2006 - 11:00:47 MST
Daniel,
In spite of your comments (below), I stand by what I said. I was trying
to kill the Searle argument because there is a very, very simple reason
why Searle's idea was ridiculous, but unfortunately all the other
discussion about related issues, which occurred in abundance in the
original BBS replies and in the years since then, has given the
misleading impression that the original argument had some merit.
I will try to explain why I say this, and address the points you make.
First, it is difficult to argue about what *exactly* Searle was claiming
in his original paper, because in an important sense there was no such
thing as "exactly what he said" -- he used vague language and subtle
innuendos at certain crucial points of the argument, so if you try to
pin down the fine print you find that it all starts to get very slippery.
As example I will cite the way you phrase his claim. You say:
"He claims ... that no additional understanding is created anywhere, in
the room or in the man, and so Strong AI is false."
How exactly does Searle arrive at this conclusion? In Step 1 he argues
that the English speaking person does not "understand" Chinese. If we
are reasonable, we must agree with him. In Step 2 he says that this is
like a computer implementing a program (since the English speaker is
merely implementing a computer program). In Step 3 he goes on to
conclude that THEREFORE when we look at a computer running a Chinese
understanding program, we have no right to say that the computer
"understands" or is "conscious of" what it is doing, any more than we
would claim that the English person in his example understands Chinese.
My beef, of course, was with Step 2. The system of mind-on-top-of-mind
is most definitely NOT the same as a system of mind-on-top-of-computer.
He is only able to pull his conclusion out of the hat by pointing to
the understanding system that is implementing the Chinese programme
(namely the English speaking person), and asking whether *that*
understanding system knows Chinese. He appeals to our intuitions. If
he had proposed that the Chinese program be implemented on top of some
other substrate, like a tinkertoy computer (or any of the other
gloriously elaborate substrates that people have discussed over the
years) he could not have persuaded our intuition to agree with him. If
he had used *anything* else except an intelligence at that lower level,
he would not have been able to harness our intuition pump and get us to
agree with him that the "substrate itself" was clearly not understanding
Chinese.
But by doing this he implicitly argued that the Strong AI people were
claiming that in his weird mind-on-mind case the understanding would
bleed through from the top level system to the substrate system. He
skips this step in his argument. (Of course! He doesn't want us to
notice that he slipped it in!). If he had inserted a Step 2(a): "The
Strong AI claim is that when you implement an AI program on top of a
dumb substrate like a computer it is exactly equivalent to implementing
the same AI program on top of a substrate that happens to have its own
intelligence," the Strong AI people would have jumped up and down and
cried Foul!, flatly refusing to accept that this was their claim. They
would say: we have never argued that intelligence bleeds through from
one level to another when you implement an intelligent system on top of
another intelligent system, so your argument breaks down at Step 2 and
Step 2(a): the English speaking person inside the room is NOT analogous
to a computer, so nothing can be deduced about the Strong AI argument.
So when you say: "Searle never claims that since 'understanding doesn't
bleed through,' Strong AI is false." I am afraid I have to disagree
completely. It is implicit, but he relies on that implicit claim.
And while you correctly point out that the "Systems Argument" is a good
characterisation of what the AI people do believe, I say that this is
mere background, and is not the correct and immediate response to
Searle's thought experiment, because Searle had already undermined his
argument when he invented a freak system, and then put false words into
the mouths of Strong AI proponents. My point is that the argument was
dead at that point: we do not need to go on and say what Strong AI
people do believe, in order to address his argument.
In fact, everyone played into his hands by going off on all these other
speculations about other weird cases. What is frustrating is that the
original replies should ALL have started out with the above argument as
a preface, then, after declaring the Chinese Room argument to be invalid
and completely dead, they should have proceeded to raise all those
interesting and speculative ideas about what Strong AI would say about
various cases of different AI implementations. Instead, Searle and his
camp argued the toss about all those other ideas as if each one were a
failed attempt to demolish his thought experiment.
Finally, Searle's response to the mind-on-mind argument was grossly
inadequate. Just more of the same trick that he had already tried to
pull off. When he tries to argue that Strong AI makes this or that
claim about what a Turing machine "understands," he is simply trying to
generalise the existing Strong AI claims into new territory (the
territory of his freak system) and then quickly say how the Strong AI
people would extend their old turing-machine language into this new
case. And since he again puts a false claim onto their mouths, he is
simply repeating the previous invalid argument.
The concept of a Turing machine has not, to my knowledge, been
adequately extended to say anything valid about the situation of one
Turing machine implemented at an extreme high level on top of another
Turing machine. In fact, I am not sure it could be extended, even in
principle. For example: if I get a regular computer running an
extremely complex piece of software that does many things, but also
implements a Turing machine task at a very high level, which latter is
then used to run some other software, there is nothing whatsoever in the
theory of Turing machines that says that the pieces of software running
at the highest level and at the lowest level have to relate to one
another: in an important sense they can be completely independent.
There are no constraints whatsoever between them.
The lower level software might be managing several autonomous space
probes zipping about the solar system and interacting with one another
occasionally in such a way as to implement a distributed Turing machine,
while this Turing machine itself may be running a painting program. But
there is no earthly reason why "Turing machine equivalence" arguments
could be used to say that the spacecraft system is "really" the same as
a painting program, or has all the functions of a painting program.
This is, as I say, a freak case that was never within the scope of the
original claims: the original claims have to be extended to deal with
the freak case, and Searle disingenuous extension is not the one that
Strong AI proponents would have made.
Richard Loosemore.
Daniel Radetsky wrote:
> On Wed, 18 Jan 2006 08:09:43 -0500
> Richard Loosemore <rpwl@lightlink.com> wrote:
>
>
>>END OF ARGUMENT.
>
>
> If you don't want to talk about Searle, don't talk about Searle, but don't give
> a set of reasons why not to talk about Searle, and expect me not to respond.
>
>
>>He proposed a computational system implemented on top of another
>>computational system (Chinese understander implemented on top of English
>>understander). This is a mind-on-top-of-mind case that has no relevance
>>whatsoever to either (a) human minds, or (b) an AI implemented on a
>>computer.
>
>
> This is a version of a response made a long time ago by Jerry Fodor. Searle
> responded, and very adequately I think. Since the mind-on-top-of-mind is
> something which is implementing a Turing machine, it is the same thing
> computation-wise as anything else implementing a Turing machine. So it is
> completely relevant to whether or not a computer (something implementing a
> Turing Machine) can be conscious.
>
> I'll be blunt: if you want to challenge Searle, use the Systems Reply. It's the
> only reply that actually works, since it explicitly disagrees with Searle's
> fundamental premise (consciousness is a causal, not a formal, process). You
> went on to make something like the Systems Reply in the rest of your post, but
> against a straw man. Searle never claims that since 'understanding doesn't bleed
> through,' Strong AI is false. He claims (in the original article; I haven't read
> everything on this subject) that no additional understanding is created
> anywhere, in the room or in the man, and so Strong AI is false. That is, the
> fact that 'understanding doesn't bleed through' is only a piece of the puzzle.
>
> Daniel
>
>
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:55 MDT