🚀 go-pugleaf

In article <1992Sep25.193632.18452@mp.cs.niu.edu> rickert@mp.cs.niu.edu (Neil Rickert) writes:
>In article <7573@skye.ed.ac.uk> jeff@aiai.ed.ac.uk (Jeff Dalton) writes:
>>In article <1992Sep24.201519.17516@mp.cs.niu.edu> rickert@mp.cs.niu.edu (Neil Rickert) writes:

>>>It is natural, and in my opinion correct, to take a verificationist
>>>position when debunking Searle's supposed disproof of the possibility
>>>of strong AI.
>>
>>And I don't.  ...
>
>>How does this verificationist position work, in your view?
>>How does it avoid defending the idea that anything that passes
>>the TT (or other behavioral test) is conscious (or whatever 
>>we take the key issue to be)?
>
>It makes no such defense.  The verificationist position just says that
>Searle's argument is not a proof (or disproof if you prefer) of anything.

What are its reasons for this conclusion?  And why can't we make
the simple systems reply (as I suggested in my message) instead?
That is, I have some suspicions about these verificationist reasons.
If all you want is to show the conclusions don't follow you don't
need _verificationist_ reasons.

>The history of science and mathematics is replete with intuitively
>appealing ideas that have been shown to be wrong.  A verificationist
>position is just protection against falling into such a trap.

Sometimes it's too much protection.

>>Well, I try to give people lots of "yes, but" opportunities
>>to agree that something other than behavior can be relevant,
>>and it virtually never happens.
>
>But you do so in such strong language, and with so many attached
>qualifications, that people react against your offers.

I'm not sure what "attached qualifications" you have in mind,
or why they would stop someone from saying they hold a differently
qualified version (which would be a form of "yes, but" agreement).
Indeed, some of my qualifications -- such as "it's possible that
it will turn out that" instead of "it's the case that" -- ought to
make it _easier_ to agree.

As for strong language, this is a result of frustration after
earlier, more moderate, versions got nowhere.  I'm thinking of
things going back a number of years, not just of exchanges in
the last year or so.

>>>                                I think it is premature to be concerned
>>>about the rights of machines, since there aren't any strong AI machines
>>>as yet.
>
>>You could almost be quoting one of my messages here, because I
>>say the same about judgements of consciousness and understanding:
>>until there are such machines (that can pass the TT) and we know
>>more about how they work, etc.
>
>I have no disagreement with this position.

Good.

>>                                But again, virtually no one from
>>the "anti-Searle" side ever says they agree.
>
>I beg to differ.  Quite a few people from the anti-Searle position agree
>(and have stated so) that this is a supportable position.

Well, there are a lot of people on the anti-Searle side.  I guess
I overstated my point by saying "virtually no one".  Indeed, some
people on the anti-Searle side (McDermott, perhaps) are explicitly
not supporters of the TT.

>It is, after
>all, the verificationist position.  Putting it in other terms, Searle
>could have said "I'm skeptical about strong AI because of the CR argument.
>I won't believe it till I see it."  This would be a completely defensible
>position.  Most of the argument is because he goes much further and
>claims to have a final and convincing disproof.

But until I see what?  In his Chinese Room argument, Searle isn't
arguing that machines couldn't pass the TT, so seeing one pass it
wouldn't be a counterexample.  What other "it" do you have in mind?

>>              All people have to do is to point out that the
>>person in the Room, or the person who has memorized the Room,
>>is simply not in a position to see whether or not understanding
>>is taking place.
>
>I can take a verificationist position on that too.  You produce the
>working CR, and then I will join you in an investigation to determine
>if and where the understanding resides.  But until you can produce a
>working CR, the argument is just so much hot air.

Sure, you can take a verificationist position on almost anything.
Have I questioned this?

>Now don't get me wrong.  The CR makes a good argument as a tuition pump
>to see why many approaches to AI have failed.  It is when Searle claims
>it is a disproof that I am bound to disagree.  And I would disagree just
>as vehemently even if I believed strong AI were impossible, for a bogus
>proof is still a bogus proof, even if you like what it purports to prove.

I agree.

-- jd