Newsgroups: comp.ai.philosophy
Path: cantaloupe.srv.cs.cmu.edu!das-news2.harvard.edu!news2.near.net!howland.reston.ans.net!cs.utexas.edu!news.unt.edu!hermes.oc.com!internet.spss.com!markrose
From: markrose@spss.com (Mark Rosenfelder)
Subject: Re: Penrose and Searle (was Re: Roger Penrose's fixed ideas)
Message-ID: <D0ID0s.35w@spss.com>
Sender: news@spss.com
Organization: SPSS Inc
References: <CzzuEu.F48@gpu.utcc.utoronto.ca> <jqbD0F4yH.E7v@netcom.com> <3c4u97$hch@news1.shell> <D0I9n9.528@gpu.utcc.utoronto.ca>
Date: Thu, 8 Dec 1994 20:02:52 GMT
Lines: 67

In article <D0I9n9.528@gpu.utcc.utoronto.ca>,
Andrzej Pindor <pindor@gpu.utcc.utoronto.ca> wrote:
>In article <3c4u97$hch@news1.shell>, Hal <hfinney@shell.portal.com> wrote:
>>Where are the internal mental states in an HLT?  What if he considered
>>saying A but instead said B?  Are you claiming that the HLT contains
>>or implies those rejected alternatives?
>>
>Of course it has to contain those rejected alternatives! How could it be
>otherwise? HLT is supposed to contain _all_ possible converations, doesn't it?
>Those "rejected" alternatives are possible answers too, aren't they?
>So why wouldn't they be in it too? 

The HLT need not contain all possible answers to every conversation-sequence;
just *one* possible answer. 

To make this clearer, let's take an example.  Suppose the conversation
so far is:

   Judge:   Well, let's get going.  What's 15 * 32?
   HLT:     Umm... about 450.
   Judge:   'About'?  Don't you know the exact answer?

At this point several responses are theoretically possible: 

   OK, 472.
   I just estimated it as 15 * 30.  If you want an exact answer, add 2 * 15.
   Sorry, I didn't write down the two numbers, what were they again?
   I'm no good at arithmetic.
   I hate these tests.  Let's say 461 and to hell with it.
   Why do you care?  
   ...
   
You seem to be imagining an HLT-- call it HLT-N-- which contains *all* these
answers, and employs some decision procedure to choose between them.
But one could just as well (in fact, more easily) create an HLT (call it
HLT-1) which contains just *one* of these answers (say, the first one).
In HLT-1, there is nothing which corresponds to Hal's "rejected alternatives".

>Consequently at some stage the program
>has to decide somehow which alternative to use and there may be quite a few
>of them. If you want to suggest that the choice might be done randomly, note
>that this is not a case for humans. A conversation usually shows a certain 
>pattern which we ascribe to personal traits. If we could not detect something
>akin to a personality we certainly would be reluctant to treat a source
>of answers as a "person".

Here you seem to be a bit confused.  There is *no way to tell*, externally,
if we're dealing with HLT-N or HLT-1; nor can we tell, externally, if HLT-N
chooses between responses randomly or some other way.  There is no pattern 
to look for; in fact HLT-1 might have been constructed by running HLT-N and 
choosing the single actual response to enter into its own table.  But even 
if it was constructed differently, HLT-1 has just as much "personality" as 
HLT-N.

>>Dalton has said, I believe, that he believes that future knowledge of and
>>understanding of how consciousness and intelligence works may give us
>>guidelines to use in judging whether a program is conscious or just
>>a mimic.  In this view, it is not possible today to give a detailed
>>answer to your question.  Broadly speaking, we might expect to look for
>>certain data structures and algorithms which could be mapped to the
>>mental states which precede speech.  We would not find these in the HLT.
>
>Not true. As I have said before, you are ignoring a very complex task of
>selecting a proper answer. This selection proces would map to the mental
>states.

This answer may cover HLT-N, but it doesn't cover HLT-1.
