From newshub.ccs.yorku.ca!ists!torn!utgpu!news-server.ecf!utcsri!rutgers!jvnc.net!yale.edu!qt.cs.utexas.edu!cs.utexas.edu!sun-barr!decwrl!mcnc!aurs01!throop Tue Jun 23 13:20:55 EDT 1992
Article 6282 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!torn!utgpu!news-server.ecf!utcsri!rutgers!jvnc.net!yale.edu!qt.cs.utexas.edu!cs.utexas.edu!sun-barr!decwrl!mcnc!aurs01!throop
>From: throop@aurs01.UUCP (Wayne Throop)
Newsgroups: comp.ai.philosophy
Subject: Re: 5-step program to AI
Message-ID: <60835@aurs01.UUCP>
Date: 17 Jun 92 14:22:46 GMT
References: <1992Jun16.213227.31307@mp.cs.niu.edu>
Sender: news@aurs01.UUCP
Lines: 83

> rickert@mp.cs.niu.edu (Neil Rickert)
>>> Steps to ai: rocks, germs, frogs, mice, chimps, humans
> I would argue that a frog already has adequate "grounding", although
> I won't claim "intentionality" since that is too tied up with a poorly
> defined "consciousness".

Fair enough.  Though I'd quibble and say that the frog's symbols
are *minimally* grounded.  To try to be a bit less vague about
what I mean by "groundedness", I'd say it has to do with a measure
of the richness of a reference.  That is, I didn't mean it to be a
boolean.

> I prefer to avoid terms like "grounding" and "intentionality" since
> they are poorly defined. [.. but ..]
> If Pavlov's dog can salivate at the ring of
> a bell, I would consider that strong evidence that the bell sound is a
> well grounded symbol for that dog. 

To rephrase the above to avoid "stealing" Harnad's term in ways he
almost certainly doesn't intend, I'd say that "I would consider 
[the dog's salivation at the sound of a bell] to be  strong evidence
that the dog treats the bell as a reference to food of considerable
richness.".   (Uh... the reference being rich, not the food...)

It seems to me that there is a considerable jump in the "richness" of
"reference-ness" of symbols between a frog and a dog.  My suspicion is
that this jump involves something other than "just" faster and more
flexible pattern matching or learning (though I may well be wrong about
that).  

John Nagle mentions that "Currently, we seem to be in the insect-level
AI era.". I think the "richness" distance between "bugs" and frogs is
smaller than that between frogs and dogs, but I wouldn't be surprised
to be wrong about this.  So I wouldn't be uncomfortable with going along
with "insect-level" as a better intuitive characterization than
"frog-level".

>>I think that the apparent facility with "higher reasoning" that
>>computers currently have IS essentially fiddling with "meaningless
>>squiggles and squoggles", and any meaning is only read in by humans
>>(or at least, to a very large degree).
> This is why I consider computer intelligence to be approximately the
> increment from the chimpanzee to the human.

Hmmmm.  Interesting.  That would fit in with what I said just above.
However, it still seems to me that the "symbol manufacture" that
computers can do isn't quite the same thing as what humans
(as opposed to other mammals) do.

So, all in all, I suppose it wouldn't surprise me if Neil's intuitions
about this are entirely right, but I don't myself have similar
intuitions.  (I hope that's an adequate way to put it...)

As an afterthought, maybe amend the steps to AI to be

         rocks, germs, bugs, mice, chimps, humans

Or maybe better still (since I think of the difference between chimps
and humans as more of a matter of degree than the difference between
chimps and rodents) (especially if one takes Koko seriously):

   entity       extra capability compared to previous entity
   ------       --------------------------------------------
   rock         none
   germ         stimulus response               algorithmic programs
   bug          "contentless' object modeling   current AI
   rat          "rich" object modeling          future AI
   ape          abstract symbol use             TT-passing ability

The above table, of course, is much more precise-seeming than it
"ought" to be.  The boundaries between these "levels" or "stages" are
much fuzzier than they seem here.  But the table does have the pleasing
feature of pithy single sylable names for each stage.

We can also add, beyond "ape":

   entity       extra capability compared to previous entity
   ------       --------------------------------------------
   god          omniscience                     oracle for true sentences        
                                                of formal systems
Or maybe we can't...

Wayne Throop       ...!mcnc!aurgate!throop


