Re: cognitive != observable? (project three)

Eray Ozkural exa wrote:
> Patty <[EMAIL PROTECTED]> wrote in message news:<[EMAIL PROTECTED]>...
>
> > Now obviously any two instances of identical phrases used 
> > against the same context vector would not necessarialy refer 
> > to exactly the same thing, but me thinks it would be close 
> > enough to start playing the language game.
>
> Yes, we can start playing the game but I was making a theoretical
> point. For true NLU, we need to resolve those referents which will not
> be immediate from a purely syntactical analysis.
>
> My point was that determining referents is an AI-complete problem. And
> that is only one of the many problems for a semantics analyzer. You
> can't really tell the referents without a world model and deliberate
> perception and imagination of the context.
>
> That might not be necessary for a chatbot or a personal assistant
> (although I imagine people will readily expect such skills from a bot
> that is seemingly competent)

Perhaps I should emphasize more that the stratagy is to 
create a platform on which we can *start* evolving.  Even a 
baby does not know its references when it *starts* to babble.

One of the probelms that I see with the development of human 
level intelligence that is no carry over from one AI project 
to another - there is no evolution of artificial 
intellegence itself.  Oh sure, we stand on each others 
successes - that level of technological evolution is always 
happening; and the literature and practices of science is, 
in a sense, an evolving AI.  Also we can look at the 
Internet, the web, google etc,  as a kind of evolving AI.

My proposal is that we define a platform, an environment, in 
which a species of AI processes can evolve using ALife 
stratagies and NN stratagies and whatever other stratagies 
someone can think of.  There is no need to restrict which 
stratagies this species of processess (i like to call them 
gremlins) can use.  If the psysical environment of these 
gremlins is stable enough,  is useful enough for us humans 
to sustain it, then these gremlins *will* evolve.  As the 
human host, you would not be speaking to a single gremlin - 
rather you would be speaking to a whole society of gremlins.

Let me give just one example of what one gremlin might do.

Given the the following fragment of the enviroment:
(SENTENCE (SUBJECT john) (VERB gave) (DIRECT-OBJECT (a 
ring)) (INDIRECT-OBJECT lucy )
There may be a gremlin that would emit the folowing to the 
enviromment:
(CONTEXT whatever#1
    (_:a#1 $type giving)
    (_:a#1 $giver john)
    (_:a#1 $recipient lucy)
    (_:a#1 $gift (_:a#1 $type ring)))
That is all this particular gremlin could ever do.  An 
instance of the gremlin would come alive based upon (VERB 
?x) in the enviroment.  If that particular behavior was not 
found to be useful, then this species of gremlin would die 
off.

Now you probably see in that one example pretty standard 
knowledge engineering techniques. But bear in mind that we 
are not limiting gremlins to knowledge engineering technques 
- but at the same time we are not excluding them.

Now what do you think?

Note: newsgroup comp.ai.nat-lang added - this discussion 
belongs there because it is more about engineering.

Patty