AI project & concepts
- From: "JGCASEY" <jgkjcasey@xxxxxxxxxxxx>
- Date: 28 Oct 2005 17:48:39 -0700
>> ... so if we here in c.a.p were to create
>> our own little AI contest for the purpose
>> of advancing the state of the art, what
>> should it be?
>
>
>
> Excellent idea, Curt. Let me see if I can
> contribute to the discussion.
>
> Stated in abstract terms, the thing that I
> value most in an AI is the ability to deal
> with concepts. These are very important
> because they supply the basis for the
> inference mechanisms, which, in turn,
> provide the support for semantics and that
> is the Holly Grail of AI. I believe that
> the number and complexity of the concepts
> that an AI is able to deal with defines
> the degree of maturity of the system, as
> is the case also for humans.
I don't think the number of concepts is
an issue. We don't have that many. As for
the complexity of a concept you would have
to be clearer on what you mean by that.
One of the advantages that human intelligence
appears to have is the ability to deal
with abstract concepts.
> However, when stated in concrete terms,
> it is a bit more difficult to establish a
> specific goal for an AI. The problem is
> that a concept can take many forms,
> depending on the focus of the system:
> A visual concept takes a very different
> form from a language concept. In these
> circumstances, and given the fact that
> real systems deal only with a particular
> subset of the environment, to establish
> a common ground, it would be required
> that all the systems to compare deal
> with the same aspect of the world.
But aren't concepts ultimately just deciding
what parts go together? The favorite example
being the concept of a bird where a bird
like the Robin is a typical example and a
penguin is not.
> Taking the DGC example, an intelligent
> system, after learning how to deal with
> several concrete instances of a particular
> obstacle, would need to demonstrate that
> it could deal with an instance of that
> same obstacle that it had never seen before.
> This would prove that the system understands
> that particular class of obstacles.
In fact I think the opposite happens. A robot
avoids "objects" and only a more sophisticated
robot classifies them according to some criteria
such as size, function or whatever attributes
are relevant to the robot.
Although there appears to be a lot of antipathy
toward GOFAI in poster to this newsgroup one
thing is clear, they may not be "true" AI but
they work, they do play chess, read car number
plates, collect tin cans and so on. Something
that hasn't been achieved by any other means.
I assume "real" AI means replacing the programmer
in the GOFAI projects. This means, I think, going
back to first principles which involves goal
seeking behavior. These GOFAI are after all goals
set by programmers needs. At the lowest level
goal seeking behavior is just a system moving
toward a state of equilibrium.
--
John Casey
.
- Prev by Date: Re: CAP Challenge
- Next by Date: Re: What is Maslow's hierarchy really?
- Previous by thread: is here anyone interested in Hegel?
- Next by thread: In memory of the extreme religious psychotics going into he depth of my mind in this NG to kill me, my soul
- Index(es):
Relevant Pages
|
|
