On Fri, 2008-03-14 at 18:46 -0400, Alan Grimes wrote:
> Florent wrote:
> > "A creation of researchers from Rensselaer Polytechnic Institute,
> > Eddie has his own set of beliefs, and the ability to reason about his
> > beliefs to draw conclusions in a manner that matches human children
> > his age.
> >
> > This includes a partially-developed "Theory of Mind", which allows him
> > to understand, predict and manipulate the behaviour of other agents
> > and of even human players, with whom researchers expect the technology
> > to be able to one day interact with in the real, physical world."
> >
> >
http://itnews.com.au/News/72057,childlike-intelligence-created-in-second-life.aspx>
>
> From the article:
> ################################
> Currently, the team is grappling with computational tractability issues
> to do with the sorting of growing amounts of knowledge that is collected
> as a artificially intelligent character matures.
> #################################
>
> This is a litmus test of AI. A true AI will have exceptionally *good*
> complexity. -- meaning this project is NOT an AI. I do congratulate the
> team though for using a darn good approach to developing the AI, putting
> it in an actual environment.
>
>
I don't understand your comment relative to the quote?
I have written some tough code, and one of the classics is a problem
that deals with the mini-max problem (minimum solution set for maximum
coverage in my case). I generated a tree structure, but it would grow
basically without bounds (the sets were unbounded, so this could be
expected), which produces a computationally intractable problem for the
hardware I had at the time. So computational tractability is I think an
issue of hardware response time vs total data or the details of the
reduction to be performed.
Your statement that true AI will have exceptionally "good" complexity I
also do not understand. Isn't complexity in general bad? As we move
farther and farther into understanding neural structures and physical
mechanisms in reality (the simulation of which is one form of AI), it
seems that nature has been quite good at discovering ways to utilize
simple structures over and over, but combining them in different ways to
achieve results that are effective for that particular organism. Seldom
is nature complex, any more than a fractal is complex as I understand
it.
It would seem that our brains are somewhat redundant, biased to provide
for damage, and structured to help eliminate the possibility of total
failure. As to the degree of goodness in the neurological structure's
complexity, at this time, we do not fully realize, or at least I haven't
encountered a good explanation of the deductive and associative process
in bulk mental processing. If you have a good reference I would like to
read it.
This field interests me intensely. But I am a bit behind, since my
last foray in coding for intelligence was 2003.
Regards,
Les H
Locked
