From: Ben Goertzel (ben@goertzel.org)
Date: Fri Jun 14 2002 - 13:07:58 MDT
In Novamente, the overall system dynamics has an "evolutionary" nature,
which is related to but different from genetic programming as normally
conceived
On the other hand, explicit GP is also used in Novamente, for some purposes.
However, your comment is apt, James: GP requires a "fitness function," and
where does that come from? In Novamente GoalNodes define fitness functions
for GP, but they must be learned themselves, based on basic GoalNodes
supplied as a priori motivations...
Also, GP has scaling problems, which means that it can only be used to
evolve relatively small program modules, which must be pieced together and
integrated into larger functional systems by other means...
All in all, I think that evolution is an important *aspect* of intelligence,
and explicitly evolutionary algorithms are potentially a useful *component*
of an AGI, but they're far from the whole story...
ben g
> -----Original Message-----
> From: owner-sl4@sysopmind.com [mailto:owner-sl4@sysopmind.com]On Behalf
> Of James Rogers
> Sent: Friday, June 14, 2002 12:15 PM
> To: sl4@sysopmind.com
> Subject: RE: Threats to the Singularity: Time...
>
>
> On Fri, 2002-06-14 at 06:35, Martin Moore wrote:
> >
> > Instead, as is currently becomming pretty vogue,
> > allowing the program to write itself might be a faster
> > approach, even though it seems 'inefficient' due to
> > its randomness. However, I still think that a lot of
> > these genetic models are good, but still focus too
> > heavily on creating "that one perfect seed". It's way
> > too top-down.
>
>
> In my (limited) experience, GP is only tractable for tightly constrained
> models. For the purposes of AGI, this would mean that one would
> effectively be required to solve the problem before you could use GP to
> "solve the problem". GP is good for finding optimal solutions within
> well-defined boundaries and metrics though, at least in terms of being
> tractable for practical purposes. AFAIK, it fairs much poorer at
> unconstrained discovery type problems.
>
> -James Rogers
> jamesr@best.com
>
This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:39 MDT