Mark Waser wrote:
From: "Ben Goertzel" <[EMAIL PROTECTED]>
Sent: Thursday, June 15, 2006 12:17 AM
Subject: Re: [agi] How the Brain Represents Abstract Knowledge
You seem to be confusing Novamente with Richard Loosemore's system...
No, I don't think so . . . . I know that I know nothing about
Richard's system :-)
The way this dialogue evolved was:
* Loosemore was saying that he was having trouble tuning the many
parameters of his system
* I said that I had had this problem in Webmind, but not in Novamente
due to having a simpler design with fewer tunable parameters
* You complained that Novamente is not sufficiently modular
Funny, my recollection (backed by direct quotes) was that it went like
* Richard was saying that he was having trouble tuning the many
parameters of his system
* I said "I'm confused . . . . Why not have multiple independent
instances of the same mechanisms with different local parameters for
different processes? Once you uncouple the local parameters from
instance to instance, making all the processes happen at once should
be no more complicated than making them happen individually in
isolation." (Which, by the way, I now think that he IS doing to the
extent possible.)
* You said "These processes all have to play together nicely as they
are acting on the same data at the same time, and need to benefit from
each others' intelligence in order to make cognition happen.
Therefore, the parameters of all the cognitive processes need to be
"tuned together" -- a tuning that will work for one process in
isolation will not necessarily work for that process when it acts in
the context of other processes..."
Okay, since I started this debate, can I interject that I an NOT HAVING
TROUBLE tuning the many parameters in my model ;-)!
It would be more accurate to say that I am relishing the trouble with
the parameters, that having trouble with them is what my entire research
paradigm is all about! Just wanted to clear that up.
Ben has put my case himself in the above quote: this is all about
making cognition happen by admitting that cognition is the cooperative
effect of a cluster of processes that simply cannot be disentangled.
What I claim is that AI research (and cognitive science, for that
matter) has boxed itself into a corner by bending over backwards to try
to force intelligent systems to have decomposable, independent modules.
Everyone does this to some extent, although some do it much less than
others.
Now, Ben would probably not go as far as I am going, and would not want
to imply that all his modules are completely entangled with one another,
but I do want it understood that close-quarters entanglement is what my
approach is all about.
Why do it this way? (1) I think that all the cognitive science
literature points in that direction (even though the cog sci folks might
not like to admit that), and (2) I can see how to utterly simplify the
things that the cog sci folks are trying to describe IF I build a model
of cognition that embraces this kind of rich interconnectedness. I can
see, in principle, how to explain an enormously wide variety of
cognitive phenomena without having to resort to (what I see as) the
baroque complexities of many AI and AGI models.
Many of these models, indeed, seem to grow in complexity as a function
of the number of aspects of cognition that they try to embrace: it
looks as if each new aspect doesn't quite fit, and as a result provokes
an adjustment of the existing mechanism to make it fit. An optimist
would say that the AGI architect is just learning about the many subtle
things needed for intelligence, and is making sensible, motivated
additions to the model...... but a pessimist would say that these are
Epicycles!
But of course, all I have so far is a written formalism that purports to
show how these many aspects of cognition can be explained within a
unified system (and fragmentary implementations that show that some of
the aspects do work), but I cannot test such an approach without
systematically building and testing a very wide variety of systems. I
am caught between a rock and hard place: I can take the mechanisms in
isolation and try to shop them to the world (in which case, as I said, I
am open to the charge that when I fixed them up to substitute for the
missing "rest of the system", I was just kludging them. Or I can
stretch out my research program and build the tools I need to test the
system as a whole, and get hammered in the mean time for not actually
doing anything that counts. Very tricky.
Richard Loosemore.
-------
To unsubscribe, change your address, or temporarily deactivate your subscription,
please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
