OK. Let me give a system engineer's perspective . . . .
I believe that a lot of the current systems have done a lot of excellent,
rigorous work both at the bottom-most and top-most levels of cognition.
The problem is, I believe, that these two levels are separated by two to
five more levels and that no one is really even willing to acknowledge that
these levels exist and are necessary and will require *a lot* of work and
learning to implement.
We are not going to get to human-level intelligence with low-level
mechanisms and a scheduler. The low-level mechanisms are not going to
miraculously figure out how to assemble knowledge into a usable, scalable
foundation for more discovery and knowledge.
Most of the systems that are highly touted are actually merely generic
discovery systems with PLANS to extend to a complete cognitive system but
nothing more -- and most of them operate at a lower (i.e. data) level
(rather than a knowledge level) than makes sense if you're truly trying to
build a knowledge and cognitive system rather than a data-mining and
discovery system.
Most of the rest (Cyc, etc.) operate at the highest conceptual level but are
massive compendiums of bolt-ons with no internal consistency, rhyme, reason,
or hope of being extendable by the system itself.
Almost all of the systems are starting out way too large rather than trying
for a very small seed with rational mechanisms for growth and ways to
cleanly add additional mechanisms.
Most of the systems have too much Not-Invented-Here syndrome and as a result
are being leap-frogged by others who are intelligently using
Commercial-Off-The-Shelf or Open-Source software.
Note: Most of these complaints do *NOT* apply to Texai (except possibly the
two to five level complaint -- except that Texai is actually starting at
what I would call one of the middle levels and looks like it has reasonable
plans for branching out.
Richard doesn't express his arguments in easy to understand terms . . . .
but his core belief that we need more engineering to solve deep problems and
less hacking to quickly achieve low-hanging fruit (and then stalling out
afterwards) definitely needs to be given more currency.
----- Original Message -----
From: "Benjamin Johnston" <[EMAIL PROTECTED]>
To: <agi@v2.listbox.com>
Sent: Tuesday, May 06, 2008 4:36 AM
Subject: Re: [agi] AGI-08 videos
Richard Loosemore said:
But instead of deep-foundation topics like these, what do we get? Mostly
what we get is hacks. People just want to dive right and make quick
assumptions about the answers to all of these issues, then they get
hacking and build something - *anything* - to make it look as though they
are getting somewhere.
I don't believe that this is accurate. Those who I speak to in this
community (including the authors of papers at AGI-08 you claim have
produced hacks) give me the clear impression that they spend every day
considering deep-foundation issues. The systems that you see are not
random hacks based on quick assumptions, but the by-products of people
grappling with these deep-foundation issues.
This is certainly my experience. Every day I'm trying to grapple with
deeper problems, but must admit that I'm unlikely to solve anything from
my armchair. To create something that can be comprehended, critiqued and
studied, I have to carefully reduce my ideas to a set of what may be
almost laughable assumptions. However, once I've made such assumptions and
implemented a system, I have a much better grasp on the problem at hand.
From there, I can go back and explore ways of removing some of those
assumptions, I can try to better model my ideas, and I can rethink the
deeper issues with the knowledge I learnt from that experiment. When I
publish work on those concrete systems, I admit that I am not directly
discussing deeper issues. However, I believe that this method makes
communication much more effective and clear (I've tried both ways and have
experienced remarkably more success in conveying my ideas with sloppy
examples than with excellent arguments) and I believe that most readers
can look beyond the annoying but necessary assumptions and see the deeper
ideas that I am attempting to express. As I work on the problem further,
I'll create systems that are closer to my own ideas and may find ways of
distilling my ideas into more formal treatments of the fundamental issues.
I suspect that this experience is shared by most people here.
Ultimately, I think that any work in an area like AGI should be read with
attention to the things left "between the lines". In fact, I think that
expecting researchers to focus only on the fundamentals first is
counterproductive: not only will you end up with a whole lot of
hypothesizing with no connection to reality or experience, but you'll have
a whole lot of talk and opinion but no understanding of each other.
-Ben
-------------------------------------------
agi
Archives: http://www.listbox.com/member/archive/303/=now
RSS Feed: http://www.listbox.com/member/archive/rss/303/
Modify Your Subscription:
http://www.listbox.com/member/?&
Powered by Listbox: http://www.listbox.com
-------------------------------------------
agi
Archives: http://www.listbox.com/member/archive/303/=now
RSS Feed: http://www.listbox.com/member/archive/rss/303/
Modify Your Subscription:
http://www.listbox.com/member/?member_id=8660244&id_secret=101455710-f059c4
Powered by Listbox: http://www.listbox.com