From: Ben Goertzel (ben@goertzel.org)
Date: Wed Dec 22 2004 - 16:28:56 MST
Hi all,
Recently David Hart -- who's been working with me on AGIRI (the nonprofit
side of the Novamente project, which hasn't been very active, but we'd like
to make it more so, because funding Novamente AGI development via the
minimal revenue from our narrow-AI consulting projects is proving
*massively* too sluggish for our tastes...) -- recently posted an email on
the list pertaining an AGIRI fundraising drive.
By way of commentary on this, I'll post here an excerpt from the informal
"Concluding Remarks" to the latest, still-not-yet-finished draft of the
ever-evolving and improving "Novamente book" (now three books). I have 6
chapters to revise out of the 29 in the book and then this version will be
ready to send off to the publisher ;-))
-- Ben G
The excerpt:
**********************
Reactions of people who have read early drafts of this book have varied
widely. Most readers are surprised and at least mildly impressed that
anyone has taken the trouble to work out a serious AGI design in so much
detail. However, not surprisingly, there is a lot of skepticism regarding
our conjecture that, when fully implemented and tuned and taught, the
Novamente design will actually lead to general intelligence at or beyond the
human level. The only complaints that really interest us are the ones made
by people who believe that some AGI design is possible. Among these
readers, the primary complaints made seem to be things like:
a.. The design is too complicated, there are too many parts to coordinate,
too many things that could go wrong
a.. Answer: Yes it IS complicated, and we wish it were simpler, but we
haven’t found a simpler design that doesn’t seem patently unworkable. Note
that the human brain is also mighty complicated – this may just be the
nature of making general intelligence work with limited resources.
b.. BOA and PTL are not enough, you need some kind of more fundamentally
innovative, efficient, or (whatever) learning algorithm. This complaint
never comes along with any suggestion regarding what this “mystery
algorithm” might be, though – most often it is hypothesized that detailed
understanding of the human brain will reveal it.
a.. Answer: This is possible, but it seems to us that a hybrid of BOA
and PTL will be enough. The question is whether deeper integration of BOA
and PTL than we’ve done now will allow BOA learning of reasonably large
(500-1000 node) combinator trees. If so, then we almost surely don’t need
any other learning algorithm, though other algorithms may be helpful.
c.. You’re programming in too much stuff: you should be making more of a
pure self-organizing learning system without so many in-built rules and
heuristics
a.. Answer: Well, the human brain seems to have a lot of stuff
programmed in, as well as a robust capability for self-organizing learning.
Conceptually, we love the idea of a pure self-organizing learning system as
much as anyone, but it doesn’t seem to be feasible given realistic time and
processing power and memory constraints.
d.. Programming explicit logical rules is just wrong; logic should occur
as an emergent phenomenon from more fundamental subsymbolic dynamics
a.. Answer: Probabilistic logic is not necessarily symbolic; in the
Novamente design we use PTL for both subsymbolic and symbolic learning,
which we believe is a highly elegant approach. The differences between
subsymbolic probabilistic logic and e.g. Hebbian learning are not really
very great when you look at them mathematically rather than in terms of
verbiage. The Novamente design is not tied to programming-in logical
knowledge a la Cyc. It’s true that the PTL rules are programmed in (though
in Novamente 2.0 they will be made adaptable), but this isn’t so different
from the brain having particular kinds of long-term potentiation wired in,
is it?
Of course, we don’t expect our answers to fully please the complainants –
but they generally seem to convince the complainants that we’ve thought
through the issues carefully. Almost no one who has seriously studied
Novamente has accused us of being naïve or foolish in our approach, though
obviously many have considered our confidence excessive, and have proposed
other directions as seeming more promising.
At time of writing, our biggest frustrations with the Novamente project are
twofold. The first is a technical frustration: existing software tools are
a pain to work with. Both Windows and Unix carry with them their own forms
of irritation, similarly with all existing programming languages: C++, Java,
LISP, functional and logical languages. All existing operating systems and
programming languages place unnecessary hassles in the way of creating a
powerful general AI. As a Novamente software developer, one spends too much
of one’s time wrestling with purely software-level issues that have no
direct connection to AI. And needless to say, the software industry does
not put much effort into creating software development environments that are
specifically conducive to AGI R&D!
The second frustration is financial: because of the radical nature of the
project, we have not been able to secure a significant amount of R&D
funding. We have sustained the project, so far, based on a combination of a
small amount of R&D funding with revenues derived from small businesses that
we’ve built by deploying portions of the Novamente system within narrow-AI
software products. This is not the right way to build a thinking machine.
What the Novamente project needs is a small group of programmers and
scientists dedicated full-time to creation of AGI. We view the current
phase of Novamente development as preliminary – our goal is to get the
system to the stage where it does things that are exciting enough that
marshalling significant R&D funding becomes easy. At that point we’ll be
able to pay a top-notch full-time team (including the authors and the other
Novamente originators) to complete the job, and progress will start going an
awful lot faster.
We mention these frustrations not out of the desire to personally vent, but
mainly because of the general point that they illustrate: neither
contemporary human society, nor the computer science community, nor even the
AI community is at all supportive of the quest to create powerful AGI
software. We believe that in hindsight, after AGI has been created, this
lack of support and enthusiasm will be viewed with incredulity. AGI is a
hard problem but it’s far from an impossible problem – it seems clear that
there are many possible solutions, and we believe that Novamente is one of
them. Almost surely it’s not the best one, but so far as we know it’s the
only likely-looking solution that’s been proposed in detail so far.
************
-----Original Message-----
From: owner-agi@v2.listbox.com [mailto:owner-agi@v2.listbox.com]On Behalf Of
David Hart
Sent: Tuesday, December 21, 2004 11:20 AM
To: agi@v2.listbox.com
Subject: [agi] Artificial General Intelligence Research - Help Wanted
Hi All,
AGIRI (the Artificial General Intelligence Research Institute) is pleased to
launch a new drive to raise $80,000 dedicated to pure AGI research with a
specific goal: to complete the next major milestone of Novamente development
by the end of 2005.
Milestone M10, shown in the Novamente Development Roadmap, is the deployment
of a Novamente system specialized to control an embodied agent, interacting
linguistically and "physically" with other agents and objects in a simulated
world. For more information on the project, see the AGI-SIM Development
Plan. AGIRI is also seeking capable programmers, for both volunteer and paid
positions.
We believe that upon completion of the AGI-SIM project, AGIRI will be
well-poised to raise the larger-dollar research funding needed to move
Novamente quickly toward its longer-term AGI goals. The AGI-SIM application
of Novamente may also open up new commercial possibilities, for instance in
the area of partnerships with firms developing mobile robotics technology.
For more information about Novamente, see the new concise but deep
single-page description located at the AGIRI home page, and the eight-page
technical paper at Novamente: An Integrative Architecture for General
Intelligence. A new series of books that will describe Novamente's
conceptual and technical under-pinnings in great detail is due for
publication in 2005.
Read more about fundraising at AGIRI on the contributions page.
Best Regards,
from all of us at AGIRI
-- AGIRI Team http://agiri.org ---------------------------------------------------------------------------- ---- To unsubscribe, change your address, or temporarily deactivate your subscription, please go to agi@v2.listbox.com">http://v2.listbox.com/member/?listname=agi@v2.listbox.com
This archive was generated by hypermail 2.1.5 : Tue Feb 21 2006 - 04:22:50 MST