RE: Review of Novamente & a2i2

From: Ben Goertzel (
Date: Sun May 05 2002 - 20:29:21 MDT

> > Eliezer, your DGI paper absolutely does NOT give a design for an AGI.
> I know that. DGI is the name of the theory. "Levels of Organization in
> General Intelligence" is the paper. Sort of like the distinction between
> the Novamente design and the Novamente manuscript.
> -- -- -- -- --
> Eliezer S. Yudkowsky
> Research Fellow, Singularity Institute for Artificial Intelligence

Ah, okay.

Well I know nothing about your DGI theory, except what is in that paper, so
I can't comment intelligently on any aspects of DGI except the very abstract
theoretical aspects described in that paper. What I know of your work is:
the DGI paper, your CFAI and Singularity papers, and Flare. None of that
comes anywhere close to a concrete AI design. I look forward very much to
reading the more detailed stuff when it becomes available.

In my view, the very general (and very sensible) ideas in your DGI paper are
consistent with a vast variety of AI designs, including Webmind, Novamente,
Peter Voss's design (insofar as I understand Peter's design), Shane Legg's
design (which is fairly similar to Peter's; and Shane is now working with
Peter), and Pei Wang's NARS design.

You have stated that you don't think my, Peter's or Pei's design fulfill the
general design criteria you've given in your DGI paper; but I think you're
wrong on this.

I want to make one thing clear: I don't aim to convince you or Peter or
anyone else with a strong intuition different from mine that *my* AI design
is better than yours/theirs. I think that the amount of hard knowledge in
this area is sufficiently small that it's to be *expected* for different
smart, well-informed people to have radically different approaches.

My aim in building a Novamente team is to gather a group of smart and
relevantly skilled people who share my own intuition about how to build an
AGI *to a reasonable extent*. Perfect agreement is not necessary; but
obviously you and I, for example, disagree too profoundly to work together
on the same system!

-- Ben G

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:38 MDT