RE: Pei Wang's NARS

From: Ben Goertzel (
Date: Sun Nov 30 2003 - 20:42:56 MST


I much admire Pei's work on NARS, and I'm glad that you appreciate it as

I encourage everyone to give Pei's work a close look. He is a deep AI
thinker, and his ideas are worth considering on the philosophical and
mathematical levels.

As a side comment, NARS has a fair bit of overlap with my own Novamente AI
system, but there are some significant differences as well, for example
(this is a non-exhaustive list):

-- Novamente uses (a generalized version of) combinatory logic to represent
complex patterns and procedures; NARS has no analogous mechanism
-- Novamente uses evolutionary learning (in the form of a generalized
version of the Pelikan/Goldberg Bayesian Optimization Algorithm) to create
new concepts, patterns and procedures
-- Novamente uses inference to carry out attention allocation, rather than
simple heuristics as in NARS [this aspect of Novamente is not yet
-- For logical inference, Novamente uses quantitative truth value formulas
that are based on probability theory rather than NARS's uncertain-logic
rules (Novamente's logic component is known as Probabilistic Term Logic,
whereas NARS is a non-probabilistic uncertain term logic)

-- Ben Goertzel

> -----Original Message-----
> From: []On Behalf Of James
> Rogers
> Sent: Sunday, November 30, 2003 6:01 PM
> To:
> Subject: Pei Wang's NARS
> I read the entirety of Pei Wang's NARS 3.0 thesis this weekend (all 181
> pages of it); I've glanced over small bits and pieces in the
> past, but never
> taken time to understand it in detail. Ben (Goertzel) has stated
> more than
> once that many of my ideas appear closely aligned with Pei's,
> something it I
> can now state with certainty is more than just apparent, and many of the
> theoretical similarities at that level explain why I've been in general
> agreement with Pei on matters that almost no one else seemed to
> "get" (from
> my perspective) in past discussions on the lists.
> >From a formal theory standpoint, I find NARS to be very, very good, among
> the best I've ever read. Most of my quibbles and questions occur
> for me at
> the implementation and representation level, which is where my general
> strengths are anyway. For example, I am not entirely sold on the
> "urgency"
> model. It seems to me that such things can be dealt with
> implicitly in the
> representation and implementation if the system is non-axiomatic. I would
> argue similar minor points elsewhere, but since implementation and
> representation is only covered obliquely this is not a criticism of the
> underlying theory.
> It is well worth the read (and quicker than the 181 pages implies) if you
> haven't been exposed to this general model of AGI. I would still assert
> that representation (and by extension, implementation) is the
> other half of
> the problem, but NARS is a pretty fine description of the first half even
> with my quibbles.
> Cheers,
> -James Rogers

This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:43 MDT