----- Original Message ----
From: Richard Loosemore <[EMAIL PROTECTED]>
To: [email protected]
Sent: Thursday, April 10, 2008 11:20:13 AM
Subject: Re: [agi] How Bodies of Knowledge Grow
... I agree that Albus is interesting. I am superficially familiar
with his approach.
From my point of view I see only one problem with his attack on the
task of building a complete AGI: he has done an "early binding" of his
architecture to a specific implementation, leaving himself no empirical
wiggle-room.
I say this, of course, because of my commitment to solving the complex
systems problem. Where he commits himself to a calculus of thought, I
see the need to construct a calculus-of-thought *generator* that defines
a space in which one choice of "calculus of thought" is a single point
in that space.
One way to sum up my ("theoretical psychology") approach is to say that
what I am doing is trying to build generators that produce spaces that
contain the human cognitive system and its neighborhood. Then, the
method is something like Monte Carlo on steroids.
Are you planning to implement a big chunk of Albus' approach in Texai,
or do you just mean that you are adopting just the perception ideas?
I'm using a hierarchical control structure in general, and specifically the one
described by Albus, as a guiding principle for the Texai architecture. It's
quite interesting - the notion of implementing an English dialog system in a
robotics-style software control framework. The framework itself is complex,
but cognitive abilities fit easily into it I believe. For the discourse node
in the hierarchy:
Albus perception = Texai utterance comprehension,
Albus behavior = Texai utterance generation,
Albus world model = Texai lexicon, grammar rules, commonsense knowledge,
Albus value judgment = Walter Kintsch spreading activation method for semantic
disambiguationSuperior to instances of the discourse node in the hierarchical
control structure, will be nodes for:
lexical knowledge acquisitiongrammar rule acquisitionBoth of these will engage
the user in English dialog to fulfill their respective missions. James Albus
did not elaborate as to how his architecture is to be implemented, beyond the
NIST Realtime Control System that was created for numerical control and
robotics applications. The implementation of RCS is mostly finite state
machines to govern behavior. Albus has a lot to say about perception in
robots, especially autonomous vehicles. I'm hoping not to have to interface
actuators and sensors myself, but rather in the future plugging their control
software underneath the framework that I am writing.
Although Albus says a given node (think agent) operates mostly in a single
chain of command, I think that a node should be a participant in multiple
control hierarchies in addition to its primary chain of command. This idea is
adopted from human organizations. For example, a truck driver is within the
Dispatch and Logistics chain of command, but is still subject to the policies
and reporting requirements from Human Resources. All control hierarchies in
Texai will be ultimately subject to a single governance.
Richard, I hope to understand more about your idea to "build generators that
produce spaces that contain the human cognitive system and its neighborhood"
as you provide the details, arising from some use cases, that you will need to
drive your initial implementation.
-Steve
Stephen L. Reed
Artificial Intelligence Researcher
http://texai.org/blog
http://texai.org
3008 Oak Crest Ave.
Austin, Texas, USA 78704
512.791.7860
__________________________________________________
Do You Yahoo!?
Tired of spam? Yahoo! Mail has the best spam protection around
http://mail.yahoo.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=98558129-0bdb63
Powered by Listbox: http://www.listbox.com
