Since everyone else is doing it . . . .
My interests are
a.. Coordinated Knowledge-Base/World-Model Building and
b.. Friendliness/Morality/Goal Development & Optimization
Fundamentally, this means developing
a.. Knowledge-harvesting agents
b.. Knowledge-collapsing/coordinating agents
c.. Knowledge-retrieval agents (includes planning, will be Friendliness
constrained)
d.. Conversational agents (includes building a model of the other conversant,
will be Friendliness constrained)
My personal belief is that a collection of the above agents with a goal and
access to a knowledge-base/world-model *is* a self-willed mind. Given that I
believe that all of the above could be developed in a fairly brief time-frame
(i.e. during *my* lifetime), I am also extremely interested in (Friendly!)
motivational systems.
For *my* purposes, I am limiting *my* concept of AGI to the concept level
and above. I am not doing any sensory work other than with textual and
electronic data (well, except for some dabbling in two-dimensional image
reading for a specific project). I think that this is a rational partition and
does not violate the General in AGI since I *don't* believe that our senses are
"Intelligent".
I have an evolving knowledge representation scheme that grew out of some
dabbling with a link parser with some of the Novamente folk (since it's my
belief that language fairly closely mirrors the structures that the
brain/intelligence is using and that language and our brains have co-evolved to
a better-than decent solution). I am trying to create a "seed
knowledge-base/world model" starting with an initial *vocabulary* of Simple
English or Basic English, rules for parsing it into my knowledge scheme
(knowledge harvesting), and rules for coordinating/collapsing knowledge
(including tracking sources of knowledge and recognizing *and maintaining*
conflicting knowledge). Translating the knowledge scheme back into English is
fairly trivial (with most of the difficulties being where to terminate the
graph and how to organize the phrases for maximum clarity) but scaling is a
HUGE problem.
One of the PRODUCTS that I'd like to quickly produce is a system that does
conflict resolution for humans (i.e. clearly representing what the disagreement
is down to the level where they agree to disagree). I'd then like to turn that
product to work on my current (pretty much complete) model of Friendliness and
how to turn that into a motivational system (or, maybe, I could just start a
huge site for arguing politics :-).
I am currently trying to get a lot of this into shape where I can ask for
collaborators. Since the core of "my" scheme is a defined *common*
knowledge-representation framework, I believe that it will be possible to
parcel out various tasks (most of which will be agent-development, translators,
and user interfaces). My personal preference is the .NET framework (solely due
to the amount of infrastructure); however, I would love to see interfaces to
work done on any platform. While my description strikes me as sounding very
logic-based, the knowledge-representation is a giant network and many of the
agents I've spec'ed act using schemes more familiar to neural networks
(spreading activation) or enzymes (yes, I am truly going back to my roots :-).
Having seen the internals of Novamente, I am very impressed but they are
currently going down a slightly different path than I would enjoy following
(doing more knowledge creation and discovery than harvesting and collapsing
coordination). I like what I've heard of Richard Loosemoore's ideas and would
like to see more. From this thread, I am intrigued by Stephen Reed and John
Rose since they seem to be on the same path that I'm following.
Oh yeah . . . . I started out as a biochemist and moved into computers to
facilitate upgrading our simulations (and even have a few published papers in
enzyme kinetics in the early 80's). I've worked in the financial arena
(including an Expert System Shell and Expert System Builder for Citicorp in the
early/mid 80's), government contracting (including a Project Manager's Support
System with an AI for scheduling, numerous Expert Systems for procurement, and
a neural network for diagnosing radioactive thallium images of the heart for
Air Force pilots), intelligence (which is why I am now *not* willing to do most
government contracting), international development (which I'm pretty sour on as
well), and a large number of data collection/sharing/analysis systems. I have
a M.S.E. in Artificial Intelligence and bailed out of doing my dissertation on
Machine Learning and Human Decision-Making when my son was born.
Mark
----- Original Message -----
From: Russell Wallace
To: [email protected]
Sent: Monday, March 26, 2007 4:02 PM
Subject: Re: [agi] AGI interests
I don't believe AI in the sense of a self-willed mind is going to happen;
fortunately, it doesn't need to. The two problems I want to help solve are the
global loss of fifty million lives a year, and the difficulty in living in the
99.999...999% of the universe that isn't Earth. Each of these is a problem not
of magnitude but of complexity, so to solve them we need better tools for
handling complexity.
At present we have marvellous tools for handling human-readable information,
but essentially all significant work still needs to be done by humans. Even our
richest, most sophisticated and flexible programs are tiny isolated fragments,
brittle and opaque, animated step by small painful step only by the constant
labor of armies of human workers. We need to do for machine-readable
information/procedural knowledge what we have done for text: create a rich,
fluid environment in which humans need merely say what we want, and the
machines will handle the details of delivering it to us. If we can build tools
that powerful, we can start making real progress on the problems of complexity.
------------------------------------------------------------------------------
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?list_id=303
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?list_id=303
