It seems to me that the great frontier for AGI can now be clearly identified
in part as real world reasoning/ real world intelligence - as distinct from
artificial, ideal world reasoning.
Robots are currently incapable of real world travel. Computers - or,
better, robotic minds/intelligences- are incapable of real world reasoning
And as I start to research this I see a great deal of AI opinions agreeing
on this frontier.
Here for example
http://cs.nyu.edu/faculty/davise/papers/singularity.pdf
The Singularity and the State of the Art in Arti cial
Intelligence
Ernest Davis
Dept. of Computer Science
New York University
New York, NY 10012
[email protected]
May 9, 2012
Abstract
The state of the art in automating basic cognitive tasks, including vision
and natural lan-
guage understanding, is far below human abilities. Real-world reasoning,
which is an unavoid-
able part of many advanced forms of computer vision and natural language
understanding, is
particularly dicult. This suggests that the advent of computers with
superhuman general
intelligence is not imminent.
Or you could take:
Foundations of
Real-World
Intelligence,
edited by
Yoshinori Uesaka et al
Computers have become increasingly powerful and far
exceed human ability to solve well-defined problems such as numerical
computation, document processing, and logical inference in ideal worlds
where all information is known and there is an algorithm for the solution
that may be stated precisely in a programming language. Nevertheless,
computers are still far inferior to humans in many other intellectual
tasks such as pattern recognition, dealing with ill-posed problems, and
learning. If computers are to develop to the point where they can perform
humanlike, flexible processing in the real world-thus opening up
a new horizon in information-processing technology-we must pursue
the fundamentals of humanlike information-processing, come to terms
with how humans process information at an intuitive level, and embody
this theoretical knowledge in various developing hardware technologies.
Or:
AI Depends on Your Point of View
Noah Shachtman 07.29.03
Even the dumbest people can look at a situation from several different
angles. But that's still a problem for the smartest of computer systems.
The Real-World Reasoning project, a Defense Advanced Research Projects
Agency program, is designed to get computers to start examining situations
in more than one way. It's part of a larger effort, spearheaded by the
Agency's Information Processing Technology Office, or IPTO, to move toward
machines that can think for themselves.
Ben's latest book is:
Real-World Reasoning: Toward Scalable, Uncertain Spatiotemporal, Contextual
and Causal Inference Atlantis Thinking Machines: Amazon.co.uk: Ben Goertzel,
Typically, of course, AI-ers are trying to adapt./ rescue logic to deal with
real world reasoning. Thus a Ben problem asks:
"were Sue and Jane both at the clinic last Tues. at 4.00 pm?"
an obviously logically adaptable (and rather artificial) problem.
A real world and more realistic version of that would ask:
"Sue and Jane were both at the clinic that time - but did they see each
other/bump into each other?"
Humans can solve that, as they solve other forms of real world reasoning,
but not by logic - by going through their imaginative experience of the real
world and living in it - in this case of clinics and being in them. Such
real "spatiotemporal", " contextual" reasoning is not logical. You might
have here for instance to think about the layout of the clinic, and which
doctors Sue and Jane were visiting.
What I see is that "real-world reasoning" is still receiving a rather
confused treatment - people aren't really aware yet of its extent.
Some may remember how I talked here a while back about the 90-odd per cent
of human reasoning that AI wasn't really touching at all and AGI-ers were v.
determinedly not thinking about - scientific, technological, business,
economics, art, history, letters, newspapers etc.. not to mention common
sense navigation and manipulation of the world (a la Woz Test robot).
That neglected 90-odd percent is real world reasoning.
Of course, a great many computer programs present models/simulations of the
real world - economic forecasts, weather forecasts, evolutionary games etc.
But this is not true real world reasoning - introduce, or suggest
introducing, one extra factor into their models, and the programs have no
capacity to deal with it. They are Chinese room modellers with no ability to
look around the world their model describes, and place and identify
other/new factors. No capacity for "spatiotemporal" or "contextual"
observation and thought.
Comments?
-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393
Modify Your Subscription:
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968
Powered by Listbox: http://www.listbox.com
