YKY (Yan King Yin) wrote:
I agree with Ben and Pei etc on this issue. Narrow AI is VERY
different from general AI. It is not at all easy to integrate several
narrow AI applications to a single, functioning system. I have never
heard of something like this being done, even for two computer vision
programs.
IMO what we need most right now is a unified knowledge representation
language, perhaps something like Cyc or better.
There are plenty of formal knowledge representation languages already,
but so what? They can be tuned in various ways, but this isn't going to
get us very far toward AGI...
I have a question: is it possible to represent an episodic story in
Cyc? For example "Little Red Riding Hood". How about in Novamente?
Yes, it is possible to represent a story in Cyc-L or any other rich
logic-based formal language -- in the sense that one can create a
representation of the story in such a language, that will be understood
by humans who read it.
But so what? The meaningfulness of the representation for AGI purposes
is very hard to assess until you have an integrated system that can do
interesting stuff based on the stories it "knows"...
On the face of it, a "representation" of a story in Cyc-L is not
necessarily any more useful for AGI than a representation of the story
in English...
A more interesting question to think about, rather than how to represent
a story in a formal language, is: How would you convince yourself that
your AGI actually understood a story? What kind of question-answers or
behaviors would convince you of this? The "textual entailment
challenge" is of some relevance here, perhaps.
-- Ben
A story like Riding Hood has the following elements:
1. vision / sensory
2. spatio-temporal
3. natural language
4. social (minds of other agents)
5. naive physics
so it is rich enough as a benchmark.
YKY
------------------------------------------------------------------------
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