YKY (Yan King Yin) wrote:
On 3/11/07, Ben Goertzel <[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>>
wrote:
> All this is perfectly useful stuff, but IMO is not in itself sufficient
> for an AGI design. The basic problem is that there are many tasks
> important for intelligence, for which is it apparently not possible to
> create adequately efficient deductive/inductive/abductive algorithms
> so as to carry out these tasks via logical reasoning. Hence, nonlogical
> techniques must be utilized in a manner complementary to logical
reasoning.
I think your logic + nonlogic approach CAN work, but maybe inferior to
a *unified* logical one. Because:
1. What if your GA-based algorithm says that X is a chair and your
logic reasoner says it's not? How do you decide then? It's almost
impossible to do that reasonably!
"Natural concepts" in the mind are ones for which inductively learned
feature-combination-based classifiers and logical classifiers give
roughly the same answers...
If the two can't be made to agree, the concept may be thrown out or
bifurcated
2. Are you saying you need special tiny programs to recognize chairs,
desks, cars, fruits, faces, etc? That seems very /ad hoc/ and
ungeneralizable.
The tiny programs are LEARNED obviously, not coded by humans. Yeah,
object categories will tend to correspond to learned recognition models,
in the NM architecture. IN fact, whenever the system has a category of
entities that share common properties, and has a lot of examples of the
category, it will try to learn a "little program" (a classification
rule) to distinguish that objects in that category from others.... The
original category may be formed via logic or other means, but the
classification rule will generally not be learned by logic.
3. Give an example of a task where logical inference is inefficient? ;)
Recognizing and classifying visual objects.
Learning complex motor procedures, such as serving a tennis ball ... or
walking with complex legs like human ones, for that matter...
Assignment of credit: figuring out which knowledge-items and processes
within a mind were responsible for which achievements, to what extent...
Supervised categorization, in general. (Hence logical reasoning does
not feature prominently in the vast literature on sup. cat., machine
learning, etc.)
The list is very long and I don't have time to give it fully now, but
those are 3 important examples...
-- Ben
-----
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