As I was thinking about a few presentations I realized that my ideas about more
human-like AI must include something that acts a little like a trial and error
method even when the subject matter is well understood. I think the component
method of AI is necessary. The components would represent concepts (data that
can be used to represent concepts but but also data that can represent things
like reactions to more 'primitive' pre-percepts. (Perception, in most
psychological literature are shaped by some kind of recognition and cognitive
reaction process. For the most part these reactions are surmised to be more
primitive than what we would consider concepts but that is in no way
established and I think that higher concepts have to be part of the perceptual
process although they usually have to very quickly applied in a perceptual
reaction. Higher concepts definitely mediate in recognition and immediate
perceptual reaction, but that mediation might be implemented with habits. When
I see a pedestrian crossing a road I usually know that it is a pedestrian and
not some other kind of object. )
The component method would be interpretable and a great deal of relevant
information can be found by looking at the components. However, concepts can be
called to act on other concepts. This interaction of concepts is not
necessarily based on a family of strongly related concepts. A tow truck is an
automobile but it is a special kind of automobile that can act on other
automobiles. You cannot drive a wrench around but it is often considered very
relevant to automobiles. An idea is not the same thing as a car but ideas made
cars possible and they shaped your car. And your idea of what a car should be
like will shape the choice of your next car.
These component ideas need to be quickly accessible in order to use them to
interpret events (like reading this.) Ideas that are strongly related to a
target subject have to be quickly accessible for stronger AI. But ideas that
are not categorized as being strongly related (in a customary manner) have to
be able to work on other ideas as well.
------------------------------------------
Artificial General Intelligence List: AGI
Permalink:
https://agi.topicbox.com/groups/agi/T1675e049c274c867-M878639a8c91dc3756d9b46cc
Delivery options: https://agi.topicbox.com/groups/agi/subscription
