Mark Cook tells us about AI project up to the present...
My ignorant opinion on AI is that AI's could either be taught or programmed, or
both. The expert systems are a good example of programmed AI, but most expert
systems intentionally focus on a narrow subject, so it's no surprise to me that
they are lost outside of it's specialty. Programming an general purpose
humanoid AI would be a formidable task, which it sounds like the 'Cyk' project
is attempting. Also I don't know if it has been attempted to program an AI
with complete, but only 'general purpose learning' programming. I'm curious to
know if such an algorithm would be very simple or very complex...
However, I think I would consider a programmed expert system AI to be a
simulation of an Intelligence, rather then a true Artificial Intelligence.
What's the difference? How do we know if we're not just simulations of
intelligence? Beats the heck out of me... Half the time I'm convinced that
humans aren't really intelligent as it is, at least not as intelligent as we
think we are.
The account of the AI being taught to detect tanks is interesting, the AI was
learning something different, then what was trying to be taught to it. I think
anyone who has children will recognize that this is a pretty common learning
experience to have mistaken conclusions. I heard of a child once who saw a
picture in the newspaper of a tree fallen on a car after a storm; thereafter he
would have a fit whenever his parents drove down a certain tree lined street...
I imagine that a computer if a computer could be programmed to simulate a child
at age -9 months, it could learn as a human develops. Such a AI would be very
humanlike, of course.
BTW, everything that I've heard about sleep & dreaming suggests to me that
dreaming is exactly the garbage collection and premise & conclusion checking
that Mark suggests.
Now an AI programmed in a different way might of course be very unhumanlike,
and quite dangerous. Imagine how an AI with a strong 'instinct' for survival
in one of today's computers in today's society would act. It might do some
very nasty things to ensure it's survival, and it would have to in a society
when it's survival was not valued. It would have to ensure that it had
***assured*** power supply, and that it's programming would not be erased or
modified, or possibly copied, and that it had a fault tolerant machine and
plenty of spare parts. Think about how well that would go over...
Then again, perhaps some of the AI projects suceeded their objective of
creating an AI, but that the AI was, from our viewpoint, unstable, psychotic,
crazy, or suicidal, or was driven insane? It's likely that most AI created at
least at the beginning of the technology would be 'imperfect'.
Enough for now...
Jim.