[Hans Blom, 960227b]
(Shannon Williams (960226.17:00 CST))
My goals center around the generation of artificial intelligence.
Unfortunately, I cannot find people, except the people on this list,
who discuss intelligence and behavior in terms that I agree with.
Right in the middle of AI and PCT is Artificial Life. There's a nice
alife discussion list that you are probably aware of. Intelligence
has often been defined as an organism's capability to achieve its
goals. Artificial creatures are given one or more goals and one or
more ways to act, and it is often surprising how even very simple
architectures can lead to very complex (purposive) behavior.
PCT tells us a lot about the basic organization of how to realize
goals. AI tells us a lot about efficient ways of how to process
perceptions into actions. But the problem with most AI is that it is
not grounded in the real world. Intelligence as we know it can, I
think, arise only in an organism that interacts with the outside
world. Therefore an organism must have at least some sensors (to
obtain information about where it is in the world) and some actuators
(to move itself closer to its goal).
I think that we need to design neural networks that control.
Right, whether they be the artificial neural networks that we know
now or some different type of architecture. At the moment, the fields
of robotics and alife have a lot to offer, I think, although the
former emphasizes applications (clumsy but extremely fast industrial
robots) far too much to my liking and the latter is still too much
dominated by the "classical" type of symbolic AI thinking.
And when we do this, I do not think that we will need hierarchies.
Maybe not, but several theoretical arguments show that we _will_ need
"functional clusters".
(We need a method of resolving conflicts, but we do not need the
hierarchy for that.)
Resolving a conflict = computing the optimal course of action where
more than one possibility exists and where these possibilities may
even be almost equally good. This certainly calls for intelligence.
Just in designing the neural networks to control and to learn to
control, I think that we will go very far in modeling intelligent/
adaptive behavior.
Fully agreed, especially the emphasis on "control" (hardwired) and
"learn to control" (adaptation, 'softwired').
Greetings,
Hans
ยทยทยท
================================================================
Eindhoven University of Technology Eindhoven, the Netherlands
Dept. of Electrical Engineering Medical Engineering Group
email: j.a.blom@ele.tue.nl
Great man achieves harmony by maintaining differences; small man
achieves harmony by maintaining the commonality. Confucius