Matt Gingell misunderstood:
> A child touches a hot stove once or twice before figuring out it's a
> bad idea. He doesn't stand in front of it for years, trying out
> thousands of random strategies, standing on his head, talking to it,
> holding his breath, till he finds one that doesn't hurt. Learning
> isn't blind search - you're misunderstanding my position if you think
> that's what I'm advocating. Learning both aquires and utilise
> knowledge and representation, improvement is deliberate and heuristic.
> Sure you need a million years if you want to evolve something from
> scratch by random variation with some simple operational fitness
> function, and you'll be running till the sun goes out doing an
> exhaustive analysis of the space of all possible programs, but those
> two possibilities don't exhaust the space of workable learning
> algorithms.
You're right, they don't spend years figuring out stoves. That's
because they already have lots and lots and LOTS of information about
the world around them before they ever get out of the womb. This
information is stored in the genes and is visible as what we call
"instinctual" behavior. THIS information took millions and millions
of years to develop, and without it, our children WOULD be spending
years (actually, years is an optimistic estimate, IMO) figuring out a
hot stove.
Experiments with babies show that babies as early as a few months are
surprised to see objects vanish into thin air, come to a full stop
without being in contact with another object, and begin moving without
a force to propel them. Babies have a very compicated idea of what
elements in their visual field constitute objects and how those
objects will generally behave.
This information is not learned from a blank slate at birth. (There
was no time.) We are born with it. Unless we program it into our AI,
we should expect the AI to require evolutionary time in order to
figure these things out. That's time we don't have.
-Dan
-unless you love someone-
-nothing else makes any sense-
e.e. cummings
This archive was generated by hypermail 2b29 : Thu Jul 27 2000 - 14:11:25 MDT