From: Matt Gingell (mjg223@is7.nyu.edu)
Date: Fri May 19 2000 - 09:47:46 MDT
Dan Fabulich wrote:
>> The hypothesis: It is possible, at least in principle, to construct a
>> machine which acquires all the skills and knowledge it requires to
>> function in the world via learning. Human provided domain-specific
>> knowledge, even of highly abstract or primitive domains, is
>> unnecessary. The question then is what is the minimal such machine,
>> and how do we go about building it.
>
> No. This is how you get a useful program given millions of years and
> plenty of trials. Alternately, this is how you get a useful program
> when you can get instant feedback as to how good your general
> "learning system" is, so finding the best one is just a matter of
> searching for it. (The latter is another way of saying the former.
> "Instant" is a point of view.)
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.
-matt
This archive was generated by hypermail 2.1.5 : Fri Nov 01 2002 - 15:28:42 MST