Hi Eric,
I haven't read "What is Thought?" yet, but I look forward to doing so.
I have enjoyed your Hayek papers and also some of your other writings on
GA's and more philosophical topics.
> As I discuss at some length, this is very different than hand coding in
> knowledge, e.g. building a big expert system. Complexity theory
> has told us that finding compact, Occam representations is an
> NP-hard problem. It is too hard to be solved by human programmers
> anymore than a human can, by inspection, solve a huge Travelling
> Salesman Problem.
I agree that finding compact representations of phenomena and processes is a
key aspect of general intelligence.
However, I think the "NP hard" comment is a bit of a red herring, isn't it?
What matters is finding ways to solve the problem in the average case, not
the worst case; and furthermore, not in the average case over all possible
inputs, but rather in the average case over inputs drawn from a probability
distribution representing "reality as we commonly it."\
I think that finding compact representations ("patterns" as I call them,
defining a pattern as a "representation as something simpler", compactness
being a handy measure of "simplicity") *in general* is an unsolvably hard
problem. Juergen Schmidhuber's OOPS system tries to solve this problem but
so far as I know it can't be scaled beyond problems of a very small size.
So I think that real "generally intelligent systems" work by combining
-- a general "compact representation finder" that deals only with very small
problems (sometimes produced via abstraction from problems that were larger)
-- a bunch of specialized "compact representation finders" that deal with
common phenomena of habitual interest to the system
The orchestration of this combination then places requirements on mind
architecture.
Anyhow this is (part of) the philosophy underlying our Novamente AI
architecture.
> AI programs typically do not "understand" because
> they do not exploit Occam's razor in the way natural intelligences
> do.
I agree, but you have to emphasize the role of knowledge here. A mind seeks
a compact representation for a phenomenon, but it seeks a representation
that is compact *relative to the knowledge it currently has in its memory
store*. This is different than seeking a representation that is compact "a
priori."
> But achieving anything like general intelligence will, in my view,
> require a massive evolutionary effort.
I disagree there. I think that simulating evolution is ONE POSSIBLE
APPROACH to creating AGI, but I don't see why you think it's the only
possible approach.
Anyway, I'm sure I'll have a lot more detailed comments once I read your
book ;-)
-- Ben Goertzel
-------
To unsubscribe, change your address, or temporarily deactivate your subscription,
please go to http://v2.listbox.com/member/[EMAIL PROTECTED]
