"the great tinkerer" <gr8tinkerer@hotmail.com> wrote:
> artificial intelligence cannot work until you can program a
> reaction to every action.
Pardon me, but bullshit! All one needs is a reaction scheme -- general
rules, and a fall-back position. Whether one even needs that for
"intelligence" is another question entirely -- do humans have a reaction
for every action? I doubt it...
> intelligence is the ability to react.
In that case, certain forms of bacteria certainly score highly on
intelligence tests...
> intelligence feeds off of conciousness and awareness.
No... both consciousness and awareness are *properties* of intelligence.
> we learn from observing: a book, a person, an email etc.
> can you teach a computer to program itself by what it sees?
Vision, perhaps not yet (any body read anything more about the Ping-Pong
robot?). However, if you're looking for computers which learn from their
past actions (and the actions of of those they encounter) you need look no
farther than the chess programs. I believe there was also a checkers
program about 1950 which "bootstrapped" itself so far that it easily beat
it's programmer...
> id like to see a computer that can install a new hard drive in
> itself when it runs out of disk space, and can fix itself when
> it has a problem.
And I'd like to see a human who could grow a second brain, or even just
regenerate a limb which was missing. My point being humans don't know how
to extend themselves yet either... although we're working on it!
> that computer is the first step towards artificial intelligence,
> the next step of course would be to "teach it to learn."
Again, computers learning from their own encounters has been done many
times[1] -- a defining property of intelligence, yes, but it's just not
quite enough.
Real AI, as Hofstadter says, will involve self reference (consciousness)
and "chunking" up. (*induction*/generalization has proven to be a
difficult concept for AI programs...)
ERiC
[1] In fact, I'm working on such a program myself right now -- an
evolutionary version of my AI engine for my 3-dimensional tic-tac-tuc-toe
program. Even just the static version usually beats me, but I'm not happy
because it's clear that it's offense is not very good; it's defense just
prevents me from ever winning...