RE: computer-go: Pattern matching

["Xu, Mousheng" <moushengxu@xxxxxxxxxxxxxxxx>]

The critical difference: hard or impossible?

Give me an example of knowledge that cannot be represented on a machine
-- not impossible;
"common sence" is very very very very very very hard to represent and
maybe even impossible -- hard and maybe practically impossible.

I vote for the latter. Given all the facts needed, a man can predict
exactly how an event would happen, as a statistician may declare. But in
the real world, the assumption "given all the facts needed" could be
impossible. Think about stock prices, then think about Go.

-- Mousheng

 -----Original Message-----
From: Elmer Elevator [mailto:bobmer@xxxxxxxxxxxxxxxxx]
Sent: Wednesday, November 10, 1999 9:00 AM
To: computer-go@xxxxxxxxxxxxxxxxx
Subject: Re: computer-go: Pattern matching


Okay, first I was thinking of writing literature -- plays that deeply
touch many
people, like Shakespeare's or Durrenmatt's -- or plays that make
peoples'
intellects and senses of humor tingle, like Alfred Jarry's "Ubu Roi" or
the
plays of Joe Orton.

I wondered what the objections to that would be. "That's not knowledge."
But of
course it is -- all great authors serve a lengthy apprenticeship,
acquiring a
great deal of relevant knowledge. (Shakespeare was an actor in other
peoples'
plays; he saw which plays drew big audiences and which turned audiences
off, and
tried to figure out why.) "That's subjective -- I don't like
Durrenmatt's plays
at all."

Well ... let me know what the other objections are. I don't think
they'll be
proper objections; I think they'll just be loopholes to weasel out of
the
original broad premise of the question. But I don't think there's going
to be a
Deep Blue of playwrighting in our lifetimes or ever.

There are, of course, machines that generate music that people actually
pay
money for and dance to -- Tekno -- but I would rather ask about the
Knowledge
that Mozart and Bach and Chopin or Charles Ives used to write their
music. Is
anyone going to teach this to a machine, and are we going to get
original  piano
sonatas of equivalent beauty and emotional involvement anytime soon?

But here's a good example of machine-resistant Knowledge. Reporters on
daily
newspapers often have to find someone very quickly -- say, within four
hours
before deadline. The person is not very clearly identified like "Jane
Smith,"
but rather something like "the ex-wife" or "one of the ex-wives" (but
not just
any of the ex-wives -- the one who had his red-headed son) of Richard
Jones. Or
"someone who rides a motorcycle who worked in a real estate office with
Laurie
Johnson in 1997." Or "a past assault victim of a man who was just
arrested for
another assault." Or "There's a guy in this county who's been struck by
lightning three times" (but you can't find an old article about him in
the
newspaper morgue; the dumb librarian probably filed the old articles
under his
name rather than under "lightning").

I won't make the computer actually get into a car and drive around the
city;
we'll restrict the hunt to the office and telephone (a very realistic
frequent
restriction).

It's certainly quantifiable and verifiable because either you find this
person
in time to interview him/her for your story (win) or you don't (lose).
And to be
fair, we won't base it on just one competitive deadline search, but
rather on a
year's worth of deadline searches. Like backgammon, there's an element
of luck,
so no expert wins every time (but backgammon is a solved computer
knowledge
problem).

All I can tell you about this kind of knowledge is that you get better
at it
year after year, that it's a skill reporters get very good at and take
great
pride in. (Police cheat -- they can use intimidation and threaten to
arrest
people to get answers.) Maybe we're arrogant, but if you posted this
question to
a bunch of reporters and claimed a computer could learn to do it as well
(same
percentage of success) as a seasoned reporter, you'd get laughed at very
vigorously.

Bob Merkin

> Patricia Hughes and David Elsdon wrote:
> >
> > Give me an example of knowledge that cannot be represented on a >
machine.
>
> In a course on expert-systems I was told, and I now believe it,
> that "common sence" is very very very very very very hard to represent
> and maybe even impossible.
> But that is because you don't know what you know so maybe I am
> cheating when I use common sence as an example.
>
> Greetz, Pieter