The Human Factor

From: Guy Macon (http://www.guymacon.com)
Date: 09/13/04 

Date: Mon, 13 Sep 2004 03:49:58 -0700

I am posting this page because I think it is interesting.
Please do NOT assume that I agree or disagree just because
I posted it. I have an opinion, but would like to see what
others think before weighing in. BTW, the rest on the
http://members.cox.net/mathmistakes/ site is well worth
looking at.
                                                -Guy Macon.

----------------------------------------------------------

Game Theory 4: The Human Factor
By Paul Cox
http://members.cox.net/mathmistakes/chess.htm

The past few months the debate about Man vs. Machine has
made headlines again as three big exhibitions have been
played pitting man against computer. The first happened last
September as US Chess champion Larry Christiansen was
defeated by a desktop computer program Chessmaster 9000.
Final Score Chessmaster two and a half, Christiansen one and
a half. While completely unofficial, it proves that at the
very least, off the shelf chess programs really can achieve
Grandmaster status.

Then in October, World Champion Vladimir Kramnik faced off
against Deep Fritz promoted as a sequel to Kasparov vs. Deep
Blue in 1997. The results of the competition was a tie, four
games to four games.

This month Kasparov himself is facing a rematch against a
computer called Deep Junior. Details are here. So far game
one was won by Kasparov, and game 2 was played to a draw,
and Deep Junior won game 3. The final three all ended in
draws for a 3-3 tie.

Are all of these signs that computer chess players are
superior to man?

Well, no actually.

Hidden behind all of the man vs. machine debate is a deep
irony that is not really being reported. All four of these
computer programs (Chessmaster and the three Deeps) have
achieved Grandmaster status not by brute force analysis of
all game possibilities, but by simulating human behavior.

You may be wondering why all of these chess programs have
the first name of "Deep". Well, the first computer to be
considered a grand master was called "Deep Thought" after
the fictional super computer in the classic novel The
Hitchhikers Guide to the Galaxy, which according to the
novel calculated the answer to the ultimate question of life
the universe and everything which turned out to be 42.

When the team that created Deep Thought was hired by IBM to
continue their research, they called the project Deep Blue
since IBM is often called "Big Blue". As far as I know
neither Deep Fritz nor Deep Junior have any connection to
either Deep Thought or Deep Blue, but the tradition has been
established.

Deep Thought played Kasparov back in 1989 and was soundly
defeated by the human. Based on the kind of game it played,
it was a true brute force player, looking at all the
possible moves and determining by a formula which is the
likeliest to best strategic move.

Humans don't think like that, they use experience and
instincts, and they learn quickly from their mistakes. A
brute force computer program has no way of realizing it made
a mistake.

So what changed between 1989 and 1997, did the increase in
calculation capacity make Deep Blue that much better than
Deep Thought? Not hardly, in fact the new generation
programs run on much less sophisticated hardware than Deep
Blue did. Deep Blue used 64 processors, calculating
200,000,000 moves per second, while Deep Junior only has 8
processors and calculates a mere 3 million chess moves per
second, yes Deep Junior is every much Deep Blue's equal when
it comes to chess.

The big difference between Thought and Blue has to do with
the software it was running. While I cannot prove this, I
(and many others) believe Deep Blue was programmed to
consider previous games played by Kasparov in the past. In
other words, Kasparov was defeated by a more efficient
simulation of himself!

Kasparov himself was of a similar opinion. "Deep Blue was
more about PR, selling the story, scaring the human race,"
he said to reporters before his latest match, "This machine
[Deep Junior] is a real opponent. It's not hiding behind a
curtain."

Regardless of outcome, the best computer chess players are
merely imitating human players, which says a lot about the
real nature of "intelligence" as compared to number
crunching.

As I stated in part 1 of the game theory series, Chess is a
two-player, non random, perfect knowledge game. As a result,
it should have an ideal strategy that will result in either
a win or a draw every time. So, far this strategy has proven
very elusive, with the computers no closer to finding it
than the humans.

The Human Factor in Simulation Games

In part 2 of game theory, I mentioned I was a fan of The
Sims by Electronic Arts. In The Sims the computer controlled
characters behave almost life like simply by following
simple rules that govern their behavior.

Well in the past month, Electronic Arts released an online
version of The Sims called The Sims Online. In this version
there are practically no computer controlled characters,
every character is controlled by a real person playing over
the Internet. The result is a very different game than the
computer controlled variety.

On the desktop game you controlled everything, Sims were
happy or miserable at your command. With Sims Online, you
only control one Sim, and he or she has to become happy or
miserable by negotiating with other Sims controlled by other
real people. There is a whole economic structure at work,
and since Sims are mirrors of their owners, there are
friendly people, control freaks, snobs, fun loving,
workaholics, etc.

Behavior of the Sims in the desktop version may have had
human like qualities, but the behavior was pretty
predictable. Behavior of online Sims is entirely
unpredictable because humans are really in control.

A few years ago, I was into playing "First Person
Shoot-em-ups" or FPS games as they are popularly known. Doom
and Duke Nukem were the popular games of the day, and I was
as good as anyone in these games. Then I tried playing
online "Death matches", where real players tried to kill
each other. I never fared so well in these games. Sims
Online is simply a less violent version of the same
experience.

To Err is Human...

Human controlled players are smarter than the computer
controlled variety, with one exception: Humans sometimes
make mistakes. Computers can beat humans in two-player, non
random, perfect knowledge games because Humans can be
counted on to make mistakes along the way. Playing checkers
against a computer is not fun, because the best you can hope
for is a draw. Playing checkers against another person is
fun, because humans make mistakes along the way allowing the
possibility of victory.

Analysis by Chess experts of those games won by computers
against grand master humans, is that in every case, the
human player made an error somewhere along the way.

...To Really foul things up requires a Computer

Computers, on the other hand, are only as smart as the
program they are running. The smartest computer players
imitate humans only without the random mistakes. This makes
the computers predictable, and since recognizing patterns is
something we humans do instinctually, computers are often
defeated in games of skill.

This all goes back to the artificial intelligence debate. A
consistent theme in science fiction is the speculation that
one day artificial intelligent personalities may one day be
superior to humans. Considering how radically different
humans and computers are, this scenario is highly unlikely.
That computers are our superior in some tasks, such as rapid
math calculation and maybe Chess, should not be a surprise.

There will always be things humans can do that computers
cannot.

----------------------------------------------------------

Relevant Pages

  • Re: Is C close to the machine?
    ... As I understand it, particularly in the case of go, humans are better at   ... playing the game than they are at programming a machine to play the game. ... biological computational and control systems learn and adapt. ...
    (comp.arch)
  • Re: Just exploring...best bang for the buck title $1k-$1.5k?
    ... I'm a pretty decent player who likes deep rule sets/games. ... but fun & somewhat deep) ... The Crypt, MNF, Rocky & Bullwinkle, and Popeye? ... R&B - great game, colorful, really fun, quick, not too tough, not too ...
    (rec.games.pinball)
  • Re: Bridge-Playing Computers
    ... there are fewer and fewer humans who a computer cannot beat. ... while bridge programs have greatly increased in ... trivial to program a computer to play a perfect game. ...
    (rec.games.bridge)
  • Re: Hydra crushes Pono in 29 moves
    ... Humans can do the same to computers if their understanding is sufficient. ... Some of these machines are online, learning from us. ... >>>There is an unstated assumption here that the only way to win is to play ... >>>the computer at its own game. ...
    (rec.games.chess.analysis)
  • Re: SW Mixed Regionals Questions
    ... forced ourselves to play catch up most of the season. ... were able to come back on top, but this wasn't a very effective game ... deep game- DTL, Barrio, and Party Van all used their deep game ... Is this a growing trend throughout the mixed game? ...
    (rec.sport.disc)