Date: Sat, 10 May 1997 07:56:20 +0100
From: Willard McCarty <Willard.McCarty@kcl.ac.uk>
Subject: Deep Blue and human intelligence
This from the Deep Blue FAQ. Note in particular Kasparov's statement,
"Chess gives us a chance to compare brute force with our abilities." Also
the commentator's remark that "Kasparov isn't playing a computer, he's
playing the ghosts of grandmasters past. That Deep Blue can organize such
a storehouse of knowledge -- and apply it on the fly to the ever-changing
complexities on the chessboard -- is what makes this particular heap of
silicon an arrow pointing to the future."
"Does Deep Blue use artificial intelligence?
The short answer is "no." Earlier computer designs that
tried to mimic human thinking weren't very good at it. No
formula exists for intuition. So Deep Blue's designers have
gone "back to the future." Deep Blue relies more on
computational power and a simpler search and evaluation
function.
The long answer is "no." "Artificial Intelligence" is more
successful in science fiction than it is here on earth, and
you don't have to be Isaac Asimov to know why it's hard to
design a machine to mimic a process we don't understand
very well to begin with. How we think is a question without
an answer. Deep Blue could never be a HAL-2000 if it tried.
Nor would it occur to Deep Blue to "try."
"If you go back to HAL in 1968," says Deep Blue
development team member Joe Hoane, "2001 came out and
a lot of people were introduced to the idea that well, you
could have a relationship with a computer. HAL in the movie
had a personality and, in 1968, people started to realize
that computers are getting interesting, that maybe we've
reached another milestone where computers are getting
really interesting... solving really interesting problems that
we couldn't otherwise solve."
Deep Blue's strengths are the strengths of a machine. It has
more chess information to work with than most computers
and all but a few chess masters. It never forgets or gets
distracted. And its orders of magnitude are better at
processing the information at hand than anything yet
devised for the purpose.
"There is no psychology at work" in Deep Blue, says IBM
research scientist Murray Campbell. Nor does Deep Blue
"learn" its opponent as it plays. Instead, it operates much
like a turbocharged "expert system," drawing on vast
resources of stored information (For example, a database
of opening games played by grandmasters over the last
100 years) and then calculating the most appropriate
response to an opponent's move. Deep Blue is stunningly
effective at solving chess problems, but it is less
"intelligent" than the stupidest person. It doesn't think, it
reacts. And that's where Garry Kasparov sees his
advantage. Speaking of an earlier IBM chess computer,
which he defeated in 1989, Kasparov said, "Chess gives us
a chance to compare brute force with our abilities."
Deep Blue applies brute force aplenty, but the "intelligence"
is the old-fashioned kind. Think about the 100 years of
grandmaster games. Kasparov isn't playing a computer,
he's playing the ghosts of grandmasters past. That Deep
Blue can organize such a storehouse of knowledge -- and
apply it on the fly to the ever-changing complexities on the
chessboard -- is what makes this particular heap of silicon
an arrow pointing to the future.
The worlds of science and enterprise are full of problems
with so many variables they can't be solved in real time. A
system like Deep Blue that can accelerate solutions by
powers of 10 is going to make a difference far beyond the
chessboard. (And P.S. - That so much of Deep Blue's
innards are "general-purpose" industry-standard hardware
is good news to any organization faced with a 7-figure
problem on a 6-figure budget.)
The way that the PowerPC chips inside Deep Blue work in
parallel to break down and solve a chess-board problem is
a pretty good analog for the way many scientists, working
independently, advance our total understanding of the
universe, or genetics...
Or the way business people confront the complexities of,
say, running an airline. Figuring THE best way to schedule
570 planes of 25 different types to 150 destinations for
best passenger revenue and most efficient fueling,
maintenance, crew deployment, and turnaround servicing is
a towering problem. On that scale, the difference between
a pretty good solution and the best solution is measured in
billions.
The shifting complexities of the chessboard are the airline
problem in miniature. For computer scientists, chess is a
laboratory benchmark. Back in computing's Jurassic age, in
1950, Claude Shannon, the chief architect of information
theory, put it this way: "The chess-playing problem is
sharply defined, both in the allowed operations and in the
ultimate goal. It is neither so simple as to be trivial, nor too
difficult for satisfactory solution."
Satisfactory solutions - to problems far beyond the
chessboard - are closer than ever before as a result of the
research that has gone into the Deep Blue system. And who
knows? As more possibilities open before us, some of
those science fiction predictions may come true. But it
won't be because of any artificial intelligence. It will be
because systems like Deep Blue helped us make better use
of the real thing.
(Quoted with thanks from <http://www.chess.ibm.com/meet/html/d.3.3a.html>)
Compare Herbert Simon in the online NY Times article, Bruce Weber, "A Mean
Chess-Playing Computer Tears at the Meaning of Thought",
<http://www.nytimes.com/partners/microsites/chess/archive8.html>.
Having designed a program to emulate the human thinking of a grandmaster,
"today he says he did not understand it would be brute force as opposed to
selectivity that would bring a chess computer to an equal footing with men
and women. But that does not diminish the accomplishment of Deep Blue, he
said, which with its powerful amalgam of brute force and selectivity, is
not unlike what humans do, if different in the ratio of its elements...."
John Searle disagrees: '"From a purely mathematical point of view... chess
is a trivial game because there's perfect information about it. For any
given position there's an optimal move; it's solvable. It's not like
football or war. It's a great game for us because our minds can't see the
solution, but the fact that we will build machines that can do it better
than we can is no more important than the fact that we can build pocket
calculators that can add and subtract better than we can."
Paul Saffo (Institute for the Future, Menlo Park, Calif.): "People who fear
machines don't need to lose any sleep just yet.... To me, the match was
interesting as a cultural event. Chess, whereas it's a difficult problem to
solve for computer scientists, is just a constrained formal problem. O.K., a
computer beat a grandmaster, but computers aren't any smarter than they were
the day before. The question I'd ask, now that this Rubicon has been passed,
is What's the new arbitrary measure? Maybe it's a computer that plays go...
or a computer that can fill in an I.R.S. form without getting an audit."
All this would seem to me to imply that in applied computing our efforts are
best directed at the development of digital resources (such as the massive
database of chess moves) rather than clever algorithms, i.e. that the best
situation is one in which trivial operations are repetitively applied to
sophisticated data. This in turn suggests that the future is in good
markup.
Comments?
WM
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Dr. Willard McCarty, Senior Lecturer, King's College London
voice: +44 (0)171 873 2784 fax: +44 (0)171 873 5801
e-mail: Willard.McCarty@kcl.ac.uk
http://www.kcl.ac.uk/kis/schools/hums/ruhc/wlm/