Artificial Intelligence
Fernando Villegas Darroui
Hi all:
I intend to develop this issue in more detailed shape in Computer Chess
Reports, but as first step I would like to know your opinion about this
statement: what DB did h as nothing to do wioth some eventual step ahead
of artificial intelligence, BUT ON THE CONTRARY, a proof that the
entire concept of such a thing has not sense because once you surpass
certain threshold of quantity you get, as Hegel said, a change of
quality. It has no sense to duplicate or simulate human thinking IF non
human thinking or analysis show to be equally or more efficiente to
certain tasks. To try such a thing as AI is like to try to develop a
vehicle that simulate exactly the mechanics of human motion, walking and
running.
Komputer Korner
Yes, you are right. I want to apologize to you about what I said
about your prediction in the match. However how could I ever
envisage in a million years how bad a match strategy, Kasparov
would adopt?. He was so scared that he didn't play his normal game
and it cost him hugely.
--
Komputer Korner
The inkompetent komputer.
chrisw
--
http://www.demon.co.uk/oxford-soft
Fernando Villegas Darroui <fern...@reuna.cl> wrote in article
<33792E...@reuna.cl>...
> Hi all:
> I intend to develop this issue in more detailed shape in Computer Chess
> Reports, but as first step I would like to know your opinion about this
> statement: what DB did h as nothing to do wioth some eventual step ahead
> of artificial intelligence, BUT ON THE CONTRARY, a proof that the
> entire concept of such a thing has not sense because once you surpass
> certain threshold of quantity you get, as Hegel said, a change of
> quality.
A statement of Lenin's always amuses me at this point.
It was made during the time of the provisional government, after a leftist
(but non-communist) deputy told Lenin that he was concerned with the
general standard of his followers (generally felt to be a low-life rabble,
at the time). The comment concerned the 'quality' oif Lenin's supporters.
Lenin replied: 'ah, yes, but quantity has a quality all of its own'.
Whilst Lenin was right, in the sense that he gained power, history has
quite a lot to say about the quality of this quantity-quality, no ? Lesson
for speed/materialists here ?
Chris Whittington
Jesper Antonsson
In article <337930...@netcom.ca>, kor...@netcom.ca wrote:
>Fernando Villegas Darroui wrote:
>>
>> Hi all:
>> I intend to develop this issue in more detailed shape in Computer Chess
>> Reports, but as first step I would like to know your opinion about this
>> statement: what DB did h as nothing to do wioth some eventual step ahead
>> of artificial intelligence, BUT ON THE CONTRARY, a proof that the
>> entire concept of such a thing has not sense because once you surpass
>> certain threshold of quantity you get, as Hegel said, a change of
>> human thinking or analysis show to be equally or more efficiente to
>> certain tasks. To try such a thing as AI is like to try to develop a
>> vehicle that simulate exactly the mechanics of human motion, walking and
>> running.
>
>about your prediction in the match. However how could I ever
>envisage in a million years how bad a match strategy, Kasparov
>would adopt?. He was so scared that he didn't play his normal game
>and it cost him hugely.
Yes, you have said that about a thousand times now. Repeating it
doesn't make it true though. Kasparov would have lost even worse
if he played his normal game. *But*, I'm not going to debate this,
since neither one of us can do nothing but guess. We'll just have
to wait and see until DB plays more games.
Jesper
Kirt Undercoffer
From an AI perspective, the statement makes little sense.
Search was one of the first techniques explored in the AI enterprise.
The problem is that given an enumerable set of state spaces that can
be explored to accomplish a goal, various search strategies can be employed
to satisfy the goal. But intensive search is pretty clearly *NOT*
how people actually think although people do in fact employ search
in their thinking (but the search is shallow). McCarthy and to some
extent Minsky have articulated a school of thought which does *NOT*
concern itself with how humans think but are more concerned with
developing effective techniques for the solution of tasks that people
would generally say require intelligence for their solution. Chess
and checkers and all games of strategy are historically held to be such
tasks.
It was also clear from studying the problem of creating
intelligent systems that search can only take the problem
so far because in most interesting cases the problem space
explodes (increases exponentially or worse). Secondly, search
alone can only be applied in well defined situations where you
have an identifiable goal state, end state(s), and operators
of some kind to facilitate the search. On the other hand,
search is appealing because it can be described simply.
Most problems are not well defined enough to be accomplishable
using search alone. Search is too low level an operation to
actually deal with most serious problems although search is
a fundamental technique found in some form in virtually all
AI systems.
Tan has repeatedly stated that Deep Blue itself has nothing to
do with AI but doesn't elaborate on what he means. He then has
said that they are embodying knowledge of some kind in the
system but doesn't elaborate on what he means by that exactly
but an AI person would assume that he meant that they are using some
form of knowledge representation and inferencing to augment the
parallel alpha-beta search that they have been using. But IBM
is historically anti-AI because of fears that their customers
will over-react with some kind of negative emotional reaction.
The Deep Blue team has conclusively demonstrated that search +
chess knowledge can produce top level chess play. Even if the
system produces moves that a human at top play wouldn't make, that
doesn't invalidate the entire play (in fact, a member of the audience
in the fifth game raised the idea that although some moves were considered
"bad" they might well not be refutable in the sense that the opponent
couldn't capitalize on the move).
Irrespective of the original intentions of the team or the current
intentions, or IBM's PR, behavior has been produced which an observer
would generally call intelligent.
Kirt Undercoffer
kund...@osf1.gmu.edu
mclane
Fernando Villegas Darroui <fern...@reuna.cl> wrote:
> a proof that the
>entire concept of such a thing has not sense because once you surpass
>certain threshold of quantity you get, as Hegel said, a change of
>quality.
Ooops ! This was by Hegel ?! Good that I stopped reading this boring
and stupid definition-philosopher. Good that Marx turned Hegel 180
degrees.
I thought this sentence was made by Frans Morsch. Hm . He once
repeated it den Haag. Did not know that Frans knew about Hegel, I
spoke with Jack London, he replied with Hegel. Heavens.
>It has no sense to duplicate or simulate human thinking IF non
>human thinking or analysis show to be equally or more efficiente to
>certain tasks. To try such a thing as AI is like to try to develop a
>vehicle that simulate exactly the mechanics of human motion, walking and
>running.
But we had efforts:
Botinniks Pionieer or Chess Sapiens
Thomas Nitsche's commercial effort Mephisto III (3 NPS)
others ...
Of course all these guys stopped one day, when the nasty brute-forcers
always killed them, overtook them, laughed about their progress.
Maybe we need new research ?
Kirt Undercoffer
Please see Drew McDermott's piece on this issue in today's
New York Times or in an unabridged version on comp.ai.
Kirt Undercoffer
kund...@osf1.gmu.edu
Fernando Villegas Darroui (fern...@reuna.cl) wrote:
: Hi all:
: I intend to develop this issue in more detailed shape in Computer Chess
: Reports, but as first step I would like to know your opinion about this
: statement: what DB did h as nothing to do wioth some eventual step ahead
: of artificial intelligence, BUT ON THE CONTRARY, a proof that the
: entire concept of such a thing has not sense because once you surpass
: certain threshold of quantity you get, as Hegel said, a change of
: quality. It has no sense to duplicate or simulate human thinking IF non
Rickard Westman
In article <33792E...@reuna.cl>, Fernando Villegas Darroui wrote:
>Hi all:
>I intend to develop this issue in more detailed shape in Computer Chess
>Reports, but as first step I would like to know your opinion about this
>statement: what DB did h as nothing to do wioth some eventual step ahead
>of artificial intelligence, BUT ON THE CONTRARY, a proof that the
>entire concept of such a thing has not sense because once you surpass
>certain threshold of quantity you get, as Hegel said, a change of
>quality. It has no sense to duplicate or simulate human thinking IF non
>human thinking or analysis show to be equally or more efficiente to
>certain tasks.
I believe the operative phrase here is "for certain tasks." Chess
is a very simple problem domain compared to the domains where AI is
yet to be successful. The outcome of Kasparov vs. Deep Blue
doesn't really say much about AI in general. AI is primarily
concerned with domains where humans are vastly superior to humans
(or at least it should be.) That hasn't been true for computer
chess for a long time, where average human players have been
regularly beaten by computers for several decades.
>To try such a thing as AI is like to try to develop a vehicle that
>simulate exactly the mechanics of human motion, walking and
>running.
It may make sense in the areas where computers are currently vastly
inferior to humans. It may even make sense for chess, even if it's
not necessary for beating Kasparov. If he can be competetive with
a machine calculating 200 million nodes per second, even though he
can only calculate a miniscule fraction of that, the "Kasparov
algorithm" must be superior to the "Deep Blue" algorithm. If the
physical symbol hypothesis of AI is correct, it should be possible
to implement it in silicon. *That* would further AI in general,
whereas tweaking search heuristics and implementing chess
algorithms in hardware would only further the narrow field of
computer chess and little less.
To make computer chess advances more applicable to other domains,
as well as providing benefit to people who are already beaten by
their chess programs, I would like to see a change of goals from
"beating humans (including the WC) in match play as thoroughly as
possible".
Some challenges which I don't think would be met by any existing
program:
1. Playing well in arbitrary positions, including closed ones, and
those requiring sophisticated positional play. That is, a chess
program today may rely on a tuned opening book which steers the
game to positions it can handle. The program which met this
challenge would perform as well as a strong human player in
arbitrary positions. The same positions could be given to
computers and humans in a way similar to duplicate bridge, and
then evaluated. The results of such a competition would be an
objective measurement of "playing well in arbitrary positions."
2. Passing the Turing Test of Chess (I just invented it, so if I'm
first, it should perhaps be called the Westman Test :-)). The
idea is similar: A program passes the test if chess experts
can't tell which games were played by an IM and the ones which
were played by a computer.
3. Get a good rating in those "rate my game" puzzles, where you
try to predict the moves of GMs and get points for correct
predictions. This is similar to 2, but probably much easier
(depending on what you mean by "good rating", of course - I'm
not familiar with the performance of current programs.)
4. Playing as well as an average club player without using any
programmer-supplied heuristics. That is, give the program the
rules of chess and let it derive its own heuristics by studying
GM games and playing games against serious opponents. I don't
think programs will be ready to gain knowledge from chess books
any time soon, but it should be allowed.
5. Analyzing the game of an average club player, and give feedback by
providing general advise that actually helps him/her, not just
principal variations.
--
Rickard Westman <wes...@snmp.com>
"Beware of the panacea peddlers: Just because you
wind up naked doesn't make you an emperor."
- Michael A Padlipsky
graham_douglass
In article <5ldmjb$p...@portal.gmu.edu>, kund...@osf1.gmu.edu says...
>
>Please see Drew McDermott's piece on this issue in today's
>New York Times or in an unabridged version on comp.ai.
What did Drew McDermott say? Can you post it here, please?
By the way, I feel that in Fernando's original post, he overlooked the point
that human chess playing skill has much to do with quantity, and not with
"quality".
Monkeys can't play chess, so some level of "quality" intelligence is required.
But even if your brain has greater capacity for knowledge than mine, I will
still beat you if I have quantity on my side - ie I have studied more openings,
I have encountered more types of position in previous play, etc.
I think that in intelligence, quality without quantity is nothing.
Maybe DB v GK demonstrates that in intelligence, quantity is more important than
quality.
>
>Kirt Undercoffer
>kund...@osf1.gmu.edu
>
>
>Fernando Villegas Darroui (fern...@reuna.cl) wrote:
>: Hi all:
>: I intend to develop this issue in more detailed shape in Computer Chess
>: Reports, but as first step I would like to know your opinion about this
>: statement: what DB did h as nothing to do wioth some eventual step ahead
>: of artificial intelligence, BUT ON THE CONTRARY, a proof that the
>: entire concept of such a thing has not sense because once you surpass
>: certain threshold of quantity you get, as Hegel said, a change of
>: quality. It has no sense to duplicate or simulate human thinking IF non
>: human thinking or analysis show to be equally or more efficiente to
>: certain tasks. To try such a thing as AI is like to try to develop a
Henri H. Arsenault
> I intend to develop this issue in more detailed shape in Computer Chess
> Reports, but as first step I would like to know your opinion about this
> statement: what DB did h as nothing to do wioth some eventual step ahead
> of artificial intelligence, BUT ON THE CONTRARY, a proof that the
> entire concept of such a thing has not sense because once you surpass
> certain threshold of quantity you get, as Hegel said, a change of
> quality. It has no sense to duplicate or simulate human thinking IF non
> human thinking or analysis show to be equally or more efficiente to
> certain tasks. To try such a thing as AI is like to try to develop a
> vehicle that simulate exactly the mechanics of human motion, walking and
> running.
It has been know for many years that in theory chess is a "solvable" game,
and is not a game at all in the sense of Von Neumann's theory of games.
Chess does require some intelligence because in practice it is impossible
(so far) to calculate all the possible moves from a starting position.
World Champion Capablanca thought that chess was worked out during the
twenties, and he proposed adding two more pieces (elephants) and expanding
the board to make the game more complicated. Just like in Tic-tac-toe, it
is inevitable that some day a computer will solve the game or at least beat
any human. I am not sure that we have reached this point.
The bottom line is that we still do not have intelligent computers.(I avoid
the question of whether we have intelligent PEOPLE...).
In case you are wondering, the area of artificial where computers break
down is in taking into account "context".
Henri
graham_douglass
In article <5lol3b$lvf$1...@news00.btx.dtag.de>, TUESCHEN.MEDIZ...@t-online.de says...
{snip}
>>I think that in intelligence, quality without quantity is nothing.
>
>>Maybe DB v GK demonstrates that in intelligence, quantity is more important than
>>quality.
>
>insulted me severely.
>You may write the next time *in intelligence as far Graham could understand
>it*. Then it would be in good order.
>Look -if you want- _intelligence_ is what?
>Look, the schism between quality and quantity is only a delusion.
>And sometimes quality comes up with sufficient quantity. Learning by doing.
>
>The reason I write so many posts. :)
I'm not sure that I need to apologise. You have not yet demonstrated that
quantity is less important than quality in the intelligence mix.
Learning by doing is the acquisition of scripts and micro scripts, which is
adding to the quantity of your knowledge.
btw - I think that the quality of your posts has improved! :-)