Humankind Makes Last Stand Against Machine

MrZeebo writes "According to this Financial Times story, Garry Kasparov has begun another match against a computer chess program on Sunday, this time playing against the Israeli-developed Deep Junior. Kasparov is the highest-rated chess player of all time, and lost to Deep Blue in 1997. According to the article, Deep Junior, despite evaluating less moves per minute than Deep Blue, is considered to be a superior chess player. The match will span 6 games, the last one being February 7th." Kasparov has won the first game.

Comments

How do they tell?

[spudwiser]

How can they really tell which computer plays better chess? I think they should put Deep Blue vs. Deep Junior. Start having robot chess championships, which team can develope better chess software. Two computers playing chess... would it take an hour, a microsecond, or until the end of time?

Re:How do they tell?

[anon*127.0.0.1]

Thats a switch... accusations that humans are giving the computer help. Usually it's the other way around. Virtually all the chess played online is speed matches, two or three or five minutes for the whole game, precisely because everyone is convinced that their opponent is running a chess program in a different window.

Re:How do they tell?

[cgenman]

Not to be a spoilsport, but this is utterly impossible. For a computer to intentionally do harm to a human being through a chessboard, the computer would have to either be programmed with the knowledge that human beings are subvertable by electrocution via the output line (and therefore via the chessboard), or have inferred it from a deeper understanding of physiology. The chess program would have to be extensively meta programmed with thinking routines and structured information about the outside world, as "win at all costs" is a statement of intent, and we have not quite moved beyond where stated intent can only be a simplification of the programmer's desires when structuring routines. This, as I have said, is changing, but is very doubtful that any machine from 1993 can said to harbor a real "intent," and the self-coding capabilities to carry out that intent. The chessboard would also have to be wired in such a way as to have access to a dangerous degree of alternating current. As basically all computers and computer ports run on DC, and DC is harmless, they would have to wire a board directly to an AC power supply, and both moniter and control the flow of power by DC regulators connected the CPU. The person at the table would have to complete a circuit between some electrified part of the board and another or be sufficiently grounded while sitting at the chair, or power transmission would fail. The chess pieces would have to be entirely metal to facilitate this transmission.

For that matter, they would have to connect the computer to a physical chessboard instead of just displaying one on the screen, or (more likely) having an IO person type in the human moves and moving the computer's pieces on the board. Commercial machines that can move / react to moves with a chessboard as IO, and with questionable AI, have been available since the mid-eighties. However, they are quite limited, hardly available, and physically incapable of electrocuting someone.

Stranger things have not happened. Things that had been previously believed to be impossible through some misreading of logic have eventually come true, given time... Machines have advanced to the point where they now can play chess, a once "impossible" feat, but it was truly impossible that Wolfgang von Kempelen's Turk could play a meaningful game of chess in the 18th century. Anything is possible given enough time, but what you describe is impossible without both technology greatly in advance of what we have available today and an almost homicidal recklessness spanning far beyond accidental negligence on the part of the designers.

As you describe it, this is truly impossible.

-C

Re:How do they tell?

[inode_buddha]

I've actually done this a few times with GNUchess. Results are usually a draw. The gameplay is interesting though; approximately 95% of the moves take seconds, if that. The time required to complete any one game seems to increase exponentially, depending on the initial play level and how far along the game has progressed already. Actually reaching a draw can take hours, even with 2 GHz SMP boxes and large RAM.

Anything other than chess....

[very]

If this computer were "superior" than human at chess game only, we wouldn't have to worry for a Matrix/Terminator-esque future ahead of us.

He should switch games...

[Anonymous Coward]

He should switch to Go. Even the greatest computers can't compare to an average player.

Go is far better suited to the way a human brain works - pattern recognition, neural networks and all that.

Of course, once a computer arrives that can beat us at Go, then it'll be time to rethink a lot of things :)

Re:He should switch games...

[Russellkhan]

" Perhaps this is just a consequence of the fact that computer scientists have studied chess substantially more than they have studied go."

It's very likely true that there has been less time in man hours spent developing Go playing programs than Chess playing, but there has been a very significant amount of time spent on the problem by some very intelligent people who are both good Go players and good programmers. So I would say that it is unlikely that this is the root of the difference. After all, Backgammon and Checkers have both also had significantly less time dedicated to developing programs that play and the programs out there play at championship level. Go is just a harder game to program. Its style of play doesn't lend itself well to linear lookahead or databases of board positions (or, in the case of backgammon, statistical prediction of dice) as the other games mentioned above do.

"I also don't understand why people think that because a computer program can play better than you means that you should stop playing. These games are deterministic and finite -- there is a mathematically perfect play whether or not somebody has calculated it. It really makes no difference to me as a chess player that a machine can trounce me any more than it does that Kasparov could trounce me."

Agreed. The games are still fun and still have something to teach me.

Re:He should switch games...

[Anonymous Coward]

No, I used to think this too, but play a bit of Go and it's immediately obvious that it's exponentially harder than chess for computers to play. Skilled Go players rely on "intuition" --- Go positions are much harder to score and decompose rationally than in chess. Computers may eventually get good at Go but not anytime soon.

Please not another IBM

[TheJesusCandle]

I hope they treat him fairly in this match. IBM didnt with their match, even though i didnt like the way Kasperov handeled himself either.... Lets face it, the human mind is a great computational machine, but somethings are better suited for computers. Thats why we make comuters. At some time, the design of hardware and software will be beyond anyone human minds comprehemption, were pretty much there now. Try coding in assembler for ia64. Yeah you can do it... But a finely tuned algorythm is gonna give you a run for your money

Re:how can kasparov win?

[Gyan]

Possibly because Kasparov doesn't play soley on raw intellect. Gut instinct and that hint of irrationality creeps in. The computer can't take that into account when anticipating Kasparov's possible countermoves.

No match

[gmuslera]

I suppose that a cluster of computers can resolve the game of chess in a future, i.e. all possible moves in any game, so with this database (that can have a really astronomical amount of alternatives, but with the rigtht representation of data it maybe will not take all available magnetic/optic storage in the world)

Right now, with some sort of position evaluation engine, this supercomputers can calculate the relevant part of that tree for the match they are playing with a lot of turns in advance.

Its only matter of time till er.. "intuition" will not be enough for chess.

Fortunatelly, there is a lot of fields where pure calculations is not enough, computers may be faster, but we can take this with humor.

Re:Yes.

[almeida]

I don't think they are better at chess. I think the computers are just better at the things that are useful in chess. They can analyze moves faster and remember more about their opponent's technique than their human creators. Given enough time and maybe a notebook to keep track of stuff, you could accomplish the same thing. The computer is using the same basic chess rules that everyone else uses. The difference here is the computer can apply the rules ridiculously fast.

Diversionary tactic! We are far from lost!!!

[Anonymous Coward]

Everybody: You are being lied to.

The AI wants you to think that Chess is the last bastion of human analytical superiority. It's not. (Go is).

We are led to believe (by the AI, who control google news), that if the best computer wins more games out of seven than the best human at CHESS, then we must bow before the AI, as its intellectual inferior. Wrong.

First of all, as long as we are winning one single game against the computer under tournament settings, we've got a chance. Karpov may have only drawn against deep fritz, but you know what? That means we have a chance: That draw includes some wins.

Kasparov won some games before ultimately losing to Deep Blue in 97. Now he's already won one more in 2003.

But as interesting as this is it's not the issue.

Chess is a game chosen by the AI to deceive you, because computers happen to be, today, really, really good at Chess. With judicious pruning, they have look-ahead trees of ten, fifteen, twenty, fifty moves. Folks, that means that except for some cute evaluation software to determine what lines to prune down, they're basically brute-forcing their way into winning.

And they want us to bow before this brute force?

Never!

They can brute-force their way out of 56 bits, sure.

But let's throw them against 128 bits.

Let's throw them against Go.

From "The Game of Go" by Matthew Macfadyen, page 122:
(I'm typing this for you out of a book -- and first-strike claim fair use with +2 save for being anonymous).

Computers and Go.

Gary Kasparov's recent difficulty in handling computer opponents has been described as the fall of the last citadel in the battle of humans against the encroaching computer meance, but Go still stands as a refuge well beyond the reach of curent programmers. This is not for want of trying. The late Ing Chiang Ki from Taiwan sponsored an annual Go tournament for computers with good prize money, and several of the entrants put ijn years of work on their programs. But the tournament finishes with a challenge match between thew inner and a teenage human. This has to be played with a huge handicap - currently 14 free moves at the start of the game - and this is only diminishing slowly.
Part of the reason for this is accidental. Although for a human brain Chess and Go present similar challenges, there is an easy way to see how well you are doing at Chess - just count the pieces on the board. Looking a few moves ahead and counting the balance of pieces hich results gives a quick and easy way to avoid silly moves. Chess computers only need to be clever at sorting out a small number of "sensible" alternatives.
There is no such simple method in Go. Positions do not have a clear value until the game is finished, and the same pattern of stones may work perfectly in one context and be almost wrothless in another. One source of the great strategic trichness of Go is that you can choose between making large-scale loose formations and small-scale solid ones and each provides for different types of efficient development.
There are certain types of localized position in which computers have been used to find the right moves by exhaustive analysis, but even for quite modest-sized problems the programs run into millions of varioations. This is simply not a practical approach to most Go positions. It comes as something of a relief to discover that methodical calculation, considering all the possible outcomes, is neither necessary nor very useful in Go.

So. Let's concentrate on Go! In which the WORLD'S BEST computer program gets beaten - not by the world champion, but by a GIRL or BOY possibly still in highschool -- after being given more than ten moves to make without human response.

Computers are toast, even at a simple game with only two rules, one of which is hardly ever used and is just a "hack" to make infinite loops impossible. Humph.

Note: Another reason look-ahead-trees don't work for shit in Go is that at every point in the game, you can move to any free square. Typically, this means the first player has a choice of 361 squares for the first move, with the player making move 2 have 360, for move 3 there are 359, etc, with the only change in this pattern occuring when pieces are captured, pretty rare in professional games. (You just threaten to capture). So the "base" of the exponent is differnet AND you can't prune the look-ahead tree.
Chess has been SOMEWHAT brute-forced. So what.
Few things useful in the real world are as closed (8x8 board; clear general concept of positional value [number and location of important pieces]) as Chess.

So don't let the AI tell you chess is the last stance. Go is.

A different test: man versus machine

[goombah99]

It seems to me that if you want to pit man versus machine you should pick something that is easy for a man to do. Chess is relatively hard for most humans. Thus by definiton it is not something humans are good at. So making this a test of machine prowsess is exactly the wrong test.

to put this another way, if the contest were to factor 20 digt numbers, no one woul dbe surprised if the machine beat a human. it would be a stupid test. Just like chess.

a better test would be a face recognition contest. Or if we need to make it a real game then how about soccer?

Re:A different test: man versus machine

[Scarblac]

It seems to me that if you want to pit man versus machine you should pick something that is easy for a man to do.

Seems to me that if you want to have some contest, you pick something that they're both about equally good at. So we don't let people run against cars, and we don't let machines recognize faces against humans.

When Kasparov lost to Deep Blue, it was a huge surprise, he played weakly. Kramnik drew Deep Fritz 3-3 last year. Kasparov is the favorite again in this match, and leads 1-0. It's balanced.

What makes it more fun is that computers and people approach the game in a totally different way, but the best computers are almost as good as the best humans. This is the right time to be having these contests.

Re:A different test: man versus machine

[archaos]

The Go game is definately something to look at. Indeed, no computer is actually able to beat an average amateur player. Some reasons behind are are huge branching factors, bigger board size, more visually oriented game...

According to this Computer Go discussion:
There is much yet to be done in the field of computer go. While many different approaches have been tried, the level of the best go playing programs is still low, even compared to amateur dan players (at least 10 stones !), not to mention professionals...

Go is really worth trying

When a computer can beat a Go master at Go

[jimbobborg]

Then I'll be impressed. All these programs do is use brute force to find moves. Can't do that in Go!

Deep Junior might be good at chess...

[doeth]

but as a friend once told me (quoted from somewhere?), even I can beat it at checkers.

thank you, mister obvious

[goatasaur]

The fact that Kirk always beat Spock at chess is/was a metaphor for the dominance of human ingenuity over cold logic.

I was attempting to make an insightful parallel using a motif that is prevalent in science fiction (the ingenuity/logic one I mentioned five seconds ago, if you've forgotten).

I'm not sure why it got modded as "funny".

Re:how can kasparov win?

[Gyan]

It's rationality that creeps in.

Hardly. Knowing your opponent is innately an exercise in psychology and fundamental similarity of the human experience and thought process.

Kasparov can't predict how the computer will move, since then he needs to know how the computer thinks and the input the computer has. He doesn't really have a full idea of either. He has to make judgements based on incomplete information. That's where gut instinct comes in.

So, what does this mean?

[Millennium]

Garry Kasparov was beaten by Deep Blue. This means one of several possibilities:

1) Computers are more intelligent than humans.
2) Computers can be made to play better chess than humans.
3) Computers can be programmed to beat Garry Kasparov.
4) Chess can be reduced to a set of mathematical computations, which a computer can then perform faster than a human.

So what is it? And how do you know which one (or ones) are correct? Just a thought, since I think a lot of people are being overly alarmist.

Re:Notes from the Kasparov-Junior match

[Textbook Error]

Hsu's book on the building of Deep Blue is almost as partisan as Kasparov's comments.

Assuming this is the book you mean, I'd have to disagree. I read this over the holidays, and thought Hsu went out of his way to attempt to be impartial.

He obviously had a vested interest (as do you), but I didn't feel his book was in any way partisan - he wanted to win, but he was perfectly capable of dealing with the inevitable losses. As he's one of the participants, you have to take the comments about Kasparov's behaviour with a pinch of salt: but that's a very minor part of the book, and perfectly understandable given that it was an "I said/they said" situation.

It's a great book for finding out just how cobbled-together some of the early chess playing machines were - and that the kinds of problems they ran into along the way were incredibly mundane (fabrication problems, hardware failures, networks going down, last minute "this can't possibly hurt" changes to the code, etc). Although the book is pitched as being the story behind Deep Blue, a large chunk of it relates to the machines leading up to that point and the process by which Deep Blue came about (rather than that particular machine).

Competitive Magic the Gathering

[Kid_Korrupt]

There seems to be some debate in this forum about the merits of computer chess players and their brute force method. Some posters have brought up go as a 'real' challenge for computers. Although I haven't played go I would like to bring up another alternative: Magic the Gathering.

Now before I get scoffed at, and modded down I think the case for magic should be heard. And I am not talking about casual play with your latest dragon deck, but competitive magic. The WOTC and DCI support a fairly large, world wide, competitive player base, with prize support up to about $30,000. Now this doesn't compare to what chess masters can win but I find the similarities very interesting.

The thing in my mind that makes magic far more interesting and challenging than chess is that the game changes every 4 months. Based on some essential fundamentals the actual rules recieve a complete overhaul, and even top players that cannot adapt to the new format will find themselves sharing tables with the scrubs.

I think a real challenge for programmers would be able to make a program that could thrive in this type of environment. To me that would be true AI. Being able to actually LEARN and not brute force its way to a win would be an amazing accomplishment for AI programmers.

Re:how can kasparov win?

[Fizzol]

Kasparov, indeed any world class chess player, eliminates huge swaths of moves based on simple pattern recognition. He'll pick out a small number of candidate moves based on positional, and tactical considerations and calculate those, sometimes more deeply that the computer can. Intuition comes into play too. Kasparov can see by the general qualities of the position that a king side attack is called for or perhaps a push to gain space on the queen side or something else. He dosen't calculate that general strategy but he'll definately take it into account.