Brain vs. Computer: Place Your Bets

dev_null_ziggy writes: "CNN reports that the current chess guru is going up against a supercomputer, amusingly titled 'Deep Fritz.' The match is scheduled for October, and the current champion, Vladimir Kramnik, stands to win $1 Million dollars if he wins. Of course, since he'll be snagging $800k for a draw, and $600k for a loss ... I'll give two to one odds on the machine."

Re:Humans has to win, right ?

[johnnydeggplant]

I agree with you that survival is an instinct found in humans, and most of the rest of the animal kingdom for that matter, but I don't think that it is a neccessary method in the live class definition. The only reason that it seems that all animals have this instinct is that the lifeforms that we consider animals have gone through a long period of survival of the fittest (think back to when you first loaded windows 3.1 on your 286, it was longer than that!) It the animal didn't have survival instincts, it wouldn't survive because, so to say, it wasn't interested in surviving (much like my programming career after windoze XP secures its strangle hold on the internet.) It we preempt those years of evolution and create a machine that is 'intelligent', it wouldn't necessarilly contain the will to survive. Intelligence is only what you define it in that if we create something and label it 'intelligent' it's still what we created. For example, I have a Psy-Duck pokemon stuffed animal with an ISO 9000 certification pin in it's 'hair' and sometimes I see intelligence in its stoic approach to life. Well, that's my stuff, Dave

computers play Chess well, but suck at GO

[andi75]

Computers play Chess quite well, they're even stronger at Checkers, and they rule at Connect 4 (a friend of mine has written the strongest freeware checkers and c4 programs [fierz.ch], check them out).

However, they suck *badly* at GO. This is because the branching factor (that is the average number of available moves) is about 30 at chess, 10 at checkers, and 7 at connect4. GO has an incredible branching factor of *over 200*. That means, the typical approach of 'alpha-beta' search breaks down.

If you're into researching new board game algorithms, try GO.

- Andreas

Re:Chess Rules Changes (Human vs. AI)

[CodingFrenzy]

A well known experiment from cognitive psychology is the following:

Novice and expert chess players are confronted with two types of chess setups and are asked to analyze the positions. Setup type 1 is an actually possible setup (e.g. can be reached by legal moves), while type 2 could never occur during the normal course of a game.

The novices do equally well for both setup types. Unexpectedly, the chess expert don't! While type 1 setups are analyzed very fast, the experts take approx as long to analyze the impossible setups (corrected for trained mind etc) as the novices.

A chess computer will probably behave as a novice in the above experiment (if the setup is not in the library), ie take the same analyzation time for both setup types.

Hence, this could lead to the conclusion that even a tiny adjustment in the rules could shift the odds heavily towards the computer, if these rules suddenly allow positions not possible previously.

Re:Chess Rules Changes (Human vs. AI)

[Anonymous Coward]

Interesting idea. Back when I was a senior in high school, playing chess caught on really big in my school. Everyone was playing it, even the football players. Anyway, we made up some new variations (some which helped the games go faster-- a necessary thing in the days of 50-minute free periods). Here are a couple of them:

1) No pawns: It makes a much quicker game, and maybe I didn't play it enough, but it didn't seem to matter who went first.

2) Only three pawns: And you can put them anywhere you want on your side of the board.

3) Secret King: The "King" is whatever piece you want it to be. Before the game, each player writes it down on a piece of paper and places it under the board. The catch to this version is that there are no "checks," per se. If your secret king is in check, you don't have to tell anyone. I kind of liked this version because it adds a sort of "poker face" aspect to it.

4) Bombs: Any _ pieces (we usually played with 2 or 3) are bombs. When they are taken, the other player loses piece that he just used to take that piece. This is done the same way as the secret king: each player writes down beforehand which pieces will be bombs.

I'm not sure how this could work with the your suggestion to use it in a match b/t computer and grand master, but I think some of them might be pretty cool. Either way, I'd be curious to see what other people think of our little way of relieving boredom during free periods.

Re:Not unusual

[max_power26]

I got it from the Article. All I can find is that the original Fritz (running on a P90) beat the Deep Blue Prototype the year before it played Kasparov.

'By winning the championships Fritz demonstrated that chess knowledge was at least as important as computing power - Fritz was using one of the least powerful computers in the tournament (a standard Pentium 90MHz PC supplied by the Chinese University of Hong Kong)'

See: http://www.dcs.qmw.ac.uk/~icca/WCCC8/chess95.html Round 5 is DB vs. Fritz

Re:That will be with "long" matches?

[Skuto]

>played with "fast" matches, allowing less time
>than normal championship "long" matches.

The match was fast in the sense that few games
were played, but Kasparov was allowed the full
thinking time.

>I had the idea no computer had been created that
>could beat a Grand Master human in a "long"
>match.

Those have existed for quite a while now. Most top
programs have no problems with 'weak' GrandMasters
(sub 2600 ELO rating) even at long timecontrols.

--
GCP

Re:Chess Rules Changes (Human vs. AI)

[kraada]

There are already variations like this. Of note is Nightmare CHess (SJG), which is an almost MTG like adaptation to chess. Basicaly, you get 5 cards, and can play one before, or after either your move or your opponents, depending on what the card says. Cards say things like: Earthquake: rotate the board 90 degrees either clockwise or counterclockwise. Promote all appropriate pawns. Earthquake counts as your move. or New Tactics, which makes pawns move diagonally and capture forward. It's a crazy game. I've also personally seen a USCF Senior Master play against some of my friends, and whomp us all, proving that it isn't all luck . . . It'd sure be interesting to see a GM play it vs a computer :)

Re:Humans has to win, right ?

[Anonymous Coward]

Maybe we are looking at it all wrong. The computer does not "beat" the person. It does not desire to win. It is a tool that does exactly what is programmed into it. It is like saying a man can beat a car in a race. You actually beat the person driving the car, not the car (or more likely lose to the guy driving the car depending on the course). Really, this matchup is Man (without the use of tools) vs. Man (using only a prebuilt tool).

And, upon checkmating the machine, a cry is heard

[wowbagger]

Fritz! They've killed Fritz! Those dirty rotten fairies!" [imdb.com]

Seriously, though, this is why chess is not a "game" in the game theory sense of the word. Every move has known, predicatable consiquences, and all the data is available to both sides during play. As a result, as computers advance, they will become better than people, because chess is a computation, not a game

Now, consider poker. While somewhat simpler in terms of the number of moves available to a player at any given time, they player cannot predict with complete precision all possible outcomes of a given play, since he does not know what cards are coming up next, what cards the other players have, and therefor cannot winnow the solution space significantly. In poker, the machine cannot easily tell if I am bluffing or if I just completed my royal flush.

Now, for a REAL computational challenge, make a computer that can play Magic, the Gathering worth a darn. Talk about "limited information" - you don't know what cards the other player has, you may not know the powers of the cards, and you may not even know what's coming up in your deck next. Make a machine that plays that well and I'll be impressed.

Re:Chess Rules Changes (Human vs. AI)

[BigGar']

There is a chess variant that was proposed by Bobby Fischer to eliminate the open book knowledege that top players and computers have. It's called Fischer Random Chess and it works by scrambling the arrangement of peices under the following rules, 1. must have opposite colored bishops & 2. one rook on either side of the king. Black and White have the same arrangement of peices & there is also some modifications to the castling rules. What this gives you is 960 possible starting positions and effectively eliminating opening theory and get's down to how well you play, not what you've memorized.

Sidenotes to the Deep Blue - Kasparov Match

[Anonymous Coward]

There have been 2 matches between Deep Blue and Kasparov, who was the strongest player in the world up until last year. Kasparov won the first one, and lost the second. There are however a few sidenotes that have to be made about the second match: - Kasparov wasn't allowed to analyse games of Deep Blue in advance, they were classified secret by IBM. This was very unfair, because the Deep Blue team had all games Kasparov has ever played in his professional career on file, and they used this in their preparation. This is very normal in chess, everybody prepares on his opponent. Kasparov could not, while Deep Blue could. - Deep Blue had very large opening and endgame databases. One could argue that they are part of the program, but it is not a very strong argument. Databases are not part of the program, it would be equivalent of Kasparov using opening books during the game. Obviously, Kasparov wasn't allowed to do this. I don't think this is entirely fair, the machine having at his disposal all opening theory in existance, every professional chess game ever played and every endgame ever analysed (we're talking about several hundreds of GB's here)... - Deep Blue wasn't a very good chess program, compared to other programs like Fritz, however it had a lot of power. And it had something else, it was designed completely to counter Kasparov's style, against any other opponent it would have played much weaker. This is, in my opinion, not entirely fair. If a chess program is superior to humans it should be superior to all humans, not to whatever human happens to be the best at that moment. - Furthermore, Kasparov simply isn't the best anti-computer player in the field. His playing style doesn't work very well on computers. Every human has another style, and I don't think Deep Blue would've been able to counter Karpov, for example, even though Karpov is obviously weaker than Kasparov. - Finally, all chess analists agreed that Kasparov played very poorly that match. They all agreed that he has played much better in the past.. All in all, I don't think computers were all that superior in 1997, and I think Kasparov would have had a big chance in a rematch, had it been fair (my first 2 points). However IBM had 'proven' a point and abandened the project... Of course, it's now 2001, so computers have become stronger. However Kramnik is not Kasparov, and I think the match will be pretty interesting....

There's still hope for us humans...

[Anonymous Coward]

As has been said already, Chess programming has very little to do with AI. The reason we get beat is that for most positions, there are only about 20 to 30 legal moves, and only a few of these are sensible, so brute force lookahead is possible.

To see how bad computers really are at strategic thinking, all you need to do is look at a game with a much higher branch factor (meaning more legal moves each turn).

One good example is the Chinese game of Go [britgo.org], which has an average of about 200 legal moves. Computers are absolutely dire at this game. Interestingly, one of the better Go playing programs is Free Software (GnuGo [gnu.org]). It still loses to half-decent humans though.

Anyone who puts 2:1 odds on Fritz is a moron.

[Supercoz]

1) Kramnik is rated about 200 FIDE points higher than Fritz. Fritz played an equal match [all games drawn] with Dr. Robert Huebner of Germany [mid 2600 elo]. Kramnik is much stronger, and he has more energy due to being 40 years younger.

2) Anti computer techniques exist. Basically, computers have excellent tactical vision but poor strategic vision. So if I try to catch one with a knight fork, it won't work, but it will happily fall into a positional trap and lose 40 mores later. Most GMs don't bother to learn these because they would rather spend their time trying to beat other humans.

3) Kramnik's style is more suited to playing a machine than Kasparov's. Kasparov is mainly a tactical player. He wins by outcalculating his opponents. Kramnik is mainly a positional player. He wins by strangling people in the endgame. When he played Deep Blue, Kasparov tried to play positionally. It just wasn't his style.

4) Most GMs use Fritz as their computer analysis program, including Kramnik. He has a good feel for how the program works and can prepare opening surprises.

Don't get me wrong, computers are getting better and better at chess. Traditionally a program gains approximately 50 ELO points for every doubling in processor speed. Given that Fritz is approximately a 2600 program, that to convincingly beat Kramnik would require about a 2900 rating, and that processor speeds double about every 18 months, I predict that in another 10 years or so a PC may be able to give the world champion a run for his money. But not today.

Forecast: Kramnik 6.5;Fritz 3.5. Kramnik +3=7-0.

Bobby Fischer

[Anonymous Coward]

Oh, Bobby Fischer is still around. He had a very
public rematch with retired ex-champ Boris Spassky back in '92 or so. (note that Spassky was probably the weakest "world chess champion" there has been)

The match showed that he still has some fire, but was quite rusty. He wouldnt be ranked in the top 10 in the world today...and isnt getting any closer by his inactivity.

Fisher is, well, seriously loopy. He'd most likely never play in a tournament, or under any other circumstance where he didnt have total control of the playing conditions. When I say loopy...i mean "the-CIA-put-a-mind-control-radio-in-my-fillings" type loopy.

After he dropped out of the chess scene (after 1972), he joined some religious sect, which bilked him out of his money. He basically orchestrated the 1992 "rematch" for the money. He got something like $5M, but he's wanted by the US Government for some minor things -- so he's living in Europe now. He still considers himself the World Chess Champion.

Re:But...

[klmth]

Not quite. They programmed Deep Blue, for example, by feeding it's heuristical engine with data from thousands of previous games. This data, together with the learning ability of the machine and the raw computation power yields to a chess computer far more effective than the people who programmed it. Not that it neccessarily is any better than a human being: Deep Blue didn't win every game against Kasparov, it merely won more games in the turney.

Chess Rules Changes (Human vs. AI)

[wass]

I think a far more interesting 'battle of the brains' tournement between computer and human would involve something that as quite, but not entirely unlike, chess. I've spoken with some friends about this idea, and they agree.

Chess players have spent nearly their entire lives studying the way pieces move around the board, whether they realize it or not. They can see several moves into the future more easily than a chess novice with equal intuition. That said, I would be far more curious to see how a chess player would handle playing a different game. What I mean by this is change the rules of chess slightly, then allow 3 days for both computer programmers and human challenger to learn/recode the new rules and strategies.

Sample rules changes could involve knights moving orthogonally 2:2 instead of 2:1, or having a borderless board, where if you move off the left you come in on the right side (like Pacman), or marking a few squares as off-limits for the whole game, etc.

I think that some of these rather simple alterations of the rules would drastically alter subsequent gameplay. I also think that a chess novice would do roughly equally well in the various scenarios (albeit rather crappy in 'classical' chess). But more interestingly, how would the chess expert do? Would these new rules to him be like learning an entirely new game? Suddenly he wouldn't have the benefit of 20+ years of practice, and would have to 'see' things as they were for the first time.

I would be very interested to see how the great chess masters would do against computers in these situations. People often hype the human/AI chess games as battles to see whether computers are smarter than people. I think 'modified' chess would be a more interesting study. Do the great chess players really possess that much more wisdom and foresight, or is it some experience acquired by 20+ years of watching the pieces move.

I posted this idea on /. a few years ago, and I got some angry replies from chess players indicating that chess is all intuition and that rules changes wouldn't matter. Well, anybody care to find out?

Re:Humans has to win, right ?

[Fred Ferrigno]

It is a rather strange thought that the human mind can create something which is superior to itself

Why? After all, the human body has been used to construct machines that are physically superior for many years.

My sentiments exactly. Humans created can openers, and they're a darned sight better at opening hard metal cans than I am.

Re:Humans has to win, right ?

[ianezz]

if this thing can think, it's going to realize that it should stay hidden to survive

If it wants to survive at all, of course.

I'd say that most (if not all?) life forms out there have some sort of survival instinct because, ehm... only the individuals (or group of individuals) that took care of surviving actually survived enough to reproduce themselves and transmit this characteristic to their descendands.

If we take an evolutionary approach, for each AI actualy willing to survive, there would be a lot of others not giving a dime on the subject, and the ones performing the selection would probably be human beings (at least in the earlier stages). Given this scenery, staying hidden is almost surely a bad move for an AI in order to stay alive.

OTOH, if you are able to tell an AI that it has to survive, you probably can also tell it to be kind towards other life forms... YMMV. :-)