Slashdot Mirror
The New (Computer) Chess World Champion
An anonymous reader writes: The 7th Thoresen Chess Engines Competition (TCEC) has ended, and a new victor has been crowned: Komodo. The article provides some background on how the different competitive chess engines have been developed, and how we can expect Moore's Law to affect computer dominance in other complex games in the future.
"Although it is coming on 18 years since Deep Blue beat Kasparov, humans are still barely fending off computers at shogi, while we retain some breathing room at Go. ... Ten years ago, each doubling of speed was thought to add 50 Elo points to strength. Now the estimate is closer to 30. Under the double-in-2-years version of Moore's Law, using an average of 50 Elo gained per doubling since Kasparov was beaten, one gets 450 Elo over 18 years, which again checks out. To be sure, the gains in computer chess have come from better algorithms, not just speed, and include nonlinear jumps, so Go should not count on a cushion of (25 – 14)*9 = 99 years."
"Although it is coming on 18 years since Deep Blue beat Kasparov, humans are still barely fending off computers at shogi, while we retain some breathing room at Go. ... Ten years ago, each doubling of speed was thought to add 50 Elo points to strength. Now the estimate is closer to 30. Under the double-in-2-years version of Moore's Law, using an average of 50 Elo gained per doubling since Kasparov was beaten, one gets 450 Elo over 18 years, which again checks out. To be sure, the gains in computer chess have come from better algorithms, not just speed, and include nonlinear jumps, so Go should not count on a cushion of (25 – 14)*9 = 99 years."
I would be curious to see how an algorithm based on millions of actual games fares against a pure mathematical model. "Based on your interest in taking queens with a pawn, you might be interested into taking a bishop with your rook".
lucm, indeed.
I don't know about the chess championship, but that is one of the best blogs on the internet. Where did it come from? In a world of 7-second attention spans, "you won't believe what happens next....", and pop culture domination, here is a guy who is talking about math, computers, and games in a friendly relaxed manner, because it is interesting to him. He talks about Godel (who apparently said, "Religions are, for the most part, bad—but religion is not"), some recent ideas in information theory, and a comparison between linear algebra and quantum computing. He uses LaTeX.
Also, from the post I learned about a game called Arimaa, which was designed to be hard for computers but easy for people. There is a bet that no computer will be able to beat a human, and you can win thousands of dollars if you do. So far it's apparently not even close. Also, got this great quote: "It’s not that chess is 99% tactics, it’s just that tactics takes up 99% of your time."
"First they came for the slanderers and i said nothing."
Moore never said anything about clock speed. He said the number of transistors. A CPU with more transistors can compute a complex problem faster. There is a science to trying to come up with problems that aren't solvable faster with more transistors (even at the same clock), but those problems are rare in the real world. Today's 3Ghz processor is faster than a 3 GHz processor from five years ago.
Recently, the big improvements have come from organizing the increased transistors into increased cores. Today's 8-core CPU can analyze eight moves at once. The older dual-core system could only analyze two moves at once. Therefore, the 8-core system is four times as fast, on a parallel problem like this.*
* Actually even better than four times, because a fractional core is needed to manage the overall process. The dual core chip could analyze 1.5 moves simultaneously, the 8-core can analyze seven simultaneously.
I'm not sure where you got the idea that chess is solved, but we're still a looong way from solving chess. We have only solved chess with 7 pieces, not the full 32, and unless quantum computers arrive in force, we have no shot at solving it in our lifetimes.
http://en.wikipedia.org/wiki/S...
I'd give my right arm to be ambidextrous...
i'd say go, probably. the rules are more minimal; the state space is larger and more connected with fewer dead-ends and "gotchas"; and it seems just as difficult/competitive. it's not unreasonable to think that games with more rules will tend to produce more specialized skills, and that specialized skills are inferior to general skills as far as "brain development" (whatever that means) is concerned.
also, dual n-back is supposedly shown to increase general intelligence rather than just skill. it's not exactly a lot of fun, but that's not the point.
"They were pure niggers." – Noam Chomsky
No it isn't. Far far from it.
I went to school with someone who was UK national chess champion for his age bracket for most of the time until about age 14. He never demonstrated any exceptional ability in anything else (not top in his class in maths at any age, for example) and didn't get exceptionally good results in his later exams. I realise that this is a single anecdote, but it left me quite sceptical about the ability of chess to train the mind. I was on my school chess team for a bit, but eventually I found purely deterministic games to be as tedious as purely nondeterministic ones.
I am TheRaven on Soylent News
To add to this, Go also uses both hemispheres of the brain. There is the logical aspect of the game, where you read out a sequence of moves, and there is also the visual pattern aspect, where a player will look for moves that make "good shape", or will look for other patterns on the board that have their own traits.
If you have to solve A before B, you can work all possible A in parallel, then all possible B in parallel. More cores is better.
Also, just because you have to solve step A before you can solve step B doesn't mean you can't START working on B, such as solving B for likely values of A, storing them in a lookup table, then selecting the precomputed answer from A. In fact I won a prize doing exactly that with my software playing a game against humans.
>. It's not hard.
Harder than it first appears, we just demonstrated.
I am disappointed to see only one mention of late Don Dailey in TFA. He is actually the guy who wrote the whole thing. I had followed his posts for years in computer go mailing list. I have learned a lot from him as an R&D engineer in an unrelated field (chemical industry). While many people adopted "improvements" only because it made sense to them, Mr. Dailey had a very systematic and methodological approach to changing the program. He had ideas and insights for improvement like many others, but he never fell in love with his own ideas. If something did not work, it did not. No matter how plausible it seemed. He also had most patience I have seen of an online person. He would carry on discussions long after it was obvious the other party was not paying enough attention or was simply stupid. He did this almost to the day he died.
Congrats Mr. Dailey. You have done it.
Gentlemen, you can't fight in here, this is the War Room!
PS, the other thing I did (successfully) was pre-compute symmetry of A and B, so I knew that any value of A in this list would give the same result in B. So I didn't have to compute the value of B, after I knew A I only had to select which bucket it was in to get the value of B.