Crazy Stone computer Go program defeats Ishida Yoshio 9 dan with 4 stones • Hacker News Board

Re: Crazy Stone computer Go program defeats Ishida Yoshio 9 dan with 4 stones

#121

Post by gcp » Thu, Mar 27, 2014, 3:59 PM UTC

sanoli wrote: post #114

S4M wrote: post #94

Earlier quoted context omitted.

My FIDE rating is slightly above 1900. I think that with 2 bishops I'll beat Magnus Carlsen easily, simply because I could sacrify one bishop for some attack and still be ahead in terms of material. But with only one bishop ahead at the start of the game, I doubt it will be enough.

2 bishops is a massive advantage in chess, but I still don't think a 1900 player could beat Magnus. Maybe a 2200 player?

2 bishops is IIRC about 700 ELO-equivalent, so this looks like a good estimate to me. (That said, 700 ELO difference is >99% winning rate, so we're in outlier territory, and as said in the other post, that would prevent me from taking bets :)

Re: Crazy Stone computer Go program defeats Ishida Yoshio 9 dan with 4 stones

#122

Post by T-hawk » Thu, Mar 27, 2014, 4:17 PM UTC

gcp wrote: post #110

throwaway5752 wrote: post #91

Earlier quoted context omitted.

My limited understanding of go is that the stone placement confers position and tempo, too, which would be undercounted looking at it purely from points.

Points (which are a game specific metric) have absolutely no relation to anything I posted. I didn't even mention them at all! It looks purely at game outcomes given the initial starting conditions.

Yes, you've got the right answer. Forget material and point comparisons which don't really port between the games. The winning percentage conferred by the handicap is what matters.

Four stones in Go sounds like a lot less than four pawns in chess, but if that's what the math says by each resulting in the same winning percentage, then the equivalency is true.

Re: Crazy Stone computer Go program defeats Ishida Yoshio 9 dan with 4 stones

#123

Post by T-hawk » Thu, Mar 27, 2014, 4:58 PM UTC

MichaelGG wrote: post #99

baddox wrote: post #42

Earlier quoted context omitted.

> Combinatorically, Go is considered pretty much the most complex game in existence Is that considering "game" to include only traditional turn-based board games? Presumably it would be trivial to make a video game that's vastly more complex (at least in terms of tree or state space complexity) than Go. Any real-time strategy game should quality.

Are top RTS games like Starcraft hard for AI? I know the default AI isn't strong, but is Starcraft in theory difficult to trounce humans? I spoke to an ex-pro player and he asserted AI could never get close, but I don't see why. It seems like being able to do precise calculations and a near infinite APM could be a significant advantage.

What makes AI difficult in these games is pattern recognition of higher level strategic abstractions. A human player can recognize that placing a Go stone or a Terran command center projects some power in regions around that spot. The human can use intuitive pattern matching to assess when he has a superior force in an area and can push to a decisive tactical victory, even if the human isn't quantifying every move in precise terms. An AI must quantify every bit of power projection somehow, which becomes impossible with present computing resources in games whose possible state space quickly explodes into 10^10 or more possibilities.

It's actually a similar problem to computer vision. Identifying a battle front from the current state of a war game and recognizing the tactical possibilities is similar to edge detection in a photograph and recognizing objects. Humans do that essentially with highly parallel computations and lookups by billions of neurons. Until we get billion-core CPUs and billion-ported RAM, AIs will not have the same capability.

Source: I've done some development on AI for Civilization. It sits somewhere between Go and Starcraft in AI capabilities. Civilization is turn-based like Go, but the state space explodes far more quickly like Starcraft when you have 100 units which can each make a dozen moves in 100! different orders on a turn. (In Civ, the order on which units act each turn is extremely important, where workers lay down railroads for other units to move, or where you attack a city with artillery before the ground pounders.)

Re: Crazy Stone computer Go program defeats Ishida Yoshio 9 dan with 4 stones

#124

Post by jim-greer » Thu, Mar 27, 2014, 5:03 PM UTC

baddox wrote: post #61

jim-greer wrote: post #59

Earlier quoted context omitted.

Hmm, would you say that poker is harder if you can raise from $10 to $50.00001 instead of $50?

I don't know. I've never played poker and don't really know how it works.

It wouldn't matter at all. It's hard for me to think of any game humans play where single vs double precision would matter.

Re: Crazy Stone computer Go program defeats Ishida Yoshio 9 dan with 4 stones

#125

Post by baddox » Thu, Mar 27, 2014, 5:15 PM UTC

MichaelGG wrote: post #99

baddox wrote: post #42

Earlier quoted context omitted.

> Combinatorically, Go is considered pretty much the most complex game in existence Is that considering "game" to include only traditional turn-based board games? Presumably it would be trivial to make a video game that's vastly more complex (at least in terms of tree or state space complexity) than Go. Any real-time strategy game should quality.

Are top RTS games like Starcraft hard for AI? I know the default AI isn't strong, but is Starcraft in theory difficult to trounce humans? I spoke to an ex-pro player and he asserted AI could never get close, but I don't see why. It seems like being able to do precise calculations and a near infinite APM could be a significant advantage.

Now that I think about it, the more interesting and challenging strategic factor in Starcraft is probably the fact that it's not a perfect information game. That's probably a big part why AIs are really good at isolated parts of Starcraft (like executing build orders in the first couple minutes of the game, or micromanaging armies in controlled situations), but really bad at actually playing a normal game and winning.

Re: Crazy Stone computer Go program defeats Ishida Yoshio 9 dan with 4 stones

#126

Post by hosh » Thu, Mar 27, 2014, 5:22 PM UTC

MichaelGG wrote: post #100

hosh wrote: post #64

Earlier quoted context omitted.

There's a lot of value for a human to get good at Go, regardless of the strength of computer opponents.

What is the value of getting good at Go?

Strategic thinking. Being able to fluidly use both linear and intuitive thought processes. Being able to apply strategic concepts learned on Go (Territory vs. Influence, life-and-death, good shapes, urgency vs. big moves, move order, etc.)

Go is a game that is abstract enough, you can easily see the same kind of patterns emerging in day-to-day life. Thus, it helps hone your ability to make decisions in face of uncertainty.

For example, a classic decision: you have a startup. Google offers an acquihire deal with you. Do you take this deal and run with the money? Do you hold out for a better acquisition deal where your product might see the light of day, or do you try to realize the potential of the company on your own? This is essentially the same decision you make when you play territorial vs. influence style, that is, realizing gains now vs. potential gains later.

Each concept in Go you learn can help you be a better Go player, but the real value is in how each of those concepts help you make decisions in your life.

You might use a computer to help you analyze things, but ultimately, the entity making the decisions for your life is you, not the computer.

Re: Crazy Stone computer Go program defeats Ishida Yoshio 9 dan with 4 stones

#127

Post by thangalin » Thu, Mar 27, 2014, 6:13 PM UTC

hyperpape wrote: post #105

thangalin wrote: post #98

Earlier quoted context omitted.

In 1997, Janice Kim (1p) gave HandTalk a 25-stone handicap: she won.[0] In 2006, Crazy Stone ran on a 4 x 2-core CPU at 2.2 GHz and won gold in a tournament.[1] In 2013, the author purchased a 4 x 16-core CPU at 2.8 GHz for tournament play.[2] I would imagine that that was the hardware used for this game. For such a slow machine the Monte Carlo method has proved devastating.[3] Consider that the the author's computer…

Current Monte Carlo programs have major problems taking advantage of increased processing power, so your major assumption is flawed. I don't have a citation at hand, but there has been discussion of this on the computer go mailing list. It's not out of the question that the engines could be tweaked to improve their performance more with better/more hardware, but right now, they have disappointing results.

"Fortunately, MCTS lends itself much more easily to parallelization" and "Although MCTS can benefit from parallel implementation, this benefit is not always as large as might be hoped" as no game-tree is constructed.[3]

I think the assumption holds: we won't have to wait 15 years; in ~7 years (or fewer) we will have 9p Computer Go players.

[1] https://webdocs.cs.ualberta.ca/~mmueller/publications.html

[2] https://webdocs.cs.ualberta.ca/~mmueller/ps/2013-CG-MCTS-Go-...

[3] http://www0.cs.ucl.ac.uk/staff/D.Silver/web/Applications_fil...

Re: Crazy Stone computer Go program defeats Ishida Yoshio 9 dan with 4 stones

#128

Post by asynchrony » Thu, Mar 27, 2014, 6:15 PM UTC

gcp wrote: post #112

bluecalm wrote: post #102

Earlier quoted context omitted.

A bishop in chess is massive. I am decent amateur (2300 ELO) and I crush the strongest computers in blitz (which are way stronger than the strongest humans) with bishop odds and it's not even interesting. 4 stones is more like a pawn, massive at professional level, not much in pro vs amateur duels. While playing vs Houdini I managed only few draws (and I lose like 95% of the time) with 2 pawns odds as well (and those…

I am decent amateur (2300 ELO) and I crush the strongest computers in blitz (which are way stronger than the strongest humans) with bishop odds and it's not even interesting. Either your rating is wrong, your are not actually playing the strongest computers, you're arguing from a too small sample size, or your definition of "crush" means "I win half of the games but those that I do win I do so convincingly". 4 stones…

Where are you getting this material advantage / ELO handicap ratio?

At any rate I'm also a decent amateur (2350 ELO) and it's absolutely unsurprising that a (2200+) player easily beats very strong computers with bishop odds.

Re: Crazy Stone computer Go program defeats Ishida Yoshio 9 dan with 4 stones

#129

Post by S4M » Thu, Mar 27, 2014, 7:23 PM UTC

sanoli wrote: post #114

S4M wrote: post #94

Earlier quoted context omitted.

My FIDE rating is slightly above 1900. I think that with 2 bishops I'll beat Magnus Carlsen easily, simply because I could sacrify one bishop for some attack and still be ahead in terms of material. But with only one bishop ahead at the start of the game, I doubt it will be enough.

2 bishops is a massive advantage in chess, but I still don't think a 1900 player could beat Magnus. Maybe a 2200 player?

Well, I could easily beat Carlsen in an endgame with one bishop and one pawn up. So if I were to start a game against him with two bishops, I could develop my pieces normally and then sacrify a bishop for a pawn to throw an attack at his king (somewhere in f7 or h7 for example). Then I'll have the initiative and still a material advantage, so what could he do? I know I am not so good, but I am confident I am not gonna fall into cheap tricks.

Also, I think some part of the difference in strength between Carlsen and I lies in his massive knowledge of the openings, which will be gone if I start the game with a two bishops handicap.

Re: Crazy Stone computer Go program defeats Ishida Yoshio 9 dan with 4 stones

#130

Post by hyperpape » Thu, Mar 27, 2014, 7:49 PM UTC

gcp wrote: post #115

hyperpape wrote: post #105

Earlier quoted context omitted.

Current Monte Carlo programs have major problems taking advantage of increased processing power, so your major assumption is flawed. I don't have a citation at hand, but there has been discussion of this on the computer go mailing list. It's not out of the question that the engines could be tweaked to improve their performance more with better/more hardware, but right now, they have disappointing results.

I don't have a citation at hand, but there has been discussion of this on the computer go mailing list. You should post it. I suspect you misunderstood. Additional computing power doesn't help some problematic situations, but the overall strength still goes up nicely. There are also some problems with parallel scaling not actually improving raw performance, but this is equivalent to the speed not actually going up.

Unfortunately, I can't seem to find it. I found other people on go forums saying that Remi Coulom might have said it, but not a citation.

Also, just to be clear, the claim that I'm considering is that around 4-6 dan KGS, you start getting markedly lower payoff in Go strength for increasing the number of playouts. The big concern seems to be capturing races.