Go has fallen to computer domination?

Discussion of anything and everything relating to chess playing software and machines.

Moderators: hgm, Dann Corbit, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
Rein Halbersma
Posts: 711
Joined: Tue May 22, 2007 9:13 am

Re: Go has fallen to computer domination?

Post by Rein Halbersma » Sun Jan 31, 2016 10:17 pm

matthewlai wrote:
Astatos wrote:So Matthew why you didn't manage to make Giraffe work?
Giraffe most certainly did work.
OK, so if you would feed the Giraffe NN-eval into a MCTS framework, would it scale to Stockfish-level peformance? (bridging around 1000 ELO IIRC).

matthewlai
Posts: 793
Joined: Sun Aug 03, 2014 2:48 am
Location: London, UK
Contact:

Re: Go has fallen to computer domination?

Post by matthewlai » Sun Jan 31, 2016 10:33 pm

Rein Halbersma wrote:
matthewlai wrote:
Astatos wrote:So Matthew why you didn't manage to make Giraffe work?
Giraffe most certainly did work.
OK, so if you would feed the Giraffe NN-eval into a MCTS framework, would it scale to Stockfish-level peformance? (bridging around 1000 ELO IIRC).
Probably not. Chess and go are different problems that require different algorithms.
Disclosure: I work for DeepMind on the AlphaZero project, but everything I say here is personal opinion and does not reflect the views of DeepMind / Alphabet.

Rein Halbersma
Posts: 711
Joined: Tue May 22, 2007 9:13 am

Re: Go has fallen to computer domination?

Post by Rein Halbersma » Sun Jan 31, 2016 10:49 pm

matthewlai wrote:
Rein Halbersma wrote:
matthewlai wrote:
Astatos wrote:So Matthew why you didn't manage to make Giraffe work?
Giraffe most certainly did work.
OK, so if you would feed the Giraffe NN-eval into a MCTS framework, would it scale to Stockfish-level peformance? (bridging around 1000 ELO IIRC).
Probably not. Chess and go are different problems that require different algorithms.
Then we have different definitions of "work". I think the question by Giorgos meant: "why didn't you manage to make Giraffe competitive with Stockfish?"

IIRC, the AlphaGo eval takes around 3ms to compute, which is about 3-4 orders of magnitude slower than a fast chess eval. An accurate eval is not enough, it needs to be fast enough as well.

matthewlai
Posts: 793
Joined: Sun Aug 03, 2014 2:48 am
Location: London, UK
Contact:

Re: Go has fallen to computer domination?

Post by matthewlai » Sun Jan 31, 2016 11:16 pm

Rein Halbersma wrote:
matthewlai wrote:
Rein Halbersma wrote:
matthewlai wrote:
Astatos wrote:So Matthew why you didn't manage to make Giraffe work?
Giraffe most certainly did work.
OK, so if you would feed the Giraffe NN-eval into a MCTS framework, would it scale to Stockfish-level peformance? (bridging around 1000 ELO IIRC).
Probably not. Chess and go are different problems that require different algorithms.
Then we have different definitions of "work". I think the question by Giorgos meant: "why didn't you manage to make Giraffe competitive with Stockfish?"

IIRC, the AlphaGo eval takes around 3ms to compute, which is about 3-4 orders of magnitude slower than a fast chess eval. An accurate eval is not enough, it needs to be fast enough as well.
Well, by "work" I meant play a reasonable game of chess. I didn't see any mentioning of Stockfish, and it seems strange to say that an engine only works if it's competitive with Stockfish. That obviously requires a lot more work than what can be done by 1 guy in 4 months, with a radically different approach, even if the approach is correct.

eval in AlphaGo is a completely different design with a completely different architecture optimized for Go and MCTS.

Giraffe's eval runs in a few microseconds.
Disclosure: I work for DeepMind on the AlphaZero project, but everything I say here is personal opinion and does not reflect the views of DeepMind / Alphabet.

bob
Posts: 20923
Joined: Mon Feb 27, 2006 6:30 pm
Location: Birmingham, AL

Re: Go has fallen to computer domination?

Post by bob » Mon Feb 01, 2016 4:30 am

syzygy wrote:
bob wrote:Nobody remembers the second computer to beat the world champion, only the first. I remain skeptical of this stuff, however, but we will see.
First was Genius. Second was Deep something, right?
No, first was deep blue. Not counting blitz games. We were rolling GMs up and spitting them out in the early 80's at blitz...

User avatar
Laskos
Posts: 10949
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: Go has fallen to computer domination?

Post by Laskos » Mon Feb 01, 2016 5:24 am

matthewlai wrote:
Rein Halbersma wrote:
matthewlai wrote:
Rein Halbersma wrote:
matthewlai wrote:
Astatos wrote:So Matthew why you didn't manage to make Giraffe work?
Giraffe most certainly did work.
OK, so if you would feed the Giraffe NN-eval into a MCTS framework, would it scale to Stockfish-level peformance? (bridging around 1000 ELO IIRC).
Probably not. Chess and go are different problems that require different algorithms.
Then we have different definitions of "work". I think the question by Giorgos meant: "why didn't you manage to make Giraffe competitive with Stockfish?"

IIRC, the AlphaGo eval takes around 3ms to compute, which is about 3-4 orders of magnitude slower than a fast chess eval. An accurate eval is not enough, it needs to be fast enough as well.
Well, by "work" I meant play a reasonable game of chess. I didn't see any mentioning of Stockfish, and it seems strange to say that an engine only works if it's competitive with Stockfish. That obviously requires a lot more work than what can be done by 1 guy in 4 months, with a radically different approach, even if the approach is correct.
From human point of view, Giraffe and AlphaGo are comparable achievements. On a 4 core i7 Giraffe plays IM level Chess. On a similar 4 core i7 with 2 GPU, AlphaGo plays Go at very strong amateur level comparable to Chess IM. In both cases, on an i7 they are in the first several thousands players in strength among millions of casual players and many thousands dedicated players, and are just below the level of Chess GM and Go Pro. AlphaGo is probably a bigger achievement only because other approaches in Go don't work well.

Rein Halbersma
Posts: 711
Joined: Tue May 22, 2007 9:13 am

Re: Go has fallen to computer domination?

Post by Rein Halbersma » Mon Feb 01, 2016 8:01 am

Laskos wrote: From human point of view, Giraffe and AlphaGo are comparable achievements. On a 4 core i7 Giraffe plays IM level Chess. On a similar 4 core i7 with 2 GPU, AlphaGo plays Go at very strong amateur level comparable to Chess IM. In both cases, on an i7 they are in the first several thousands players in strength among millions of casual players and many thousands dedicated players, and are just below the level of Chess GM and Go Pro. AlphaGo is probably a bigger achievement only because other approaches in Go don't work well.
I don't agree at all. One should always compare to the state-of-the-art. Otherwise, it's a bit like showing that a solar-powered and self-driving car is more energy efficient and better at navigating than a Formula 1 car, and then not only suggesting that the remaining order of magnitude in speed difference makes the solar car close to competitive, but also that it will easily generalize to planes and trains as well. :?

User avatar
Laskos
Posts: 10949
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: Go has fallen to computer domination?

Post by Laskos » Mon Feb 01, 2016 9:04 am

Rein Halbersma wrote:
Laskos wrote: From human point of view, Giraffe and AlphaGo are comparable achievements. On a 4 core i7 Giraffe plays IM level Chess. On a similar 4 core i7 with 2 GPU, AlphaGo plays Go at very strong amateur level comparable to Chess IM. In both cases, on an i7 they are in the first several thousands players in strength among millions of casual players and many thousands dedicated players, and are just below the level of Chess GM and Go Pro. AlphaGo is probably a bigger achievement only because other approaches in Go don't work well.
I don't agree at all. One should always compare to the state-of-the-art. Otherwise, it's a bit like showing that a solar-powered and self-driving car is more energy efficient and better at navigating than a Formula 1 car, and then not only suggesting that the remaining order of magnitude in speed difference makes the solar car close to competitive, but also that it will easily generalize to planes and trains as well. :?
Well, I don't think that comparing to Stockfish is the way to decide whether an approach to some specific AI problems like Chess works or not. Even the term AI was invented in contrast and to be compared with human intelligence. "Intelligence" seems crucial here, otherwise F1 car is also a very successful AI application, as it runs much faster than humans. The question was "why Giraffe didn't work" and it seems a weird question, in fact it is shown to work better than 99% of humans in a field associated with common language use of the term "intelligence". Saying that Giraffe didn't work because there is Stockfish state of the art is like saying that an approach to image recognition doesn't work because applied to recognizing celestial objects there are much better tools.

Werewolf
Posts: 1349
Joined: Thu Sep 18, 2008 8:24 pm

Re: Go has fallen to computer domination?

Post by Werewolf » Mon Feb 01, 2016 9:09 am

Laskos wrote:
matthewlai wrote:
Rein Halbersma wrote:
matthewlai wrote:
Rein Halbersma wrote:
matthewlai wrote:
Astatos wrote:So Matthew why you didn't manage to make Giraffe work?
Giraffe most certainly did work.
OK, so if you would feed the Giraffe NN-eval into a MCTS framework, would it scale to Stockfish-level peformance? (bridging around 1000 ELO IIRC).
Probably not. Chess and go are different problems that require different algorithms.
Then we have different definitions of "work". I think the question by Giorgos meant: "why didn't you manage to make Giraffe competitive with Stockfish?"

IIRC, the AlphaGo eval takes around 3ms to compute, which is about 3-4 orders of magnitude slower than a fast chess eval. An accurate eval is not enough, it needs to be fast enough as well.
Well, by "work" I meant play a reasonable game of chess. I didn't see any mentioning of Stockfish, and it seems strange to say that an engine only works if it's competitive with Stockfish. That obviously requires a lot more work than what can be done by 1 guy in 4 months, with a radically different approach, even if the approach is correct.
From human point of view, Giraffe and AlphaGo are comparable achievements. On a 4 core i7 Giraffe plays IM level Chess.

Giraffe is only a single core entity I think, at least for the end user.

User avatar
Laskos
Posts: 10949
Joined: Wed Jul 26, 2006 8:21 pm
Full name: Kai Laskos

Re: Go has fallen to computer domination?

Post by Laskos » Mon Feb 01, 2016 9:14 am

Werewolf wrote:
Laskos wrote: From human point of view, Giraffe and AlphaGo are comparable achievements. On a 4 core i7 Giraffe plays IM level Chess.

Giraffe is only a single core entity I think, at least for the end user.
I have run it on my i7, but I don't remember. It doesn't matter too much, I can say the same for 1 core of i7.

Post Reply