mersenneforum.org Göogle's AlphaGo AI is playing Lee Sedol (~ the best human Go player)
 Register FAQ Search Today's Posts Mark Forums Read

 2016-03-10, 03:04 #1 Dubslow Basketry That Evening!     "Bunslow the Bold" Jun 2011 40
 2016-03-10, 08:31 #2 LaurV Romulan Interpreter     Jun 2011 Thailand 863610 Posts Darn! 2-0. Very nice match. That mid-game, wonderful! Watched it live. Thanks for sharing the links! (I didn't know about the event)
 2016-03-12, 12:39 #3 axn     Jun 2003 26×73 Posts And now, 3-0.
2016-03-12, 17:42   #4
MooMoo2

Aug 2010

22·3·47 Posts

Wow, things changed a lot over the past two years:

http://www.wired.com/2014/05/the-world-of-computer-go/
Quote:
 "There is chess in the western world, but Go is incomparably more subtle and intellectual,” says South Korean Lee Sedol, perhaps the greatest living Go player
Quote:
 I ask Coulom when a machine will win without a handicap. “I think maybe ten years,” he says. “But I do not like to make predictions.” His caveat is a wise one. In 2007, Deep Blue’s chief engineer, Feng-Hsiung Hsu, said much the same thing. Hsu also favored alpha-beta search over Monte Carlo techniques in Go programs, speculating that the latter “won’t play a significant role in creating a machine that can top the best human players.” Even with Monte Carlo, another ten years may prove too optimistic. And while programmers are virtually unanimous in saying computers will eventually top the humans, many in the Go community are skeptical. “The question of whether they’ll get there is an open one,” says Will Lockhart, director of the Go documentary The Surrounding Game. “Those who are familiar with just how strong professionals really are, they’re not so sure.”
Quote:
 I am a fairly strong Go player (5 Dan). I am actually in the process of writing a go playing AI myself. I have spoken with the worlds top minds in the field. I can say without hesitation that the Pseudo Monte Carlo routine, is just a fraction of what makes a strong computer Go player. I am an electrical engineer and control systems are my life. I am pouring everything I am, morning, day, night, into solving this exact problem. My personal belief is that we are a VERY long way from beating the top humans. I play even against the strongest computers in the world, and a professional player is simply leagues and leagues above me where I can't even fathom the depth of their reading and insights. I am standing exactly from the computers vantage, and I can tell you, there is a long long long way to the top. For comparison, the difference between me and a 9p is greater than the difference between me and a beginner. It's not even half way. It is absurd how good the human brain is at this game. The game get's really abstract at the highest level - and the best computer programs like Aya, ZenBS, CrazyStone, or manyFaces only use Pseudo Monte Carlos for move generation. Move generation is nothing compared to positional assessment. That is where the magic is. And for every way that you can write a program to have positional awareness, a human can just out-abstract it. It's really unbelievable how adaptive we are.
Quote:
 I am willing to offer this guy a prop bet: I think the best computer can beat any go player by the end of 2025. I'll bet any amount up to $10k (responding to the guy above) Quote:  I'm not the guy in question, but I'm willing to take that bet. I can place the amount in escrow, if you're actually serious about it. About 7% return on an investment is pretty good, even if it's only$10k. The key to realize is the way that Go will get better isn't raw CPU power, that can only help up to a point, because the branching factor is so high. The key is we need better understanding of the game and game theory in general on a fundamental level, and that's a software and human creativity problem, which follows a much slower curve than Moore's law. It will definitely happen, eventually, but I don't think 10 years is even remotely realistic.
Quote:
 Well, sure by the 2025 computers will be able to compete against some professionals and maybe occasionally take down a win from a strong one. However, to reach a point where it can beat "any go player" will be far, far away in the future.
From 1997:
http://www.nytimes.com/1997/07/29/sc....html?emc=eta1
Quote:
 ''It may be a hundred years before a computer beats humans at Go -- maybe even longer,'' said Dr. Piet Hut, an astrophysicist at the Institute for Advanced Study in Princeton, N.J., and a fan of the game. ''If a reasonably intelligent person learned to play Go, in a few months he could beat all existing computer programs. You don't have to be a Kasparov.'' When or if a computer defeats a human Go champion, it will be a sign that artificial intelligence is truly beginning to become as good as the real thing.
Quote:
 winning the \$1.4 million prize promised by the Ing foundation to a program that beats a human champion may be an impossible dream. The offer expires in the year 2000. Go programmers are hoping it will be extended for another century or two.

2016-03-13, 22:02   #5
ATH
Einyen

Dec 2003
Denmark

2·32·7·23 Posts

Anyone saw any stats on DeepMind in terms of TLOPs and amount of RAM?

http://www.nature.com/nature/journal...61.html#access

It is behind a pay wall but the tables below can be seen and in table 6, 8, 10, 11 shows that the "single-machine" version goes up to 48 CPUs and 8 GPUs (and 40 search threads), while the distributed version, which is the one playing Lee Sedol, goes up to 1920 CPUs 280 GPUs (and 64 search threads).

According to this it is probably that version they are using, I wonder what the TFLOPs throughput is.
http://www.theverge.com/2016/3/9/111...ligence-impact
Quote:
 DeepMind founder Demis Hassabis says that although AlphaGo has improved since beating Fan Hui in October, it’s using roughly the same computing power for the Lee Se-dol matches, having already hit a point of diminishing returns in that regard.

2016-03-13, 22:40   #6
R. Gerbicz

"Robert Gerbicz"
Oct 2005
Hungary

25478 Posts

Quote:
 Originally Posted by ATH According to this it is probably that version they are using, I wonder what the TFLOPs throughput is.
Yes, they are using that, see for example: https://en.wikipedia.org/wiki/AlphaGo "The version of AlphaGo playing against Lee uses 1,920 CPUs and 280 GPUs.[20]"

 2016-03-14, 02:26 #7 LaurV Romulan Interpreter     Jun 2011 Thailand 22·17·127 Posts Cheaters hehe Joking apart, that second game is the brightest game I ever saw, as much as my knowledge allows me to understand the game... (I used to play somewhere between 3 kyu and 2 dan, depending on luck ) Last fiddled with by LaurV on 2016-03-14 at 02:28
 2016-03-15, 09:38 #8 Dubslow Basketry That Evening!     "Bunslow the Bold" Jun 2011 40
2016-06-02, 02:50   #9
ewmayer
2ω=0

Sep 2002
República de California

7×23×71 Posts

An 'energetic coda' to this story:

Another Way of Looking at Lee Sedol vs AlphaGo | Jacques Mattheij
Quote:
 Lee Sedol used about 20 Watts of power to operate. By contrast, AlphaGo runs on a whopping 1920 CPUs and another 280 GPUs for an estimated power consumption of approximately 1 MW (200 W per CPU and 200 W per GPU). That’s 50,000 times as much power as the amount of power that Lee Sedol’s brain uses ... So now the interesting question (to me at least) is: How long before a computer will beat the human Go world champion using no more power than the human.
Sure, but can the human mine bitcoins in his spare time? Uh-huh - I didn't think so.

We may be able to get some idea of the 'when' by looking at the power curve for computer chess programs from Deep Blue to the present - how close is a 20W power-efficient compute rig running a state of the art chess app to playing at world champ level?

2016-06-02, 03:49   #10
LaurV
Romulan Interpreter

Jun 2011
Thailand

22·17·127 Posts

Quote:
 Originally Posted by ewmayer An 'energetic coda' to this story:
Correct! I mean, incorrect, they didn't count the energy used by the humans operating the computer's team, I saw at least 2 guys there including the one who was placing the pieces on the board, his (their) brains also consumed energy.

2016-06-02, 03:57   #11
retina
Undefined

"The unspeakable one"
Jun 2006
My evil lair

160616 Posts

Quote:
 Originally Posted by LaurV Correct! I mean, incorrect, they didn't count the energy used by the humans operating the computer's team, I saw at least 2 guys there including the one who was placing the pieces on the board, his (their) brains also consumed energy.
Since we are being pedantic, they also didn't include the wattage from Lee when he moves his arms, beats his heart, breathes in/out air, etc.

It would be possible to make a robotic arm to move the pieces and connect it to a compute "brain" and still come out using less total energy than one human. The robot arm could average less than one watt, leaving about 100 to 200 watts energy budget for the compute engine.

 Similar Threads Thread Thread Starter Forum Replies Last Post Orgasmic Troll Puzzles 31 2012-09-12 21:47 Fusion_power Lounge 5 2010-02-08 09:26 tha Science & Technology 0 2008-03-02 16:11 ewmayer Programming 2 2006-02-04 19:22 drew Lounge 10 2006-01-19 05:25

All times are UTC. The time now is 13:46.

Sat Aug 15 13:46:02 UTC 2020 up 2 days, 10:21, 1 user, load averages: 1.50, 1.65, 1.63