Sighris (sighris) wrote in tessier_ashpool,
Sighris
sighris
tessier_ashpool

  • Location:
  • Mood:

AI and the game of Go

Some interesting news, Andrew Okun just reported:
MyungWan Kim, 8p, lost a 9-stone handicap game by 1.5 points to an 800-processor core European supercomputer running MoGo. {For those of you who don't know the ranking system, 8p means 8-Dan Black-Belt Pro, so one of the best in the world, and the 9-stone handy-cap given to the computer is a HUGE advantage, so I was hoping it would do better... but it did win... sort of...}

David Doshay, a California computer go researcher, said the win was significant not only in itself, but because the Monte Carlo method-based MoGo software showed strong evidence that its strength is dependent on the time and computing resources available to it, meaning it will get stronger over the years as computers get more powerful, without the need for additional coding.

Footnotes:
1) For an earlier IEEE article on AI, Chess and the game of Go, see:
http://community.livejournal.com/ai_research/8253.html
2) Andrew Okun will be joining the AGA Board of Directors in September.
AGA = American Go Assoc.
  • Post a new comment

    Error

    Anonymous comments are disabled in this journal

    default userpic
  • 19 comments
It's a bit hard to swallow that heavy computing brute-force methods are the way forward. I don't doubt the result, I had just hoped that Go-research would have advanced by making a breakthrough in pattern recognition or something. :(

On the other hand, an 800-processor machine is massively parallell, which could be taken as showing that a more "brain-like" interconnected architecture is the way forward. :))

Then again, monte carlo in parallell is parallell but not "interconnected" in the sense of a recurrent neural network. :((
I feel the same way as you...
- I had expected more from the AI community than this.

But still... this shows some progress... but I wonder if it is progress in the "correct" direction!?!

Anyways, there is more here:
COMPUTER BEATS PRO AT U.S. GO CONGRESS: (8/7/2008 5P PST) In a historic achievement, the MoGo computer program defeated Myungwan Kim 8P by 1.5 points in a 9-stone game...
http://www.usgo.org/index.php?%23_id=4602
- also see the below comments by LJ user Krayle.

Namaste,
Sighris
- who is hoping for the sigularity... but not holdin his breath! ;)


Remember that a nine stone handicap in a game of Go makes both players play in a way vastly different from how they would play an even game. There's even a saying in the Go community that "the only thing you can learn from playing handicapped games is how to play handicapped games".
In my opinion, this MoGo computer is about 1d (an amateur first dan) while the usual Go programs are about 5-6 kyu now, so they're not very far behind. A talented player can reach the 1d level in something about three years, some even faster. It's the path from 1d to 8p what takes the rest of the life ;)
Actually, this MoGo computer is above 1d, MyungWan Kim is quoted to having said that it was about 3-dan, sometimes making 5-dan moves. A 1-dan would, mathematically, get a 9 stone handicap from a 10-dan amateur, which would be about 7 or so stones below an 8-dan pro. Remember that 1-dan pro is one rank above 6-dan amateur by Japanese rating, or 9-dan amateur by AGA rating.

Also, the computer has a really annoying endgame strategy that should be worked on. It fills in its own territory until it's winning by 1.5. It was actually winning by more than 5 until it started doing that.

You actually learn a lot from playing handicap games, just not joseki or territorial balance. You learn a lot of fighting; attacking, defending, invading, etc. Besides, handicapped games are for playing a more "fair" game, not necessarily learning. You do indeed miss out on joseki, opening, etc.
I would like to see it playing an equal game with a 3d then. From my experience, the more handi stones you get, the bigger difference in the style of the game when compared with an equal one. With nine handi you just have to be good at defense, you don't have to know almost anything about attacking, whereas your opponent has to attack as mad ignoring many potentially insecure situations. I think nine handi stones is far too much to judge the level of the player. I twists the style of the game too much. Three handi stones, four, maybe five, but not nine.

Hey, is that Fujiwara-no-Sai from Hikaru no Go on your avatar? :)
Er ... yeah, it is >.> Added it a long time ago figuring no one would get it ^_^;

I'm just saying that MyungWan Kim said it was 3-5 dan, but you're right; I can't learn much about someone's style from a 9H game, but I'm not a Korean 8p, though he obviously wasn't judging it's fuseki/joseki skill, since those don't really work in handicap games. All he can judge is the fighting (which I've noticed are the strong points of Go-AI, fights can be brute-forced, for the most part; judging which joseki to use to get a better whole board position can't be).
Pointless. So a computer able to play Go is considered to be intelligent?
It's been half a century since people are trying to build true AI and they are still clueless.
Could not even reverse engineer a brain of the simplest fly.
In the long run, yeah. I agree. All Deep Blue accomplished was killing Chess for humans.
It didn't "kill chess"... tens of thousands of humans still play Chess, can learn from it or at least use it as a mental exercise &/or mental combat!

But this is about coming up with more powerful computational algorithms for a task which humans find challenging... and it is an early sign of things to come (if Ray Kurzweil is even half right).

http://www.wired.com/medtech/drugs/magazine/16-04/ff_kurzweil

Interesting times are ahead!
Sighris
p.s. I friended you, please do the same!
This wasn't about computational algorithms. It used brute force, so this was just about the power of an "800-processor core European supercomputer".

...Just like Deep Blue. That's where computers have us beat. Voluminous immediately accessible comparatively infallible memory.
Do you know anything about the SlugGo program by David Doshay? - mainly I want to know if it is a more "intelligent" program (which was run on a less powerful computer at a previous Go Tourny). I played against it and although I didn't get to finish my game (because Bowen needed to leave) I thought it played at ~about~ a 9k level... but with a more powerful computer (and a few tweaks?) maybe ( I am hoping ) it could play at a much higher level?

Let me know if you know anything, because I am curious.
Sighris
> Pointless.

I disagree...

> So a computer able to play Go is considered to be intelligent?

No... but...

> It's been half a century since people are trying to build
> true AI and they are still clueless.

Yes, progress has been VERY slow... but...

> Could not even reverse engineer a brain of the simplest fly.

Well, a fly's brain is not so simple!
- Give the human's some more time, they will create better computers and programs, and maybe someday something more interesting will show up.

For example I find the "evolving" programs which have been making a lot of money on the stock market interesting!
Sighris

You wrote "[They can] not even reverse engineer a brain of the simplest fly."

You might find this interesting:
http://www.popsci.com/future-human/article/2008-07/mapping-human-mind

A 150 pound human against a parallel computer system that probably weighs 10-20 tons.

Where's the equality?
Who said anything about "equality"???
- this is just about a limited type of "computing" (playing Go)!

But it has signs for the future!
Sighris
It just seems misleading for people to suggest computers can match human intellect at a game when the human is playing against a computer the size of a building.

They should measure how much a human brain weighs with its respective support in the form of arteries, heart, etc needed to keep it functioning.

Then, they should impose a limitation as to the maximum size of the computer the human is playing against. For example, say it is found that a bare minimum of 50 pounds of brain tissue and support systems are required to play chess....

Impose a limit of 10 times that weight the computer team must adhere to.

So, the computer unit--including power supply can't weigh more than 500 lbs.

That would be the start of leveling the field in these types of challenges, I think.

It just seems ridiculous to me. Would people take anyone seriously if they bragged that a truck weighing 40 tons was able to carry more weight, faster, than a human being?

Its ridiculous.

Level the field so people realize the truth as to where technology stands in comparison to biological systems.
You asked:
> "Would people take anyone seriously if they bragged
> that a truck weighing 40 tons was able to carry more
> weight, faster, than a human being?

uhmmm... Yes... if it was the first truck and up until that point everyone had been carrying everything on their backs!

Again, who said anything about "equality"?

This is just one small step in computing; but another "milestone" has been crossed (sort of, as I wrote above: I had expected more from the AI community than this... but still it shows some progress... I just hope it is progress in the "correct" direction!) now I hope we can get to a program which can play a Pro-Dan level player WITH-OUT needing ANY handi-cap stones!

Again, who said anything about "equality"?

Man against machine challenges--challenges in general, require a degree of equality to be worthwhile, do they not?

This is just one small step in computing; but another "milestone" has been crossed (sort of, as I wrote above: I had expected more from the AI community than this... but still it shows some progress... I just hope it is progress in the "correct" direction!) now I hope we can get to a program which can play a Pro-Dan level player WITH-OUT needing ANY handi-cap stones!

That's just it.

The only "progress" that has been made is in the form of hardware.

Brute force implementations are hardly bragworthy.

Especially considering that they're likely to be billions of times less efficient than the biological systems some researchers would claim they are claiming "victory" against.

I would rate this as a defeat... Of course, that's just my personal opinion.

XD.
>> Again, who said anything about "equality"?
>
> Man against machine challenges -- challenges in general,
> require a degree of equality to be worthwhile, do they not?

No.

I guess I am not concerned about something you are concerned about. To me this is not a contest where I am concerned about "fairness".

Let me give you an example of where I am coming from, I am interested in how fast we humans can get a motorcycles and jets to go (NASA's X-43A was designed to operate at speeds greater than Mach 7) and I do not care if the motorcycle weighs more than a jaguar (the animal, not the car) or the jet weighs more than the fastest bird (a falcon I think).

Now of course I would also like the computer (running the AI) to be small, light and cheap too, so that I could afford one... but I know we need to take this one step at a time.

BTW, I agree with you that "brute force" is not the best way to develop an AI computer program to play Go. You might find an earlier IEEE article on "AI, Chess and the game of Go" interesting, see what I posted here:
http://community.livejournal.com/ai_research/8253.html