Chinese School of Chess

topazg · Post by **topazg** » Thu Jan 31, 2013 3:33 am

Please keep things on topic to Go and chess, rather than making me hungry talking about stir fries

EdLee · Post by **EdLee** » Thu Jan 31, 2013 5:01 am

burrkitty wrote:Computer programs can beat the best chess players in the world. In Go it isn't even close.

About 20 years ago (before 1996 anyway), Gary Kasparov was on Letterman and
Kasparov said he thought the computer would NEVER beat the best human players.
About 10 years ago, the computer was pretty bad at Go -- way more than 9 stones from pro -- one could say it wasn't even close.
Today the computer is at around 5 stones from pro (19x19) -- it is getting very close now.
(On 9x9, I think the computer is already near pro or at pro level.)

burrkitty · Post by **burrkitty** » Thu Jan 31, 2013 6:08 pm

Sorry! It was more about precision vs simplicity but some how cooking became the analogy we were using. My bad for making you hungry. Maybe we need a cooking thread in the off topics :p

5 stones from pro (which is what? Shodan? 3d? What changes d to p? Money and a org?) still seems like a long way to me. Comparing computing from 20 years ago to today isn't even fair. Mathematically, going from a 19x19 to a 9x9 which is the difference between a calculation that yields a absurdly huge number and one that yields less possibilities than a chess board. That's just the math! Not even a discussion on rules, just math! 5 stones is SO FAR!!!!! ((((；ﾟДﾟ)))))))

topazg · Post by **topazg** » Thu Jan 31, 2013 6:15 pm

In 2006, the best computer engine for Go I could find was around 10k. Now it's a reasonably strong KGS 5d. Even if it's still 5 stones away from professional strength, that level of improvement is still very impressive. I would be surprised if computer Go wasn't at KGS 8-9d by 2020.

burrkitty · Post by **burrkitty** » Thu Jan 31, 2013 6:23 pm

I wouldn't. That level of play gets closer and closer to needing to be a creative AI. That tech has been 10 years away for 30 years now.

topazg · Post by **topazg** » Thu Jan 31, 2013 6:39 pm

burrkitty wrote:I wouldn't. That level of play gets closer and closer to needing to be a creative AI. That tech has been 10 years away for 30 years now.

I don't know if you follow the chess world, but much the same was said of chess. The top engines didn't get there by number crunching better, they got there by having a better heuristic appreciation of a position. I don't see why the same isn't possible with Go.

EdLee · Post by **EdLee** » Fri Feb 01, 2013 4:40 am

burrkitty wrote:pro (which is what? Shodan? 3d? What changes d to p? Money and a org?)

Many people, perhaps including yourself, have some misunderstanding and confusion about this.
This seems off topic, but I'm happy to discuss this more in PM if you'd like.

SmoothOper · Post by **SmoothOper** » Fri Feb 01, 2013 7:42 am

topazg wrote:
burrkitty wrote:I wouldn't. That level of play gets closer and closer to needing to be a creative AI. That tech has been 10 years away for 30 years now.

I don't know if you follow the chess world, but much the same was said of chess. The top engines didn't get there by number crunching better, they got there by having a better heuristic appreciation of a position. I don't see why the same isn't possible with Go.

The thing is they didn't get there by shear number crunching or heuristics. They got there by storing all winning positions in things called "End Game Table-bases", I am pretty sure these don't work for Go.

skydyr · Post by **skydyr** » Fri Feb 01, 2013 7:46 am

SmoothOper wrote:
topazg wrote:
burrkitty wrote:I wouldn't. That level of play gets closer and closer to needing to be a creative AI. That tech has been 10 years away for 30 years now.

I don't know if you follow the chess world, but much the same was said of chess. The top engines didn't get there by number crunching better, they got there by having a better heuristic appreciation of a position. I don't see why the same isn't possible with Go.

The thing is they didn't get there by shear number crunching or heuristics. They got there by storing all winning positions in things called "End Game Table-bases", I am pretty sure these don't work for Go.

It's not terribly different from having pre-evaluated joseki, though it comes at a different stage of the game. One certainly might speak of playing to get an opponent or computer 'out of book'.

xed_over · Post by **xed_over** » Fri Feb 01, 2013 11:13 am

EdLee wrote:
burrkitty wrote:pro (which is what? Shodan? 3d? What changes d to p? Money and a org?)
Many people, perhaps including yourself, have some misunderstanding and confusion about this.
This seems off topic, but I'm happy to discuss this more in PM if you'd like.

why not just discuss it here so the rest of us could benefit?
don't be afraid to go off-topic. that's what healthy discussions do.

SmoothOper · Post by **SmoothOper** » Fri Feb 01, 2013 1:14 pm

skydyr wrote:
SmoothOper wrote:
The thing is they didn't get there by shear number crunching or heuristics. They got there by storing all winning positions in things called "End Game Table-bases", I am pretty sure these don't work for Go.

It's not terribly different from having pre-evaluated joseki, though it comes at a different stage of the game. One certainly might speak of playing to get an opponent or computer 'out of book'.

And in chess they have both openings and table-bases. I think the difference is that by doing the preprocessing off-line, they can eliminate much of the computation that is needed to play the game, IE they can terminate the searches early in chess. However, constantly searching through the end of the game to evaluate winning moves, is quite a burden computationally.

EdLee · Post by **EdLee** » Fri Feb 01, 2013 2:24 pm

xed_over, good question. A short answer is there are huge, nasty cans of worms in many places.

A few random samples:
Q W E R T Y U

msgreg · Post by **msgreg** » Fri Feb 01, 2013 2:40 pm

EdLee wrote:xed_over, good question. A short answer is there are huge, nasty cans of worms in many places.

A few random samples:
Q W E R T Y U

burrkitty wrote:5 stones from pro (which is what? Shodan? 3d? What changes d to p? Money and a org?)

I took the original question to mean "what is the rank equivalent between amateurs and pros". Based on my small amount of reading, it seems like the skill level of 1d-9d and 1p-9p overlap somewhat. Where is the overlap?

Here's the Sensei's page on Worldwide Rank Comparison

But it doesn't list professionals. Perhaps I mistook the question :-(

EdLee · Post by **EdLee** » Fri Feb 01, 2013 3:36 pm

burrkitty wrote:pro (which is what? Shodan? 3d? What changes d to p? Money and a org?)

msgreg wrote:I took the original question to mean...

The tone of voice was unclear to me. I could not tell whether it was rhetorical, a joke, sarcasm, or a genuine search for information.

EdLee · Post by **EdLee** » Fri Feb 01, 2013 3:39 pm

msgreg wrote:it seems like the skill level of 1d-9d and 1p-9p overlap somewhat. Where is the overlap?

This varies from country to country, and even from individual to individual.
For example, in China, some amateur 6-dans are actually ex-pros -- they quit their pro status.
In China, an amateur 7-dan is usually a national champion (in China) -- they're near or at pro level.

Life In 19x19