Life In 19x19

From US. Go Congress

"After their presentation, the two took questions from the audience, many of whom wanted to know things like when an AlphaGoBot on KGS will be available and whether a strong version of the program would be available in the near future for desktops or handhelds. Most were answered cryptically with “Under discussion,” but in response to a question about how strong AlphaGo is today, Huang — who had earlier showed a graph charting improvement of one rank a month — did say that it was possible that the program could now give a top professional two stones, but that this has not yet been tested."

see http://www.usgo.org/news/2016/07/huge-a ... -congress/

Didn't some pro once say he could not be beaten if he took four stones from god? Seems like AlphaGo may be a semi-god.

I can give professional 2 stone handy too.
I may not be able to win but sure can give handycap..

AGA just edit the news and now it's not stated 'top' pro, just pro. The article is now go on like this

"Most were answered cryptically with “Under discussion,” but in response to a question about how strong AlphaGo is today, Huang — who had earlier showed a graph charting improvement of one rank a month — did say that it was possible that the program could now give a professional two stones, but that this has not yet been tested. He also said that commentaries will be released soon on all five AlphaGo-Lee Sedol games, as well as three games between AlphaGo v18 (the version that played Lee Sedol)."

DrStraw wrote:Didn't some pro once say he could not be beaten if he took four stones from god? Seems like AlphaGo may be a semi-god.

There is a difference between God in go (someone who plays perfectly) and omnipotent being.

Someone who plays perfect moves couldn't maybe win by giving 4 handicap stones since they couldn't play tricky moves or overplays since they would be losing from the start and any move they think they can play will lose the game.

Omnipotent being would know which tricky moves to play in order to confuse the human, in other words that kind of God would know when you would make a mistake.

Then again I am not a pro nor a God so I don't really know.

I can give a 360 stone handicap to a pro.

blablabla....talk is cheap. When can we see alphago play again?

Aja Huang just log in to KGS so we storm a lot of questions to him
- self play games will released next 2 weeks (building website right now)
- games are very complicated so gu li and zhou ruiyang helped a lot
- pro he's referring to (give 2 stones handicap) is andy liu

pookpooi wrote: - pro he's referring to (give 2 stones handicap) is andy liu

Meh, so that's no new information: Lee Sedol crushed Andy on 2 stones*, could probably win on 3, and AlphaGo beat Lee Sedol. So unless AlphaGo is worse at handicap games than even ones (which could be possible, as some of handicap Go skill is the psychological understanding of the kind of mistakes weaker players make and inducing them, and I expect LSD is better at this metagame than an AI) I would already expect AlphaGo to be able to beat Andy Liu on 2 stones after the match in March.

*in the go9dan match, Andy actually gave him a good game and may have been leading at some points in their Ing Cup match so maybe Andy improved since then, plays better on the board, or something.

Uberdude wrote: So unless AlphaGo is worse at handicap games than even ones (which could be possible, as some of handicap Go skill is the psychological understanding of the kind of mistakes weaker players make and inducing them, and I expect LSD is better at this metagame than an AI)

But I think this is misunderstanding what a neural net can learn.

True, being able to do well in handicap games is NOT simply determined by difference in skill level in even games. And true, up to this point a particular neural net might not (yet) have LEARNED how to play differently in a handicap game. But you are treating this as a limitation of the PROGRAM (that is implementing the neural net) as opposed to the training of the net (how much training in handicap games). You are thinking that there is something necessarily human about "learning the typical mistakes the weaker player might make in tricky situations".

In other to train a neural net, you have to be able to score it's initial (pre-trained) response right or wrong in order to train it to get it right.

Uberdude wrote:

pookpooi wrote: - pro he's referring to (give 2 stones handicap) is andy liu

Meh, so that's no new information: Lee Sedol crushed Andy on 2 stones*, could probably win on 3, and AlphaGo beat Lee Sedol. So unless AlphaGo is worse at handicap games than even ones (which could be possible, as some of handicap Go skill is the psychological understanding of the kind of mistakes weaker players make and inducing them, and I expect LSD is better at this metagame than an AI) I would already expect AlphaGo to be able to beat Andy Liu on 2 stones after the match in March.

*in the go9dan match, Andy actually gave him a good game and may have been leading at some points in their Ing Cup match so maybe Andy improved since then, plays better on the board, or something.

I believe this is the cleverness of Aja Huang to answer the question he doesn't want to answer, and can't be avoid either.
Andy Liu's rating (from goratings.org) is 3152, plus two stones (230 x 2) = 3612 and this number is roughly the AlphaGo's rating (3604 as of now). So it's the another way to say 'nothing new' but sadly USGo website got it wrong and made it to the headline that is spreading like a wildfire on the internet LOL

Mike Novack wrote:But you are treating this as a limitation of the PROGRAM (that is implementing the neural net) as opposed to the training of the net (how much training in handicap games). You are thinking that there is something necessarily human about "learning the typical mistakes the weaker player might make in tricky situations".

No I am not.

I like the Go community professional lamenting. Especially that east asian style kind of arrogance "our feng shui amd honorable traditions will beat any computer". Nope, will not.
It's exactly what the chess community went through. In a few years, maybe decade(s), that will have stopped since any human Go player will be defeated by affordable computer programs. Now we're in a transition phase.

Sneegurd wrote:I like the Go community professional lamenting.

I've seen very few in the professional community lamenting. It's mostly the amateurs.

Many pros are very interested to see how things shape up.

oren wrote:I've seen very few in the professional community lamenting. It's mostly the amateurs.

Many pros are very interested to see how things shape up.

most pro before lee sedol match think lee sedol will win, but this is just their prediction, not lamenting
but one day during the match, I roam the internet and found janice kim 3p facebook, she said she don't want AI to win because go is the queen of the game and AI will make it unpure. after alphago lost in game 4 she post demis hassabis sad face and said something like lol you said you just want to test alphago but why you're sad when it lost, this is from my memory though cause after that most of her status go private.
edit: found it https://www.facebook.com/photo.php?fbid ... =3&theater