Alpha Go vs. Alpha Go (Game 8) - Go From Another Dimension

[@Tasermaxx]

Trying to learn Go from AlphaGo is like trying to learn colors from a rainbow. You pick a hue, and AlphaGo has millions of other hues to choose from. You'd spend ten lifetimes just trying to figure out indigo.

[@jjjccc728]

I enjoyed your commentary. The go school Yunguseng Dojang also has a great commentary on the same series It is interesting to see pro Go players like the teachers at Dojang struggle with these games. Keep up the good work.

[@ClawsNGloves]

Alphago is like water taking the form of whatever shape .

[@Borthralla]

What probably happened with the ko at the end is that black knew it was behind and realized the only way it could win would be to introduce variance. AlphaGo saw that playing simply, which is easier to read, would always lead to a loss, so it was forced to complicate the game and make it unpredictable. We haven't really seen this behavior yet (because it's never been in a losing situation against people), but it makes sense. When you're ahead, simplify the game, but if you're behind, complicate the game as much as you can to reverse it. That's probably something AlphaGo learned from playing itself.

[@DontbtmeplaysGo]

I agree: Alphago would rather have a "small" 60% chance of losing by 100 points than a huge 80% chance of losing by 1 point.

[@david203]

It would be even more impressive if A G had been taught that strategy by its programmers instead of learning it on its own. Just my opinion.

[@thom1218]

+David Spector Alpha Go is designed to learn on its own, unlike most contemporary software. A primary mission for DeepMind is to develop artificial intelligence that is able to create new practical knowledge by building on what humans already know in most cases. Go is an area where it has succeeded in doing exactly that. Its programmers designed it to search for tactics that are superior to the training data it was given, which consisted of professional human game play. Alpha go first discovers the tactics within the training data, and it then goes on to find deeper strategies to defeat those tactics. This is the product of AlphaGo, and this type of artificial intelligence strategy can be applied to many other fields of study. If programmers hard-coded any preexisting knowledge into this type of endeavor apart from training data, then that would absolutely defeat the purpose. If said knowledge was in fact known to be good, it should be considered part of the training data, which reflects the human state of the art that should be built upon or discarded by the AI's deep search.

[@dannygjk]

David Spector I am a programmer and neural nets are the future. Forget the old ways.

[@ChibiRuah]

as a double digit kyu i cant say much on what is good or bad from one as strong as alpha go. but man that 3-3 invasion so early even hurts my soul. But i guess it must be valid enough because even if its weak alpha go is strong enough to with it anyways but yea some moves seem really alien.

[@pianoforte611]

You probably don't have control over this, but it would be awesome to have the hypothetical moves in a slightly different color to distinguish them

[@CattyRayheart]

yay I was hoping for this analysis.

[@david203]

Me, too, and it's so interesting to see the strange positions that result from playing with itself. Some look like beginner's moves, yet make sense in the context of the actual direction and strategy of play.

[@Goombario37]

Wait, you wrote the, "So you want to play Go?" Series? I own all those books and I'm subscribed and I had no idea you were related to those books or had written books

[@sundaygolessons]

Yeah that's me :P

[@aditsu]

8:00 - 8:20 dude, that's EXACTLY the same result...

[@Wemdiculous]

Seems like alpha exchanged a wing for a 3-3 invade so it could get corner territory and not be sealed in. He played it in both cases only when its opponent would be willing to give up influence on one side for the other. Dont know why thats important other than getting sealed in is worse than not getting sealed in, but that is when alpha plays 3-3.

[@Wemdiculous]

Prolly because its so confident in its ability to invade without space for 2 line extensions

[@coryjay9384]

I enjoyed this project

[@dreadrath]

Even Alpha Go can't Alpha Go.

[@spanishmackerel]

Thanks for the video.

[@alwayslookingatself]

Great title

[@david203]

Why doesn't Alpha Go have a dictionary of Josekis and do good moves in the beginning game? Many of the moves here look odd and make for later difficulties that could have been avoided by playing more like people play, IMO.

[@sundaygolessons]

Actually I like Alpha Go's approach better. Joseki is only good or bad in a full board context. One joseki may be horrible in one position or perfect in another. Alpha Go tailors each individual move for the full board position. We saw an example of this in the pair Go game where the human player finished a local joseki but Alpha Go had a better, more global move.

[@david203]

Oh, yes, for sure. But when most of the board is still empty, playing a joseki really is the best sequence, according to the best go players in the world. It would not take much to add josekis, but they would prevent A Go from falling behind if the opponent decides to play one through. After the game gets going, I think I agree with you, though.

[@dwaltrip77]

David Spector the best players in the world are about 2 stones weaker than AlphaGo, so it is difficult to say if the josekis they know are still the best moves

[@steliostoulis1875]

you got handsome dude

[@yagzyalcntas553]

6/10

[@LapahnYT]

WutFace voice

[@jkim1967]

4 dan trying to understand alpha go is not right. i will understand if you are at least 7 dan.

[@jkim1967]

your variations are typical of low dan.... i will view other video.

[@minghei2010]

爸爸抓狂了 我們有領樹苗 但是 還沒有種完 因為后里今天才有水進來 我昨天還花光了身上的錢 買種子 但是我要拿樹苗去種在后里 但是爸爸不高興 my father is crazy , we have got tree babies, but i havent finished growing them because Holi (后里) today the water will come into the ditch i spent all of the money (90 NT) to buy seeds but i want to grow tree babies in HoULi (后里 but my father was not happy about this , and i dont know why