Ke Jie vs. Alpha Go - Match 1 Post Game Analysis

[@LuisYu]

Read that AlphaGo objective is to win by slightest margin and securing a winning position is more important than increasing the winning margin.

[@Ekvitarius]

Alphago apparently wants to maximize its chances of winning, even if that means winning by a narrower margin. I suppose if we gave it the goal of maximizing its score it would pay more like a human

[@MadaxeMunkeee]

@@Ekvitarius humans maximise score for two reasons - 1 is that we care about the winning margin to some extent, but 2 is because we have some uncertainty about our ability to estimate the margin correctly and we’d hate to lose because we were wrong. AlphaGo is a lot more confident than any human about its ability to calculate the winning margin accurately. I don’t think it’s actually trying to maximise the probability of winning once it has an advantage in the endgame. It just doesn’t care what it does anymore, I think. To a human it seems weird because we always try to win but AlphaGo knows it already has.

[@Anwolf123]

I do not have time to watch the games myself, please keep up these commentaries for the next matches! :) Very interesting

[@sundaygolessons]

I'll be posting right after the game tonight.

[@Trotskisty]

Super analysis. Very insightful. Thanx.

[@Ayratzay]

I thought you are dead...

[@Trotskisty]

Undead.

[@tysonliu2833]

As a Chinese i used to learn go growing up but wouldnt be able to admire its beauty until i got older. I think it's one of the best intelligent games ever invented and always think its such a shame that only few people in the west knows that let alone knowing how to play, but i think your annalist is pretty profound and can tell that you yourselft are an experienced player. Just wondering where you from and where did you learn go playing? Is there a go community or sth in the northameria?

[@lettuan1982]

12:01 Actually, many professional players think this is the highlight of the game. AlphaGo took the lead from then on. Ke Jie himself also mentioned this after the game as well.

[@sundaygolessons]

Yeah it is amazingly effective, but subtle and took some time to understand :P

[@macnolds4145]

I love this channel. Thanks!

[@sundaygolessons]

No prob!

[@okomxxi8645]

4:00 suggested variation is bad for black. He is too low (2nd line) and white gets sente.

[@gromtaj]

Great commentary. Thank you!

[@GregKulevich]

Thanks Jonathan

[@jejeakle]

What was the score because they did put translate at the end?

[@jejeakle]

Bảo Ngọc thx

[@simoncarlile5190]

I'd be very interested to watch a game where the current incarnation of AlphaGo plays a previous incarnation (such as the one from earlier this year or even last year's).

[@sundaygolessons]

I think in the presentation they said that the current version of Alpha Go could give the old version three stones handicap, which you know, IS INSANE :P

[@zetlaoupah1773]

so Ke Jie has to be 3 stones stronger than Lee Sedol to win third game , insane indeed xD

[@palfers1]

It's the "effortlessly" that's the key for me. What does it take to get AlphaGo to break a sweat??!!

[@alonamaloh]

Handicap! I am afraid AlphaGo has gotten to the point where humans don't have a chance in no-handicap games. The next event against pros should be used to determine how much handicap would make the games competitive.

[@jamesgrist1101]

all go players will be disappointed that in all the variations being played with alphago at this summit there is no handicap variation. That would have been by far the most interesting part of the summit. Everyone just has to speculate how many stones ahead AG is. I guess he could give top 9d players 3 (+2) black stone start and win most his games.

[@sundaygolessons]

I dunno if Alpha Go can give Ke Jie two stones yet. That is really a lot. But I would love to see it tried.

[@jamesgrist1101]

one reference I'm using is todays chess engines vs top GMs. Stockfish can give top players 3 move start, or a pawn, and win. Also AG2017May will be far stronger than AG2016March. Those 60 wins in 60 games, and the fact that AG didn't need a clock in his 1st game vs KJ. I'd guess AG could easily give them a stone, and 2 at the very limit of his powers. One thing I'm hoping will happen is versions of AG will be preserved and made available, so players can play AG2016March etc, including handicap. So one day we will know if AG2017May can give 2 stones to top 9dans. We'd all love to see it, hope Hassibis switches his brain on next time he organises an AG event.

[@Kasparovwannabe]

james grist Chess computers are farther ahead than GO computers, relative to human strength.

[@tzakman8697]

go seems too difficult to me even i like it ....... i play a lot of chess but this game is much more difficult... were can i find free lessons for general strategies?

[@stephanospavlou4749]

sign up on online-go.com and check out senseis.xmp.net/?PagesForBeginners

[@jejeakle]

This was alphago's first counting game right?

[@sundaygolessons]

Not entirely sure. I think Lee Sedol usually resigned, even if the score difference wasn't that great, because he knew what the end result was.

[@Blubpaule]

There were counting games when Alphago played as Master as far as I remember

[@TYUIO11111]

Alpha go doesn't care about how much it wins by it seems it cares more about winning in general whether by alot or small it only cares about winning, which is interesting in that it exposes the greed humans have within them a machine only looks at results not gains

[@ladyki9517]

Alpha plays territorial and flexible as black, and as white it plays for influence while also being flexible.... based on the games and comments, It seems a bit clear to me AlphaGo think white is better from the very beginning of the game ( I can't help but think that's because of Komi, especially the 0.5 bs )... even Lee second thought Alpha was better as white, and I heard AlphaGo wins most of its game against itself as white.

[@AneurysmXX]

SM Sai the 0.5 is to ensure the game doesn't end in a tie. it's not bs.

[@ladyki9517]

Everyone knows it's to make sure tie doesn't occur, but that's the issue... if two people of the same strength play a perfect game ( tbh only alpha is capable of that ) and the 0.5 automatically gives a win to white... it's bs.... the .5 is simply not fair... the Komi fine, but rlly .5 added? Rlly? No one say it isn't bullshit... no matter how small it may be, if AlphaGo thinks it's enough to make white have a higher probability of winning... I don't think anyone can argue the .5 is just stupid.

[@hodgeconjecture1028]

google should design alphago program that can be used by humans to play go. Namely the program telling human players, at each turn, the probability of every sequence of following every choice, and let human choose. This would be more fun, and more fair to human, regarding the number of times a machine can do in the same time compared to human.

[@spinakker14]

there's actually a program just like that - or at least how I understand it, I only started playing now it's called Leela, check it out

[@dannygjk]

er... You kept confusing me as to who was white/black.

[@Dirtfire]

AG is white.

[@kuopi0913]

nice commentary. thank you! -Taiwan bro here

[@sundaygolessons]

多谢

[@pascaldupuis2651]

I am very far to be an go expert. I understand (from others sources) that alphago won with slightest margin (half point). But, this commentator kept telling that alphago dominated the game. It would have been nice if he explained this apparent contradiction.

[@HungNguyen-kb4ib]

He did explain if you listen closely. It goes something like this: When you are already winning by a lot, you'll get to choose between moves that give you 95% chance of winning by a large margin and moves that give you 99.99999999% chance of winning by a half point. Human players would very likely go after the first choice, but AlphaGo simply chooses the second. It makes that decision at 27:20, and the rest, as the commentator said, is a one-way street.

[@pascaldupuis2651]

Thanks for your answer. The commentator should have been more precise and say the Alphago dropped intentionally some points and give an estimation of this number of points.Furthermore, how can a human can compete with a program witch can sure establish of an half point win when the best analysts were bugling hard to estimate the score at this moment of the game.

[@feanorfeuergeist]

The way to beat AlphaGO is to play the utmost retarded shit you can think of , that will confuse it because its only good aganst top level humans.

[@markh1462]

This is true cuz machine learning algo can only learn or extrapolate out of what it has seen. So, we have to find the state space regions that had not been well covered by it.

[@feanorfeuergeist]

Exactly ! Also it doesnt seem to like shape vs. shape and jumps somewhere , what if you would just continue your shape and take his stones and let him place his fancy stones that will be useful in 300 turns?

[@Tsuyara]

How do you know it is only good against top level humans? AlphaGo defeated the Nr. 1 ranked player and numerous other Pros, but you propose that it'd be that easy to defeat?

[@feanorfeuergeist]

Yes , i beat alphaGO by drawing a smiley onto the board.

[@SaveHipHop100]

feanorfeuergeist stupid theory....if you make nonsense moves AlphaGo will just continue gathering territory.

[@8DX]

Gender neutral pronouns for AlphaGo, please! They deserve as much. =8)-DX

[@ubertoaster99]

*IT* has no use for gender, surely?

[@8DX]

Which would be my point? But if you're already going to anthropomorphise the robot, implying intention, goals and human-like cognition, they/them are the neutral pronouns, rather than "it".

[@dannygjk]

You are assuming, "anthropomorphise the robot, implying intention, goals and human-like cognition". Besides, "intention, goals and human-like cognition", is partially appropriate in the case of an AI that uses neural nets. It's especially understandable when the annotator has no computer science knowledge/experience. Humans are merely used to giving an entity a gender. Using the proper, 'it', could get confusing. For one thing when you are talking about stones and positions etc. you may find yourself having to word your sentences carefully to make it clear that you are not referring to AG. I admit that myself I prefer to say white and black because it's a good habit due to it working whether the player is AI, male, or female. Also I find it's easier to look at the position objectively when thinking in terms of black and white.

[@ubertoaster99]

Using 'they' for a known individual is just awful, if that's what you're suggesting. Anything would be better, anything at all. People anthropomorphise computers all the time and don't give them a gender. "It's eaten my files", "It's playing up", "It's deliberately misinterpreting me". "It" works fine.

[@markh1462]

"Humans are merely used to giving an entity a gender" False. This is only true in european languages. Eastern languages like chinese do not even assign gender to 3rd person pronoun (there is no "he" or "she", only one gender neutral "ta"). And, thank goodness, asian languages do not assign gender to objects! That is why many asian ppl I know sometimes use she to refer to a guy instead of he, or vice versa, by mistake when they speak english. I personally think that the western way of injecting genders into everything is unnecessarily complicating and lots of times just dumb. I can't find any legit reason why we have to overcomplicate things like this.