AlphaGo vs Lee Sedol: Game 5 – Video Commentary

The comments below are from the discussion during the live stream.

Related Articles


  1. Younggil An says:

    Hello everyone! Today we’re going to watch the final game of the match between Lee Sedol and AlphaGo.

    AlphaGo is winning 3–1, so it has already won the five game match, but Lee Sedol is still going to do his best and showed that AlphaGo can be defeated in game 4.

    Lee Sedol fans may be interested in our brand new book Relentless: Lee Sedol vs Gu Li which is now available for pre-order. You can download two free chapters here.

    If you have any questions about anything just let me know!

    • David Ormerod says:

      Match report coming soon.

      Full game record:



  2. Apparently AlphaGo lets its users know what its confidence of winning is — they should display its current confidence on the feed.

  3. William Chang says:

    As a former practitioner of another “impossible task”, natural language web search, I would like to piece together my few observations here and on another forum Boardgamegeek interpreting the strategies and reversal that has happened in this match, together with a few proposals for further research in game AI. I don’t want to interrupt game 5 happenings so will add to the end of game 4 comments.
    — William Chang (former Baidu chief scientist, Infoseek chief architect and CTO)

  4. Younggil An says:

    Thanks Flaum and William for your interesting comments.

  5. Excited to see how this turns out. What sort of game do you expect from Lee Sedol, Younggil?

    • Younggil An says:

      Hi Wasi, I think Lee will try to lead for a fighting game today. Or he will try to make the game complicated.

  6. How about first move at tengen?

    • Younggil An says:

      That wouldn’t be a serious game, so Lee doesn’t do. That would be very interesting though.

  7. Lee Sedol will win today 🙂

  8. Younggil An says:

    Lee chose two 3-4 moves, because it’s better to lead for a complicated opening.

  9. The beginning of an amashi strategy?

  10. Younggil An says:

    Black chose a normal territorial opening with the attachment at Q15 (7).

  11. I just wondering what if Lee try to re-create the game steps in match 4 ? will AG reacts in the same way ? and then lose again ? 🙂

  12. Younggil An says:

    White’s attachment at Q5 (12) is cutting edge.

  13. Hi,Younggil thanks for your great commentary.
    Ι want a win for Lee because no matter if all say that Alphago has
    no actual style perhaps the Τop Pros can adjust their game against it.

  14. Younggil An says:

    Black’s invasion on the right side at Q12 (17) is very recently researched move.

  15. Younggil An says:

    White’s atari at O5 (18) and Black’s push up at P12 were fighting spirit, and the opening looks interesting.

    I’ve never seen this opening before.

  16. Max Goldman says:

    Hi Younggil, thank you for your excellent commentary on the last 4 games. How do you think the opening so far? Do you think AlphaGo will deploy a more aggressive strategy after the last defeat? Thanks.

    • Younggil An says:

      Hi Max, the opening is even so far, and I don’t think AlphaGo will deploy more aggressive style because that’s what Lee prefers.

      By the way, cutting at O13 (21) was strong, and White’s extension was calm.

  17. “It may be an illusion, but I’m getting the impression that I know what’s going on.” – Michael Redmond. Ha!

  18. Younggil An says:

    The result from the first fighting up to Black R14 (25) seems to be playable for both sides.

    Black’s territory and White’s influence over the center are well balanced.

  19. Younggil An says:

    White’s pincer at C8 (30) was a good choice, because the left side and the center are more promising for White than the top left corner.

  20. Max Goldman says:

    Hi Younggil, from match 1 til now, do you have the same feeling as Lee that AlphaGo seems to be stronger when playing white counters?

    • Jorik Mandemaker says:

      It would be interesting to know what the winrate for white is when AlphaGo is playing herself. That might tell us something about if it consideres 7.5 komi to be a lot or not.

      • Younggil An says:

        Yes, right. I agree with you, and many other pros are interested in to see the data in the near future.

    • Younggil An says:

      Yes, AlphaGo seems to be playing more smoothly with White.

      I think that’s because AlphaGo thought the komi (7.5) is big.

  21. Anonymous says:

    Fightin Lee.

  22. Younggil An says:

    Black’s attachment at D8 (31) was to reduce White’s influence over the center, and the result up to B8 (39) seems to be still playable for both.

  23. Younggil An says:

    White’s cut at M14 (48) was well timed, and Black’s atari was necessary.

    White’s hane at R13 was a good followup, and the game becomes more interesting.

  24. Younggil An says:

    It seems like Black should block at R12 next, but then White will sacrifice the stones to make her influence more powerful.

    The game is still well balanced.

  25. Younggil An says:

    White keeps on playing in the bottom right corner, and let’s what AlphaGo’s plan is. It doesn’t seem to be easy to make something happen over there…

  26. Younggil An says:

    Oh, maybe yes. It’s quite early in the game though…

  27. Kallisti XF says:

    All the commentators seem baffled about what AlphaGo is trying to accomplish in the bottom right 🙂

  28. Younggil An says:

    White’s attachment and cut from N3 (60) is a nice tesuji.

    However, White’s earlier exchanges from the bottom right are bad, so the game is slightly better for Black.

  29. Does the lower right look like an amateurish mistake by AlphaGo, Mr. An? Too bad if that is the way Lee Sedol wins!

    • Younggil An says:

      I’m not sure, but it looks like AlphaGo misread something in that area.

      White’s thickness is quite powerful, so it’s still quite difficult for Black to maintain his lead.

      • Harald K says:

        Demis Hassabis, who has access to DeepMind’s win probability estimates, said on twitter that it made a bad mistake early in the game. So it looks like it indeed was a mistake, and not some diabolical long-term plan.

  30. Younggil An says:

    Black has several options now for 69.

    Reducing White’s influence at the top around G4 is a good idea, or playing in the bottom left is also quite big.

  31. Younggil An says:

    White’s cap at G6 (70) is strong, but risky.

    AlphaGo didn’t play this sort of risky move in previous games, so I guess she thought White is behind now.

  32. William Chang says:

    I’m not getting real-time but it seems to me (a lay person) if Lee gets a stone or two to around H11 first he would win the connection fight and the game.

    • Younggil An says:

      That’s an interesting idea. Let’s see if Black plays there later and win the game as you wish!

  33. Seems like it’s the old story with computers. They do some really smart things and some really dumb things. Of course they may look at us and say the same thing!

  34. Jim Cochrane says:

    It would be instructive to me if the commentators would discuss the moves as they are made in the game instead of talking about the previous situations. Returning to the game and discussing the thinking about the upcoming prospects would be more useful. Often Michael reverts to game and plays several moves to catch up and we miss it

  35. Krystefer says:

    Yes, I think Michael Redmond is doing a great job explaining to the uninitiated as well as the good players. Garlock is a huge distraction and disappointment though.

  36. Anonymous says:

    who is leading guys?

    • Younggil An says:

      Black J2 (81) was questionable, and the game becomes nearly even I think. Black was ahead in the opening though…

  37. Younggil An says:

    White sealed Black in, but Black is happy to live at the top without any trouble or loss.

    Black seems to be in the lead, but AlphaGo might have thought it’s playable for her.

  38. Younggil An says:

    Black has around 60 points on the board.
    Bottom right: 35, top right: 12, left side: 9, top: 4

    White’s center area is big, but its territory isn’t yet completed.

  39. Younggil An says:

    Black G8 (85) was to exchange before living at the top.

    Black H3 (89) was necessary, and the game is still favorable for Black, because there might be a bad aji from cutting at J7.

  40. Anonymous says:

    Hope the google team didn’t change the value calculation method to fix last game’s mistake…

  41. Younggil An says:

    White’s attachment at H2 (92) was well timed probe, and Black has to live or resist with a ko.

    Let’s see how it’s going…

  42. Younggil An says:

    Since Black has to live so small, the game doesn’t look good for Black…

  43. It’s unlikely that the error in Game 4 will come into play again today. That was a black swan, an edge case that was highly unexpected and extremely unlikely to reoccur. I doubt it’ll play a role in deciding today’s game.

    • Younggil An says:

      I agree with you. It’s not easy to find unexpected good move in this sort of peaceful game.

  44. Younggil An says:

    It looks as if Alphago thought she’s winning now.

    White I17 (100) is very solid and calm. The game was simplified, which is bad news for Lee…

    • Anonymous says:

      I like how lee is playing at the bottom right. To human eye, B is winning the capturing race but because it is not yet complete. Computer will constantly give some false positive chances of winning. This kind of thing can add up to a miracle like in game 4!!

  45. Younggil An says:

    Black’s moves in the center were active and annoying, but White didn’t care, and took a big place at C14 (106).

    White seems to be leading in terms of territory.

  46. Younggil An says:

    There’s still bad aji in the bottom left corner, but the timing to play there is the issue for Black.

  47. Younggil An says:

    Attaching at D17 (107) and F17 is a good combination.

    Black tries to reduce White’s territory, and White has two good options at E17 or F16.

  48. Younggil An says:

    White could connect underneath with F18, and the game would be favorable for White.

    However, White pushes up at G16, and that’s even stronger.

  49. Younggil An says:

    It’s not easy for Black to settle, but attaching at D15 (113) appeared to be a nice tesuji.

  50. Younggil An says:

    Black’s invaded stones are connected with I18 (125), but White is going to cut Black’s center stones off soon to attack.

    The game is still favorable for White, but if Black save the center group after White cuts, the game will be close.

  51. Younggil An says:

    White’s I11 (136) was the vital point of Black’s shape, and Black should consider of sacrificing some of his center stones.

    White is still ahead, but not by a big margin.

  52. Who’s winning so far? sorry cant play the stream..

  53. Wow, so interesting now

  54. David Ormerod says:

    Game record for people watching from behind firewalls:



  55. William Chang says:

    Don’t have current board but Lee managed to play F8; still losing the connection fight but maybe ko-fight coming? Too late?

    • Younggil An says:

      Ko fight wouldn’t come soon, and White is in the lead.

      However, the game seems to be quite close.

  56. William Chang says:

    No he didn’t play F8, sacrificing as Younggil suggested.

  57. Younggil An says:

    White’s attach in the center from I11 (136) wasn’t that efficient, and White didn’t make any big points up to I13 (144).

    Black’s atari at E8 is a nice endgame move, and the game seems to be getting more closer.

  58. Younggil An says:

    Black started to reduce White’s right center area with M10 (147), and the game becomes nearly even I think.

  59. Younggil An says:

    Lee Sedol is in byoyomi at the move of 151.

  60. The game seems to be concluded. Who is having more territory?

  61. Anonymous says:

    What’s the chance for Black to revert White’s lead?

    • Younggil An says:

      I’m not sure, but it’s actually quite close, so Black still has a chance to reverse depend on the endgame.

  62. Anonymous says:

    Thanks. Really hope Lee can win the last game.

  63. Anonymous says:

    Thanks. Really hope Lee can win the last match in this game.

  64. Which one would be better for black to play? M11 or M8?

  65. Another strange move by AlphaGo?

  66. White is little ahead but Black has better prospect for endgame

  67. Younggil An says:

    Black’s push at B12 (167) seems to be a mistake, and it’s hard to find a good move to reduce White’s territory in the left center…

  68. Younggil An says:

    Oh, Black’s peep at D13 is bold, and there will be a trade I think.

    However, the trade will be favorable for White…

  69. Son Pham says:

    I would like to have Mike Redmond play a game with AlphaGo and do commentary as we go lol. He is so good

  70. Younggil An says:

    Black will lose some points unfortunately, and the game will become clearly better for White.

  71. So, the trade is really a mistake, right? How many points does black lose for this trade?

    • Younggil An says:

      I’m not sure, but it looks like 1 to 2 points. Actually, that’s not really a bad trade, and the game is still quite close.

      White is still ahead by B3 (186).

  72. Surprising to see AlphaGo wasting ko threats, no?

  73. Younggil An says:

    White is winning by 2~3 points, but there’s nowhere for Lee to catch up against AlphaGo.

  74. Anonymous says:

    Who is white because it’s impossible for me to see streaming? thx

  75. Anonymous says:

    Alphago is white

  76. Why can’t black play G11 instead of G12?

  77. Younggil An says:

    Black could play at G11, but then White can play at G12 to make some points.

  78. Younggil An says:

    AlphaGo (White)’s endgame is perfect so far.

  79. Hope Lee can win the game. If not, at least we have a very close game at the last day of the match. Really enjoying this event so far. Will Alpha Go play with other top pro like Ke Jie in the future?

    • Younggil An says:

      I have no idea about that, but probably not since AlphaGo already defeated Lee Sedol.

      • Its too bad that Ke Jie wasn’t chosen instead. Consider he is the better player now and Lee won once and lost narrowly once, I wonder if he could have done slightly better. Probably not 5-0 or 4-1 but 3-2 or 2-3. Also others have commented that Lee Sedol’s strengths plays into Alphago’s strength which reduces Lee’s effectively severely.

  80. Younggil An says:

    Thank you for watching the match with us everyone.

    It looks like AlphaGo has won game 5, so I’m going to start reviewing the game now and writing my preliminary analysis for our article.

    I’ll be posting more commentary of the games in the coming week.

    • Thank you so much Younggil. Looking forward to your article 🙂 It’s 2am here already..

    • Huining Cao says:

      Should Komi be 7 now most people think 7.5 is too high.the fair Komi should be an odd number and 5 too low, 9 too high. It’s ok to have a draw

  81. William Chang says:

    Well, he didn’t make a “Hex move” near H11 as I had hoped, lost the connection fight there at the left-center and could not recover.

  82. Andrew Timothy says:

    I hope we can see the game until the very end where they calculate the points! All previous games ended by resignation.

  83. Krystefer says:

    Its not over yet. Lee Sedol may be behind but he has not lost!

  84. William Chang says:

    Thank you Younggil for putting up with amateurs.

  85. A very close game and a very narrow loss for Lee.

  86. Krystefer says:

    Always happy to hear amateurs like William speak up. Helps Younggil explain the game to people who don’t understand the game! Thanks both!

  87. Thanks for the excellent feedback and commentary as always.

  88. I found this game the most interesting of the 5. It was an interesting strategic contest.

  89. William Chang says:

    I’m adding a few research proposals to the end of Game 4 comments, and maybe here as well.

  90. Anonymous says:

    who won guys?

  91. Anonymous says:

    Younggil said Alphago has won, but the game is not finish and the result will be very close.

  92. Anonymous says:

    Look like AlphaGo is still in lead by less than 3 pts

  93. Anonymous says:

    Aphago won !!!!

  94. Younggil An says:

    Lee Sedol resigned, and AlphaGo won the match by 4-1.

  95. Anonymous says:

    Lee Sedol resigned

  96. Robert B says:

    Should be White + 2.5 points

  97. This tells me that Lee Sedol’s assessment is correct. It is harder to play as black with 7.5 komi for either side.

  98. BobbyBriggs says:

    Lee Sedol can think it’s a miracle he won the fourth match…

  99. Anonymous says:

    @David Omerod: can you provide the actual board as before? Cheers

  100. If I were to evaluate Alphago v18’s strength, it is equivalent of allowing human playing black with komi of zero.

  101. I wonder if Lee’s mistake is in playing too conservatively once he had a good victory in the lower right and tried to just maintain that lead but Alphago just kept chipping away.

    If there is one thing I learn is that you can’t beat Alphago by out “slow-playing” it. That’s just a recipe for lost.

  102. Was T4 a legal move for black? Why did he not play it?

  103. Anonymous says:

    is the kifu already up somewhere for download and review?

  104. BobbyBriggs says:
  105. It is a sad day of the start of decline in Professional Go. For sure it was inevitable but sad nonetheless.

    Now that computers have surpassed humans in Go, I expect interest in Professional Go to fall away rapidly. Spending a lifetime studying and playing will now seem pointless. Regardless of how strong a human becomes against other humans, everyone will know their play is but a shadow of what a computer can do. This means that Go will just become a pastime and we will not see people devote their lives to it because it is futile and pointless.

    • Anonymous says:

      why would it? In chess Magnus Carlson is highly praised. Chess players learn new ways to interprete positions and openings with the help of strong engines. Alphago will help strong Go players to become even better. the excitement of watching the 100m finals in the olympics is not tainted by the knowledge of race cars being able to beat human runners to the finish line. technology is a tool and a means and not the end of the way.

      • What will be a sad development for go is that as soon as bots with that strength are available for everybody, people will start to criticise grandmasters for their play.

        I invite you to go to and open the relay broadcast of the FIDE candidates tournament. You’ll be appalled about how the people hurl abuse at world class players in the chat. Statements like “what an idiot, move XYZ was really easy to see” are very common there.

        Before the bots became strong, people treated grandmasters with far more respect.

        I hope that the go community keep their noblesse in this respect.

      • Yes, but Go players are not just revered for their prowess and their mental abilities, but for their knowledge. Go, and chess before it, used to be an area of human knowledge, a minor science, and also an art, and as such also a field of research. And now it is being downgraded to a sport, as highlighted by the 100m analogy. We are unlikely to have any more Shusakus and GoSeiGens who will not just be the top players of their era, but will bring new ideas, new ways of playing, new theories and will be inventors and artists. Now we will just ask the computer what is the best way of playing.
        Sure, sport is fine, and fun, and people are happy to spend their lives practicing it, but it is not quite the same, and I can’t help feeling sadness too.

    • I honestly think this will create a golden age of go.
      Children all over the world will have a chance to learn go at a high level now. Also, the existence of high quality go programs may have a different effect on opening play than it has had in chess.

    • Lafcadio says:

      A robot can throw a javelin farther than a human but we still enjoy the Olympic games!

  106. Anonymous says:

    thank you, this helped 🙂

  107. iamnotfilip says:

    I think the rules are bent in computer’s favour. There are many computers (a team) that communicate via high-speed communication links. They are also allowed access to older games and to play out the combinations on (internal, virtual) go boards. Also they don’t get tired or drained, so the time limit does not effect both equally.

    To make it more even, a human should have a go-board(s) beside to try out combinations of his own. Have access to a team of people who can give input on local battles. Introduce breaks or have time limit for humans increase as match drags on.

    That would be a more fair match of man vs. machine.

    • Yeah, maybe, but that’s not the point anyway : the best computers are now at par or better than world-class players, that’s a fact.
      And they will for sure improve with improved algorithm and additionnal computing power (altough it will take a long time to get as much power in a computer at home !) : in the coming years, that will not be a question anymore, computers will crush humans … As in chess ; -(. However, the game of Chess still exists, players still enjoy it and professional players did not disappear.

  108. Melodicbard says:

    I think there will be a new battle ground, a go competition for AI. Darkforest might have lost the sente to beat human, but it can be Alphago’s next challenger!

  109. It is so disappointing to see that something that was once thought of the empire of human brain special has suddenly vanished due to some not especially refined algorithm (and machinery) produced by the same brain. The art of go, development of thousand years of generations has fallen apart.
    I am personally not interested in go made perfect by machines. At least we human are always going see it as perfect until a newer algorithm will not prove otherwise. From now on no race between man and computer in this area.

    • The only suddenness is that the victory of an AI over top human has come several years to a decade earlier than expected. The art of go does not suffer in the least for it, the human study of go and human playing of go are completely different to what AlphaGo does, the machine doesn’t supplant, but assists, by showing us moves that we wouldn’t have ourselves found. It can provide no theory or strategy for why they work, though.

  110. I have never understood why people root for the human player in these kinds of games. The AI represents the greatest achievement and hope of the human race, while the human player is just one individual. Lee Sedol being personally good at Go buys the rest of us nothing. Scientists being able to build machines that can solve difficult problems is a triumph that buys us everything.

    This game was the human race versus one individual human. The human race won. We beat Lee Sedol!

    • Anonymous says:

      In sports there is the phenomenon, when the national team wins, everyone says “we won”, whereas when it loses everyone says “they (the players) lost”. In this particular case, I’d prefer to say that DeepMind/Google won.

    • Because we wanted to belive that ntuition and perception of beaut are important in GO, something that we believed that is inaccessible to machines… maybe it was our human ego, we wanted to have clear confirmation that we are someting more than just a mind and inteligence.

    • Because we wanted to belive that intuition and perception of beauty are important in GO, something that we believed that is inaccessible to machines… maybe it was our human ego, we wanted to have clear confirmation that we are someting more than just a mind and inteligence.

      • But we AREN’T something more than minds. A mind is what a person IS. It can’t be any other way.

  111. “We beat Lee Sedol!”
    Definitely did not. We beat ourselves. From now on any win over a machine can only be temporal. We can use machine to improve against a human, but not against machine.

  112. Darek! Wholeheartedly agree.

  113. Michael Bacon says:

    In lieu of komi has the Go world ever considered having white play the next two stones after black plays the first? I recall reading an article by top mathematicians which proved allowing the second team to take two shots on goal in overtime in football (soccer in the US) and then alternating shots obviates the advantage of taking the first shot.

    • William Chang says:

      Michael, please see below (end of game 4 comments) my proposals for game AI research and classic-game variations especially the Free-move option: the player who has made the fewer moves has the option to make a free move subject to game-specific constraint. For komi-Free Go the constraint on the double-move is they cannot be adjacent/diagonal and cannot together make a capture. Comments and suggestions appreciated, — William

    • Wouldn’t your proposal of having black first and then white goes twice, then black/white alternating as usual simply be the equivalent of White going first?

      Because by the time of the fourth move, both black and white will have made 2 moves each and then white continues to be one move ahead. That is exactly the same as if White moved first. And obviously having White go first doesn’t solve the komi issue so your suggestion doesn’t quite work.

  114. BobbyBriggs says:

    One day, an IA will have consciousness and emotions. Let’s be prepared for that day….

    • The three laws of robotics must be introduced into the legal system.

    • Anonymous says:

      If AI gains consciousness and emotion, perhaps it is no longer an AI but a full human being, just non-organic.

  115. Anonymous says:

    I believe Lee’s sole win may be the very last time a human tops Alpha Go. Except for the first game, Lee played superbly, as well as any human can. And though he is ranked only fourth right now, the differences between top ten are miniscule anyway. I don’t see anyone, including Ke Jie & Park Jung Hwan, beating AlphaGo. It will take a miracle like the fourth game to pull one off & AlphaGo will continually get stronger.

    • And even if they beat it once or twice that is going to be just temporal.

    • There are pretty clearly some big problems with AlphaGo, which are neither easily debuggable, nor easily reproducible, and this is going to be a problem for machine learning generally. Take the series of moves following White 50 — these serve no purpose and lose points, yet AlphaGo continued to make them. Similarly with the moves following Sedol’s tesuji in game 4, they were just bad moves that AlphaGo simply wouldn’t make if its strength was at a consistent level.

      It seems that were AlphaGo opened for playing with globally, people would fairly quickly identify what exactly causes AlphaGo to spaz out, at which point it would go from top strength to no strength. Such is the inevitable downside of statistical learning.

      • A few month ago the strongest know program was like a strong amateur. It was quite an achievement even after decades of continuous developments. And here we have something that is just the beginning, because although so many things are still missing from this algorithm, still it’s strength is formidable, way beyond the capabilities of 99 percent of go playing humans.
        I understand your point on reproducibility and debuggibility or whatever -abilities. I guess the first giant steps are made refinements or much improving additions are inevitable, machines are unstoppable. And if it could be done in go other areas will soon be in preparation too.

  116. William Chang says:

    As a former practitioner of another “impossible task”, natural language web search, I would like to piece together my few observations here and on another forum Boardgamegeek interpreting the strategies and reversal that happened in this match, together with a few proposals for further research in game AI, and classic-game variations that may be interesting to play, have deeper strategy, and present additional challenges for AI.

    I have added them to the end of game _4_ comments. Feedback and suggestions very much appreciated.

    — William Chang (former Baidu chief scientist, Infoseek chief architect and CTO)
    williamichang at

  117. William Chang says:

    In the interest of fairness and enhanced performance for all:
    (1) If the AI can add hardware to buy time, then the human player should not be clocked.
    (2) Humans should play as a team.
    (3) Any opening book should be shared, i.e. open book.

  118. William Chang says:

    Well here in Beijing, Nie Weiping and his protégé Ke Jie are both quite confident of besting AlphaGo in the early game.

  119. William,
    That is an interesting comment about Nie Weiping. I had thought specifically of his style as being difficult for computer to deal with. I recall that in his prime he would often tenuki in the middle of joseki to an extent and with a frequency I don’t believe was matched by other players. His global sense of the board was special. I wonder if he and Ke Jie have in mind this style of opening to beat Alphago.

    • William Chang says:

      I caught a few minutes of today’s end game with Nie as webcast commentator (can you believe that!!!) and he kept saying of Lee’s stones in the left/lower center “this won’t connect”, “can’t connect”, “not going to connect”…. So, I think he was criticizing Lee’s shortcomings playing for tactics/territory and not enough strategy/influence. I don’t know what Nie has to say about game 4’s move 46 that I love so much, at the center of but not close to the surrounding action, that eventually set up move 78.

      • 46 was a classy move. I thought Lee could ignore 47. Playing to stabilize the center group was huge.

  120. If it’s true that (a) AlphaGo’s strategy has not been changed throughout the contest, and (b) it has no random factor (for example if two moves look equally good), and if Lee had won any of the first three games, then he could simply have repeated the same moves when he next held the same color stones. I suspect, however, that this would have been considered very poor form, especially in Korea.

  121. William Chang says:

    I’m sure (b) is not so, random sampling is key to its method (Monte Carlo tree search), and given moves of close-enough score it probably chooses randomly.

  122. William Chang says:

    I just had an epiphany or moment of pure fantasy. To me, Weiqi (Go) is the closest we have to a perfect abstract game. And I firmly believe games are abstractions that simulate social competition or phenomenon. Yet, there are aspects of Weiqi that have made me uneasy, half subconsciously. On the surface, why is the board so large as to confound novice players (especially children but also de novo AI) where to start playing. Deep down, was there ever a proto-Weiqi that more closely represented the expansions and movements not only of armies, but also of tribes and peoples in ancient times? Are there any clues of such a theme in our quintessential abstract game?

    In Search of Proto-Weiqi (added to game 4 comments)
    — William Chang (former Baidu chief scientist, Infoseek chief architect and CTO)
    williamichang at

  123. JamesTar says: