Hey there! Sign in to join this conversationNew here? Join for free

Google AI beats world Go champion in 4 out of 5 games Watch

    • TSR Community Team
    • Thread Starter
    Offline

    20
    ReputationRep:
    TSR Community Team
    (Original post by chazwomaq)
    Yeah, millions, I think.

    This is really amazing. The difference in complexity between chess and go is astronomical. And the AI approach is totally different. Neural networks over brute force.

    The weird thing is, nobody knows how AlphaGo does it. Neural networks cannot really be probed to understand strategy or things like that. The computer can learn but cannot tell us how it learns.

    The scope for using AI to solve problems of science is amazing, and it really feels like we are on the brink. Some have speculated that we won't have professional scientists or diagnosticians in the future, just people who manage the AI computers!
    It was interesting watching the reaction of Myungwam Kim who's a 9 dan pro reviewing the game for American Go Association... he said he thought that he'd see computer Go take over in his lifetime but that this is "too soon".

    It really feels like a significant moment in AI technology. You wouldn't be able to tell AlphaGo was a bot just from playing style - that's quite a big deal.
    • PS Reviewer
    • Study Helper
    Offline

    14
    ReputationRep:
    PS Reviewer
    Study Helper
    Puddles the Monkey


    I've watched the first game in its 4-hour entirety and it's absolutely fascinating. From what I could understand whilst watching it is that the computer is weighing up the fine margins between winning and losing. It's not going for the decisive win, yet it's clever enough to know that giving up a few points here, to gain a few MORE points elsewhere in a delicate and tactical manner is going to give it the slightest of victory margins... It's playing a war of attrition rather than a full on assault.

    It's a level of comprehension I can only dream of :lol:

    Time to watch match 2 in it's 4-hour entirety . See you at lunch!
    • TSR Group Staff
    Offline

    4
    ReputationRep:
    TSR Group Staff
    This AI is another big step towards the eradication of the human race. As soon as a self aware AI is created, the first thing it would do would be to wipe out the human race.
    • PS Reviewer
    • Study Helper
    Offline

    14
    ReputationRep:
    PS Reviewer
    Study Helper
    (Original post by Nirgilis)
    Puddles the Monkey


    I've watched the first game in its 4-hour entirety and it's absolutely fascinating. From what I could understand whilst watching it is that the computer is weighing up the fine margins between winning and losing. It's not going for the decisive win, yet it's clever enough to know that giving up a few points here, to gain a few MORE points elsewhere in a delicate and tactical manner is going to give it the slightest of victory margins... It's playing a war of attrition rather than a full on assault.

    It's a level of comprehension I can only dream of :lol:

    Time to watch match 2 in it's 4-hour entirety . See you at lunch!
    My question was answered in the next video at 47 minutes...

    "AlphaGo tries to maximise it's probability of winning, but it doesn't care about the margin. If it sees two moves, one giving it an 80% chance of a 20-point victory, and one giving it a 90% chance of a 1-point victory, AlphaGo will pick the second choice. This goes against some of the instincts of Go players, particularly amateurs [...]".

    Astounding
    • TSR Community Team
    • Thread Starter
    Offline

    20
    ReputationRep:
    TSR Community Team
    (Original post by Nirgilis)
    Puddles the Monkey


    I've watched the first game in its 4-hour entirety and it's absolutely fascinating. From what I could understand whilst watching it is that the computer is weighing up the fine margins between winning and losing. It's not going for the decisive win, yet it's clever enough to know that giving up a few points here, to gain a few MORE points elsewhere in a delicate and tactical manner is going to give it the slightest of victory margins... It's playing a war of attrition rather than a full on assault.

    It's a level of comprehension I can only dream of :lol:

    Time to watch match 2 in it's 4-hour entirety . See you at lunch!
    Yeah, you should try watching the AGA reviews of the games here. They're useful and a bit more in depth then the official stream. Plus the dynamic between Andrew Jackson and Myungwan Kim is hilarious.

    We don't really know *just* how strong AlphaGo is because it plays to win and will take a secure win with less points - so there's lots of times when it looks like AlphaGo has made a mistake when actually it's just solidifying it's position to make a surer win. Pros do this too but AlphaGo can do it to such a fine level :eek3:

    I want to see AlphaGo play itself.
    • TSR Community Team
    • Thread Starter
    Offline

    20
    ReputationRep:
    TSR Community Team
    I also hope Google releases a transcript of the games showing AlphaGo's level of certainty at each move so we can see how early on it was sure it was winning.
    Offline

    0
    ReputationRep:
    Alphago's play is extremely reminiscent of a player who is often considered the best ever, the reigning champion before Lee Sedol, Lee Chang Ho. They both play an extremely solid, safe style, often ignoring seemingly better moves for another move has a higher chance of victory in the long run. Lee Chang Ho also happens to be the one of the only players to have a 50%+ win record against Lee Sedol, which suggests that the Alphago matchup is stylistically bad for Lee Sedol. Not to mention that Lee Sedol threw away game 1 with complacency and two gigantic blunders. Lee Sedol has a very tough match up against it. Alphago (like Lee Chang Ho did) has an extremely strong (some will call it perfect) late game, which is why I believe Lee Sedol has to play the hyperaggressive style he used to play in the early 20s. I think an unconventional beginning (maybe start with the first move on the centre of the board) into constant battling will force out Alphago to overuse its time and make mistakes. Also, props for google for making Alphago. I am seriously considering switching my major to Computer Science to study A.I.s!
    Offline

    2
    ReputationRep:
    (Original post by The Financier)
    This is worse. Deep Blue's approach was primarily brute-force searching and took a supercomputer that was near the top 250th most powerful for its time. AlphaGo in comparison uses neural networks to act in a similar way to how a human thinks and learns and doesn't need as many resources.
    That's why it's so amazing, for the first time ever, humanity is creating real artificial intelligence.
    • TSR Community Team
    • Thread Starter
    Offline

    20
    ReputationRep:
    TSR Community Team
    (Original post by Hiamint)
    Alphago's play is extremely reminiscent of a player who is often considered the best ever, the reigning champion before Lee Sedol, Lee Chang Ho. They both play an extremely solid, safe style, often ignoring seemingly better moves for another move has a higher chance of victory in the long run. Lee Chang Ho also happens to be the one of the only players to have a 50%+ win record against Lee Sedol, which suggests that the Alphago matchup is stylistically bad for Lee Sedol. Not to mention that Lee Sedol threw away game 1 with complacency and two gigantic blunders. Lee Sedol has a very tough match up against it. Alphago (like Lee Chang Ho did) has an extremely strong (some will call it perfect) late game, which is why I believe Lee Sedol has to play the hyperaggressive style he used to play in the early 20s. I think an unconventional beginning (maybe start with the first move on the centre of the board) into constant battling will force out Alphago to overuse its time and make mistakes. Also, props for google for making Alphago. I am seriously considering switching my major to Computer Science to study A.I.s!
    Do you play Go?
    Offline

    2
    ReputationRep:
    Kim Jong Un would nail the little *******. Balls back in humans court.
    Offline

    15
    ReputationRep:
    (Original post by Puddles the Monkey)
    Do you play Go?
    Go is not even the most popular board game in China. China invented another chess game that became more popular than Go. Go is now more played in Japan and Korea than in China.

    Also Mahjong.
    • TSR Support Team
    • Clearing and Applications Advisor
    Offline

    20
    ReputationRep:
    TSR Support Team
    Clearing and Applications Advisor
    3-0 :emo:
    Offline

    1
    ReputationRep:
    Every part of who I am is breeding a hatred for AI -- my ego, love for poker etc etc

    (Original post by HucktheForde)
    Go is not even the most popular board game in China. China invented another chess game that became more popular than Go. Go is now more played in Japan and Korea than in China.Also Mahjong.
    Watch Akagi right now
    Offline

    0
    ReputationRep:
    It is astounding to think that an artificial intelligence can now beat human intelligence. Technology nowadays seem to be better and better each day. However, there might (and will) come a time where humans will be too reliant on artificial intelligence. This can also be worrying because this might slow down the process of human intelligence and would affect the human generation. Though, I do applaud Google for creating such technology, but it is worrying for our future generation.


    Posted from TSR Mobile
    Offline

    15
    ReputationRep:
    (Original post by YourName___)
    Every part of who I am is breeding a hatred for AI -- my ego, love for poker etc etc



    Watch Akagi right now
    no thank you
    • TSR Support Team
    • Clearing and Applications Advisor
    Offline

    20
    ReputationRep:
    TSR Support Team
    Clearing and Applications Advisor
    3-1! :woo:

    From what I know, Lee made a series of slightly aggressive moves which a typical player would go conservative against. AlphaGo went on the attack move-for-move but it allowed Lee to respond well and as it was in a state that it wasn't familiar with it snowballed into several bad moves that handed Lee the game.
    • TSR Community Team
    • Thread Starter
    Offline

    20
    ReputationRep:
    TSR Community Team
    (Original post by The Financier)
    3-1! :woo:

    From what I know, Lee made a series of slightly aggressive moves which a typical player would go conservative against. AlphaGo went on the attack move-for-move but it allowed Lee to respond well and as it was in a state that it wasn't familiar with it snowballed into several bad moves that handed Lee the game.
    :yay:

    Lee made one particular move (79?) which AlphaGo didn't read properly.... it got confused, and didn't realise the significance of that move until about 10 moves later, when it suddenly 'realised' what had happened.

    After that AlphaGo realised it was behind and reacted to losing in the usual bot manner (which I thought was really interesting) by basically losing its **** and playing loads of random bad/pointless aggressive moves in an attempt to turn the game around. I'm still not 100% clear on why bots do this :beard:

    Lee Sedol reaction to bad AlphaGo move:


    • Section Leader
    • Clearing and Applications Advisor
    Offline

    21
    ReputationRep:
    Section Leader
    Clearing and Applications Advisor
    (Original post by Puddles the Monkey)
    :yay:

    Lee made one particular move (79?) which AlphaGo didn't read properly.... it got confused, and didn't realise the significance of that move until about 10 moves later, when it suddenly 'realised' what had happened.

    After that AlphaGo realised it was behind and reacted to losing in the usual bot manner (which I thought was really interesting) by basically losing its **** and playing loads of random bad/pointless aggressive moves in an attempt to turn the game around. I'm still not 100% clear on why bots do this :beard:

    Lee Sedol reaction to bad AlphaGo move:


    That was pure:
    • TSR Community Team
    • Thread Starter
    Offline

    20
    ReputationRep:
    TSR Community Team
    jneill

    Yeah it's amazing, he looks flabbergasted Then he even looks at the screen to make sure the guy placing stones for AlphaGo got it right :lol:
    • TSR Support Team
    • Very Important Poster
    • Welcome Squad
    Offline

    20
    ReputationRep:
    TSR Support Team
    Very Important Poster
    Welcome Squad
    (Original post by Puddles the Monkey)
    :yay:

    Lee made one particular move (79?) which AlphaGo didn't read properly.... it got confused, and didn't realise the significance of that move until about 10 moves later, when it suddenly 'realised' what had happened.

    After that AlphaGo realised it was behind and reacted to losing in the usual bot manner (which I thought was really interesting) by basically losing its **** and playing loads of random bad/pointless aggressive moves in an attempt to turn the game around. I'm still not 100% clear on why bots do this :beard:

    Lee Sedol reaction to bad AlphaGo move:


    Does this suggest that feints work well against AI?
 
 
 
  • See more of what you like on The Student Room

    You can personalise what you see on TSR. Tell us a little about yourself to get started.

  • Poll
    Would you like to hibernate through the winter months?
    Useful resources
  • See more of what you like on The Student Room

    You can personalise what you see on TSR. Tell us a little about yourself to get started.

  • The Student Room, Get Revising and Marked by Teachers are trading names of The Student Room Group Ltd.

    Register Number: 04666380 (England and Wales), VAT No. 806 8067 22 Registered Office: International House, Queens Road, Brighton, BN1 3XE

    Quick reply
    Reputation gems: You get these gems as you gain rep from other members for making good contributions and giving helpful advice.