Want to read Slashdot from your mobile device? Point it at m.slashdot.org and keep reading!

 


Forgot your password?
Close
typodupeerror
×
AI Google China Japan Software Games Technology

Google's AlphaGo AI Secretively Won More Than 50 Straight Games Against World's Top Go Players (qz.com) 139

Posted by BeauHD from the don't-tell-anyone dept.
An anonymous reader quotes a report from Quartz: When Google's artificial intelligence program AlphaGo made history by taking down Korea's Lee Sedol -- one of the world's best Go players -- in a landslide 4-1 victory in March, Chinese player Ke Jie was skeptical. He famously wrote on Weibo the next day, "Even if AlphaGo can defeat Lee Sedol, it can't beat me," and has since agreed to take on the AI at an undecided time. But now even Ke, the reigning top-ranked Go player, has acknowledged that human beings are no match for robots in the complex board game, after he lost three games to an AI that mysteriously popped up online in recent days. The AI turned out to be AlphaGo in disguise. On Jan. 4, after winning more than 50 games against several of the world's best Go players, Ke included, a user registered with an ID of "Master" on two Chinese board game platforms came forward to identify itself as AlphaGo. "I'm AlphaGo's Doctor Huang," the user "Master" wrote on foxwq.com, according to screenshots from Chinese media reports. Taiwanese developer Aja Huang is a member of Google's DeepMind team behind the AI. Since Dec. 29, Master has defeated a long list of top Go players including Korea's Park Jung-hwan (world No. 3), Japan's Iyama Yuta (No. 5) and Ke in fast-paced games. He won 51 games straight before his 52nd rival, Chen Yaoye, went offline, forcing the game to be recorded as a tie. By Jan. 4 when the test was completed, Master had racked up 60 wins, plus the one tie, and zero loss, according to numerous reports (link in Chinese).
This discussion has been archived. No new comments can be posted.

Google's AlphaGo AI Secretively Won More Than 50 Straight Games Against World's Top Go Players More

Google's AlphaGo AI Secretively Won More Than 50 Straight Games Against World's Top Go Players

Comments Filter:

  • When will google (Score:1)

    by Anonymous Coward

    When will Google make an AlphaMale AI/robot?

  • Wargames much? (Score:5, Funny)

    by Kjella ( 173770 ) on Wednesday January 04, 2017 @04:35PM (#53606683) Homepage

    He won 51 games straight before his 52nd rival, Chen Yaoye, went offline, forcing the game to be recorded as a tie.

    So the only way to win is not to play.

    • Re: (Score:3, Interesting)

      by PseudoThink ( 576121 )
      Star Trek TNG depicted nearly this exact scenario back in July, 1989, with Episode 21 of Season 2, entitled "Peak Performance":
      https://www.youtube.com/watch?v=yIRT6xRQkf8 [youtube.com]

    • He won 51 games straight before his 52nd rival, Chen Yaoye, went offline, forcing the game to be recorded as a tie.

      So the only way to win is not to play.

      Meh, gamers ragequit when losing online games all the time. Nothing spectacular here.

    • Re: (Score:1)

      by Anonymous Coward

      For Tic Tac Toe and Global Thermonuclear War, that is the correct approach.

      • Re:Wargames much? (Score:5, Funny)

        by ShanghaiBill ( 739463 ) on Wednesday January 04, 2017 @06:14PM (#53607229)

        For Tic Tac Toe and Global Thermonuclear War, that is the correct approach.

        Years ago, in NYC's Chinatown, there was a chicken that could play tic-tac-toe. You pay $2, and you can play against the chicken. I watched it pay a dozen times, it would always win or draw. When my turn came, I played, and it was a draw. They my cousin played, and lost. As we were walking away, I say "Dude, you just lost to chicken." He was quiet for a bit, and then said, "Yeah, but the chicken got to go first." Me: "Yeah, but still, it was a chicken." Him: "Well, yeah, but the chicken plays everyday. I was rusty." Me: "Yeah, but it was A CHICKEN. You are a HUMAN. Shouldn't that count for something?" Years later, I still rib him about it every time we meet. He definitely wishes he had chosen not to play.

         

        • Me: "Yeah, but it was A CHICKEN. You are a HUMAN. Shouldn't that count for something?"

          Learn the answer at any Chinese restaurant.

        • There was a home electronic game based on that exact scenario - "I took a lickin' from a chicken". Had a goofy looking robot chicken in a clear box, with inputs that would let it kick your ass a several games.

    • He won 51 games straight before his 52nd rival, Chen Yaoye, went offline, forcing the game to be recorded as a tie.

      So the only way to win is not to play.

      No. The only way to not lose is to quit before you are beaten.

      Resistance is futile, winning is not an option.

  • AI Go Players? (Score:3)

    by TechyImmigrant ( 175943 ) on Wednesday January 04, 2017 @04:36PM (#53606687) Homepage Journal

    I'm waiting for the AI Rust players.

  • Don't forget; the Master Control Program started off as a chess program. Remember Encom!

  • It's official (Score:5, Informative)

    by Artem S. Tashkinov ( 764309 ) on Wednesday January 04, 2017 @04:49PM (#53606777) Homepage
    An official confirmation [twitter.com] from Demis Hassabis, a co-founder of DeepMind.

  • Ok for the next trick (Score:4, Interesting)

    by FunkSoulBrother ( 140893 ) on Wednesday January 04, 2017 @05:05PM (#53606867)

    I'll be impressed when you write an AI that can competently play Civ5.

    I'm not really sure if it's a more difficult problem than Go or not (I'd think so with all of the decisions to be made), but holy hell is the shipped AI in all Civ games useless.

    • Re: (Score:3)

      by es330td ( 964170 )
      I think the problem with Civ et al AI is that the individual PC doesn't have the horsepower to be a competent opponent. Given a webapi to Google AI I imagine developers could create games that are effectively unbeatable by humans.
    • They are already attempting something just as hard, they will be learning how to play Starcraft 2 competently

      • SC2 is a joke. Small map? Zerg rush kekeke. AI with insane APM to micro will always win that. Medium to large map? Protoss Stalker ball (again, with micro) to Void Rays.

        The reason the SC2 competitive scene hasn't taken off is the game is fundamentally shitty. It's a rock-paper-scissors game 90% of the time, and Blizzard balances it not for fairness but for a desired outcome (equal win/loss rates across all races at higher levels of play). This was a HUGE problem during the initial release because all

        • SC2 is a joke. Small map? Zerg rush kekeke

          Hits a wall, then dies.

        • And compared to BW no one gives a shit. Because it's boring as fuck, they can't balance it fairly,

          SC2 is such a great game right now. It long ago surpassed BW in terms of potential strategies, balance, and general fun.

    • Re: (Score:2, Insightful)

      by Anonymous Coward

      I'll be impressed when you write an AI that can competently play Civ5.

      Knowingly or not, you actually wrote a very legitimate (and probably harder) problem in AI. How do you make a good AI opponent? Note this is a different question than what was done in Go recently, which is "make a computer program as skilled as possible."

      You need to make an AI which will make mistakes, pretend like it doesn't know everything about the universe, including the player's civilization state (see The Computer Is A Cheating Bastard [tvtropes.org] trope and related tropes within that article), have multiple civil

  • But... (Score:1)

    by Anonymous Coward

    This is not AI, simply because the rules of Go were programmed into the computer to start with. If it had to figure out the rules and the idea of winning by itself then that would be amazing...

    However it was taught what a good move is by some point or similar system, that's hardly self learning....

    • I've always thought poker would be a better test for AI. Not in a card counting way, the computer would have to be forbidden from card counting since it is illegal to play the game that way. The computer instead would have to study the other players through a camera and understand what their tells are, and whether they are bluffing or not. Basically any game where the psychology of the other players matters as much as the rules of the game would be a better test.

      • Basically any game where the psychology of the other players matters as much as the rules of the game would be a better test.

        That's a completely different class of game. Literally, a completely different class.

        Psychology essentially means having some game-relevant information which some players know and others don't. At least some players (probably all - I've never been interested to learn how to play poker beyond what's necessary for answering statistics problems) only have partial information on the stat

        • Right, so what I'm saying is that it's not really interesting to see a computer play Go because the computer already has all the information it needs and is just a matter of doing a deep calculation on it. No different then pressing buttons on a calculator because the math in the calculator is already known. A more interesting application of AI is where it must build its cache of information about the world and make decisions based on it successfully. In fact I propose the former isn't really AI, just a

    • Re: (Score:3)

      by haruchai ( 17472 )

      This is not AI, simply because the rules of Go were programmed into the computer to start with. If it had to figure out the rules and the idea of winning by itself then that would be amazing...

      However it was taught what a good move is by some point or similar system, that's hardly self learning....

      Are you kidding? Then only a very small %age of humanity could be considered intelligent as there are a lot of things that a vanishingly small number of children would NEVER figure out without help.

      • You're looking at it the wrong way. The fact of the matter is, AlphaGo can do one thing in it's life, play Go. Make an exhaustive list of the things in life a preschooler may be capable of (identify a color, identify an animal, read a book) and quickly the preschooler looks vastly more intelligent. If a person was born that could do nothing but beat everyone at Go, they wouldn't call him/her intelligent, they would call them a 'savant'. They used to be called 'idiot savants' but, political correctness..

        • Re: (Score:2)

          by haruchai ( 17472 )

          You're looking at it the wrong way. The fact of the matter is, AlphaGo can do one thing in it's life, play Go. Make an exhaustive list of the things in life a preschooler may be capable of (identify a color, identify an animal, read a book) and quickly the preschooler looks vastly more intelligent. If a person was born that could do nothing but beat everyone at Go, they wouldn't call him/her intelligent, they would call them a 'savant'. They used to be called 'idiot savants' but, political correctness..

          The advantage I give to humans is that a single gifted, well-honed human brain is still capable of besting just about any single-machine AI at many difficult tasks.
          AlphaGo, otoh, is running on hundreds of CPUs & GPUs.
          But as far as the savant argument goes, I don't think we'll hold the high ground for very long, perhaps a decade or so and we'll see AIs that are multi-talented.
          The real issue is not what they're capable of but who they can replace. A few companies are now looking to offload many middle man

          • Perhaps AIs will become multi-talented or have some intuition about the world, but I believe approaches like AlphaGo will not tend to be expanded into a multipurpose AIs. AlphaGo is too much like a drag car designed to drive a 1/4 mile as fast as possible. The only way to get to a daily driver car is to scrap everything and start with a different approach.

    • Re: But... (Score:1)

      by Anonymous Coward

      You are actually wrong. AlphaGo was deep learning and evolved from playing all these pros. Thats why it cant be outsmarted using unconventional moves. There is minimal preprogramming involved.

      • "Deep Learning" is a buzzword. It's a search algorithm or a trained neural net (that is implementing a search algorithm). How do I know? Because it's effective at playing Go. Go is an incredibly simple game with a massive search space. And it was absolutely programmed with the rules of the game, the win/loss conditions, etc., or at least explicitly trained on them in its bootstrapping phase.

        If you can't take AlphaGo and teach it to play Parcheesi in an hour without adjusting its code, is it really (or

        • Neural nets don't implement search algorithms ...

          'Artificial Intelligence' is a term like 'Autism' ... it has a well defined meaning for the people working in those fields. And laymen like you are ignorant about that meaning and try to impose what you think it should mean on your readership ;D

          • Neural nets can implement just about anything they're trained to. It won't necessarily be the best or most accurate (it can easily fall into a local min/max issue until it's trained on data that bumps it out) but it generally works if you give it simple rules and goals, then let it train. A "search algorithm" here refers to a decision tree for making moves.

            • A neural network is not able to analyze a decision tree. Hence it is not a search algorithm etc.
              It is a very 'simple' input pattern versus output pattern matcher/generator.

              The 'not so simple' task is to find a fitting topology (how many layers etc.) and to train efficiently and successful.
               

        • You mean, something like this? [wired.co.uk]

    • It is indeed true that one neural network was trained using a collection of games, but the version of AlphaGo that played against Lee Sedol last year, was using two neural networks, and the second one (for evaluating the positional strength of a board configuration) was trained by letting AlphaGo play against itself. It is not known how he current version of AlphaGo works, whether any additional neural networks were added, but if it has become stronger, it has done so by playing against itself. It should be

  • "He won 51 games straight before his 52nd rival, Chen Yaoye, went offline, forcing the game to be recorded as a tie."

    Typical rage quit after getting rekt, ruining the game for everyone involved.

    • Re:Typical (Score:5, Insightful)

      by Frans Faase ( 648933 ) on Wednesday January 04, 2017 @07:15PM (#53607553) Homepage
      You seemed have to missed the fact that many of the top professional players were lining up to play against this bot. They view it as free training lession, not to beat an AI bot, but to beat their human counter parts. Since Lee Sedol played against AlphaGo, he has gained in strength, so much even that a certain point, using a certain method, AlphaGo was the strongest player, not because it had played more games, but because Lee Sedol had won so many games. Ke Jie, to be considered the strongest player at the moment, has made remarks that humans have only touched at the truth behind go, after he played against Master(P). Most go players have a very high regard for the game, as they sense that it is much deeper than human mind can consider. For this reason, I guess, many professional go players find this a very exicting time, because it will enhance their understanding of the game. In this view it is very unlikely that a professional player will use a trick to force a tie.

    • Comment removed (Score:5, Interesting)

      by account_deleted ( 4530225 ) on Wednesday January 04, 2017 @07:38PM (#53607653)
      Comment removed based on user account deletion

  • There was some question of if it was a fluke against Sedol, now it's confirmed that machine has bested man at Go. First Chess, now Go. I'm wondering where the goalposts will be moved to now.

    I can imagine some kind of Turing test where a woman converses with two suitors, and has to choose which to go on a date with; one is a man, the other a machine (cue the jokes). Lines of dialogue used to have to be pre-programmed, but with all the deep learning that modern AI can do, with access to project gutenberg/wiki

    • Re: (Score:2)

      by Fwipp ( 1473271 )

      Have you listened to men? ELIZA would win that test.

    • Re: (Score:1)

      by Anonymous Coward

      Ok, time to increase the board size to 37x37.

      Let's be honest. It's quite an achievement. But the last stand against AI should continue to stand.

      In fact when chess fell to AI we learned nothing about AI, but when go effectively fell last year we had learned something from it. The play looks like human play because it's playing like a human.

      • Ok, time to increase the board size to 37x37.

        People have been playing with different (larger) sized boards for decades if not centuries (ISTR that some Japanese professionals were studying 21x21 games in the 18th century ; 20x20 was considered a solution to the "mirror go" problem in the 15/ 16th century, until more elegant solutions were developed). The high-dan players who have worked on larger boards say the balance of the game between territory and influence changes considerably.

        37x37 would be an unnec

    • He knew Sedol was beaten yet had the hubris to say, "Even if AlphaGo can defeat Lee Sedol, it can't beat me". Actually sounds like a lot of posters here, "AI will never take my job"

    • Re: (Score:1)

      by umghhh ( 965931 )

      This test is useless - most designers of the machines are hetero men thus they would have chosen a girl for tests as in this movie [imdb.com].

      The there is this other thing - in our quest to improve our lot we went on to improve all other things too with economic efficiency being the best factor to decide what is better. Now what would be the final logical conclusion of a system that was set to improve that to the maximum possible value?

    • Before that there will be an AI scammer, that will be very convincing in make you give some money to it

    • There was some question of if it was a fluke against Sedol,

      Not from Sedol. No other opinion matters.

  • So just any anonymous doofus can go on a website and play against the world's top players? Interesting...

    • Re: (Score:2)

      by ffkom ( 3519199 )
      I would assume their AI had to play numerous games against increasingly strong online players before those top players were offered as opponents by the online service.
      • It's a pretty safe bet that the AlphaGo team has some pretty good players onboard who could seed the stealth account with some high-ranking initial opposition to establish it's strength. After that, something that shows up as a confirmed 7 or 8 dan professional would pretty soon attract more ranked opponents. If they didn't want to wade through the low ranked people, then you can set tthe minimum rank for possible opponents - or at least you could the last time I played go online (8 or 10 years ago).
  • ... but now that he's elected, the guy they hired to act as their "candidate face" - based on AI-authored tweets and speeches - refuses to drop out of his role and will actually rule the US. Google's programmers meanwhile are frustrated how little of a challenge it was to have their AI beat Hillary. ;-)

  • Pride goeth before a fall (Score:2, Interesting)

    by Anonymous Coward

    As we see yet another instance of hubris in action, this time the assertion of Go players and hangers-on that "Go is so much more complex than chess, so it will never be mastered by a machine". Computational complexity or large problem space has little to do with either play-ability or ease of mastery.
    Next challenge?

    • It wasn't hubris, and yes, go is much more complex than chess. People didn't overestimate the game, they just underestimated how fast AI techniques would advance.They thought this would come in the 2020's not 2016.

    • Re: (Score:2)

      by Herve5 ( 879674 )

      (...) "Go is so much more complex than chess, so it will never be mastered by a machine".

      I am not this old, but I *perfectly* remember having owned a book on computing that addressed the same topic 30 years ago with the previous-level example: "tick-tac-toe is lost to the AI but chess is so much more complex, so it will never be mastered..." :-)

    • the assertion of Go players and hangers-on that "Go is so much more complex than chess, so it will never be mastered by a machine".

      Having been playing Go for 33 years now, and knowing people who've been trying to program Go for 32 years, I've never heard either a player or "hanger-on" make that assertion. Even in the days when the best program in the world could be beaten flat by a human with a couple of evening's teaching.

      Come to think of it, I've never met a "Go hanger-on" who was not also to some degree

  • ...humans are still better at making up games :)
  • that AlphaGo is so lucky!

  • After chess, checkers, poker, and Arimaa fell, the significance of Go with its huge search space was whether its huge search space and geometrical intuition represented a fundamentally harder challenge for computer algorithms, or just a different challenge, one that we hadn't figured out yet.

    Many were saying that Go posed a fundamentally harder problem than chess. I took the view that Go would fall hard once it finally fell, but it was unclear when that day would arrive. I thought it was more "different"

  • ... is futile.

  • AI does better than humans in chess, Jeopardy, soon driving and surgery, and now Go? Come on, man, put this "better than humans can do" power to real work and build a sexbot that can fuck the holy hell out of me!

    Don't mod this down! You know you want this.

  • Now do it on about 20 Watts while still having the robustness and flexibility of a human brain.
    Let's not lose sight of how beating someone in a complex task doesn't equate to what humans intelligence.

    Don't get me wrong, though, this is an impressive point we've reached in technology.

Slashdot Top Deals

"Confound these ancestors.... They've stolen our best ideas!" - Ben Jonson

Close