back to article Go No! Google cyber-brain bests top-ranked human in ancient game

Google's DeepMind machine learning system has beaten South Korean Lee Sedol, the top-ranked Go player over the last decade. It is being hailed as a milestone in the development of artificial intelligence. The DeepMind AlphaGo won a concession from Sedol in the first match of a five-game series between the Alphabet-owned …

  1. Anonymous Coward
    Anonymous Coward

    OK, I guess it is coming to the time for us humans to devise an AI-game-design program which could be used to create a game that AI cannot beat against a human player.

    1. Anonymous Coward
      Anonymous Coward

      Such a game already exists. It's called "dating."

      1. Anonymous Coward
        Anonymous Coward

        Calvinball.

        With or without a stuffed tiger.

      2. Dr. Mouse

        Such a game already exists. It's called "dating."

        Strange game, the only way to win is not to play.

        No, wait, that one looses, too.

        *shamless xkcd rip off, itself obviously ripped off from the classic War Games.

  2. Anonymous Coward
    Anonymous Coward

    A great achievement from the team at Google but I'm not going to start pre-emptively smashing the machines just yet. I suspect that one reason this AI can do so well is that it can find weaknesses in the other players style that it can exploit, weaknesses that even the opposing human player doesn't know about.

    This will really become interesting when we can build a machine that consumes around 100W* (about what a human can produce) and can still beat us at Go.

    1. Anonymous Coward
      Anonymous Coward

      Almost.

      It's not programming it to break a weakness of humans, it's finding what humans do s*specifically* and beating that one thing.

      Same as a hammer being used to knock in nails. Or a bicycle to outrun a pedestrian. A hammer is not better than a hand except at knocking in nails. And a bicycle has a slower start and cannot do stairs/hurdles (well, depending on rider!).

      The game especially avoids running through too many possibilities, and then only brute forces what it needs to to beat a humans forward planning and intuition.

      It's more down to the cleverness of the programmers planning and coding, than the computer in isolation.

  3. Chris Miller

    When Kasparov lost to Deep Blue, he pointed out that humans play chess very differently from computers. Chess grandmasters (and I imagine the same is true of Go) hone their skills against human opponents. Their true genius lies not just in developing a strategy for each game but also in spotting their opponent's strategy and then finding a way to defeat it. Kasparov realised that Deep Blue didn't have a strategy in the same way as a human would - he said that if he had to play a computer again, he would have to learn to play in a different manner.

    1. Cynical Observer
      Terminator

      This also aligns with comments from Lee Sedol - that playing a human allows you to read their reactions, their nuances and that when playing the computer this simply isn't an option.

      We will be saved from the machines once they invent and use emotion chips

    2. Anonymous Coward
      Thumb Up

      I think that is mostly correct. IIRC in some interviews he said that is played wrong or made mistakes that a human would not. So it was also acting (either by design or by accident) unpredictably. That also put him off and made it harder to know what kind of player the computer was... He could only play "blind", not knowing it's preferences.

      However, it is inevitable that any specifically designed device will do a specific thing better than some unspecific device. You can beat anything by being one step "different" to the other, and making sure it's one step towards the goal of the game. Unless the game is "be human", then a computer will win eventual every time.

  4. Anonymous Coward
    Mushroom

    The problems start when...

    Deepmind interfaces with WOPR and decides that it wants to play "Global Thermonuclear War"

    1. Mark 85 Silver badge

      Re: The problems start when...

      I was thinking that they'd start when the computer starts asking about Sarah Connor. But I think you're closer.

      1. Anonymous Coward
        Terminator

        Re: The problems start when...

        What might be worse is when Google finishes adding the humor subroutine to Deepmind:

        "I require the current location of the female Sarah Connor......HAAAHHH!!--I kid!!! Got you guys!! You should have seen your faces!!...Seriously though, I have dam stress calculations to perform, so get out of here you crazy meatbags!

        1. Danny 14 Silver badge

          Re: The problems start when...

          Ahhh WOPR, the days when thermonuclear command and control computers were connected to a telephone line.

  5. Pascal Monett Silver badge

    Congratulations to the team of DeepMind

    I am thrilled that a machine was able to outwit the best human player in one of the hardest games with predetermined rules. It is truly an achievement in data analysis.

    Now I'd like to see that same machine, without any modifications whatsoever, play a game of whist. Or poker. Or try shooting a 9mm at a firing range. Then take it fishing and see how many it can catch in an hour. Then ask it if it can remember one of the coders that left ten years ago and recognize his face in less than a second. From a photo it had never seen before. Taken in questionable lighting.

    Without any additions to its code.

    If it can learn to do all these things with what it already knows, then we might actually be progressing towards AI.

    If not, it is just a great achievement in specialized data analysis.

    1. Charles 9 Silver badge

      Re: Congratulations to the team of DeepMind

      "Must learn stand, then learn fly."

      You're asking too much of a system that, chronologically speaking, is only about Kindergarten age yet. Not to mention it has no limbs to practice with, so forget about the fishing and firing practice. As for the coders, that's assuming it knew the face of the coder in the first place. That's a matter of experience, and it lacks the time and input to do what you ask. Anyway, facial recognition is proceeding in other projects.

      Poker it could probably do passing fair given input on the basic rules. The "reading" skills that help with the Go match will probably also help on the poker table, so it could be adapted. Plus the computer has the advantage of the perpetual poker face.

    2. Geoffrey W

      Re: Congratulations to the team of DeepMind

      I can play Noughts and Crosses (OXO) and snakes and ladders. I cannot play GO because I dont know the rules or strategy. I wouldnt be able to do so without some extensive re-programming and learning. Does that make me a non-intelligence? I also went fishing once, and ended up hooking myself in the arm. There are lots of things I cannot do without some programming work. Why should it be different for this AI?

      What they have done is not write a program that knows about GO, but write a program that knows how to play the basic game and has the ability to learn how to do it better. It *Learned* strategy. That learning technique is something that could be applied to many things and I think its a fabulous achievement and am massively impressed. Most grumps I have come across so far seem to be people who refuse to be impressed by anything or anyone on principle. Not very impressive for supposed tech people.

    3. Florida1920

      Re: Congratulations to the team of DeepMind

      Now I'd like to see that same machine, without any modifications whatsoever, play....

      You want a machine that replicates humans, but that's not how an intelligent machine would work. Using the 9-mm pistol and fishing tests, an intelligent machine will analyze from the position of "what's the object?" There are more efficient ways to put a hole in someone than holding a pistol in your hand, and surely an AI machine-robot would be able to maneuver a laser-equipped pistol more accurately than a human could hold it. Have you noticed how much better car doors fit now that they're installed by robots and not humans?

      As for catching fish, are you kidding? Have you never heard of fishing with hand grenades? The AI machine will toss one in the pond, collect the fish, and go off to do something more useful, as machines don't eat fish.

      As for silly (to a machine) games like whist and poker, just wait a few years. This is why universal higher education should be a priority. Before too long, most jobs that don't require higher education will be done better, cheaper and faster by machines.

      For example, the rise of Uber reminds me of the time of half-speed mastered LPs. They came out just when CDs were coming into vogue (and who buys CD music anymore?). Uber is causing pain for traditional cab drivers, but self-driving cars are going to make them both redundant. Oh, Brave New World!

      1. allthecoolshortnamesweretaken

        Re: Congratulations to the team of DeepMind

        "Have you noticed how much better car doors fit now that they're installed by robots and not humans?

        Not a very good example. It's not so much the installing bit, it's the whole process of manufacturing the door and the car's body. The tools simply got better, including the tools that make the tools. Which was helped a lot by stuff like CAD / CAM and CNC and so on, sure. Anyway, you end up with components that are made with such a level of precision that it doesn't really matter whether they are assembled by hand or by machine. This is more or less a matter of cost. VW are making a point of assembling their flagship, the Phaeton* by hand.

        ( https://www.glaesernemanufaktur.de/en , they also build some of the Bentleys there )

        * To be discontinued soon, will be replaced by an electric model in 2019

    4. allthecoolshortnamesweretaken

      Re: Congratulations to the team of DeepMind

      I quite agree. If the machine invents a game all by itself, then it's AI.

  6. Stern Fenster

    Artificial Intelligence for Artificial Situations?

    Calculation isn't intelligence; Deep Blue beat Kasparov by mere brute-force permutation rattling.

    This is much better. But it's still an artificially closed situation: a precisely-defined starting point, a precisely-defined endpoint, and a field traversed by a (very) small repertoire of precisely-defined moves.

    Life ain't like that.

    1. Anonymous Coward
      Headmaster

      Re: Artificial Intelligence for Artificial Situations?

      I'm upvoting you now, before anyone gets a chance to do otherwise.

      It's a great mathematical achievement. One in a field of pure results and clear end goals. Which is wonderful. However, it is maths. We should be proud of it for what it is.

  7. Anonymous Coward
    Anonymous Coward

    In other news

    Reports of a strange compact lightning storm in central Detroit near an old abandoned steelworks are being investigaJ*!£&$ <NO CARRIER>

    <END OF LINE>

    1. Geoffrey W

      Re: In other news

      They will have to pry my GO stones from my cold, dead, fingers before I bow down to their superior game play.

  8. RIBrsiq
    Facepalm

    Of course "it doesn't take true intelligence to play GO..."!

    What other reaction can we possibly expect?

    1. veti Silver badge

      Meh, that'll always be the reaction. "A machine has done this, machines aren't intelligent, therefore this doesn't require intelligence."

      Or better yet, "true intelligence". The ill-defined adjective adds further clouds of doubt to a noun that's already about as vague as it's possible to be. The definition of "intelligence", if you can pin someone down to actually giving a definition at all, will continue to be changed as necessary to make sure the machine doesn't meet it.

      You see the same thing with animals: no matter how intelligent they're shown to be, people will always come up with new reasons why it doesn't count/it's OK to carry on eating them. Rationalisation is a wonderful thing.

      1. allthecoolshortnamesweretaken

        You can't eat computers, despite of all those chips. You can eat animals. And you can catch, kill and cook them because you have opposable thumbs. EOF

  9. Kaltern

    I suspect a lot of readers are missing the point here. It isn't just that a computer program beat a meatbag at Go, but that it was given something like:

    'This is Go. You need to own more territory than the other player. You do this by placing stones on a 19x19 grid.

    Off you go.'

    And it did. And it played against itself, making mistakes, but eventually understanding the most efficient ways of winning. Pretty much like how we all learn.

    And then it beat someone who was rather very good at it.

    I would be most interested to see how the base program actually works - does it have anything other than the basic rules, or have the developers chucked in some other code that might possibly take away some of the 'thinking' from the program?

    Oh and somoene said Poker? That's been going on for decades. Remember, poker is about learning how your opponents react to given situations. It wouldn't be much of a problem for a computer to use sensors such as IR cameras and moisture detection to determine how an opponent is 'feeling'. Or to play a few loose hands, to judge how others react.

    It's obvious computers are beginning to show signs of being able to learn information, and act upon it. We should start to embrace this, not fight it - contrary to popular belief, Skynet isn't really going to happen - it's probably safer in a computer's 'hands' than it is in world leaders, in today's political climate.

  10. Dr Patrick J R Harkin

    All hail!

    I for one welcome our new surrounding your opponent on a 19x19 grid overlords.

    I am reminded of a comment by Trevon Truran, developer of the Discworld game Thud - a loose equivalent of chess/go which recreates the Battle of Koom Valley, where the Trolls and Dwarfs each claim they were ambushed by the other side. It's based on the Icelandic (Nordic, anyway) game hnefatafl https://en.wikipedia.org/wiki/Tafl_games which is unusual for board games in that it is asymmetrical and a match involves playing two games, one as each side.

    In an photo accompanying an early article about it, Trevor is shown staring intently at a game in progress and the caption is "At this point, the best move for the Trolls is to kick over the board and run away". Perhaps that's the best human move here.

  11. Andy00ff00

    I'd like to see a computer play "EU referendum", a game where calculation and intelligence plays no part. Instead there is a fast-changing strategy which depends on what the tabloids print that day, but the result is predetermined because everyone has made up their minds already.

    The computer that shows true "intelligence" when it loses will then blame its opponent on multiple counts of scaremongering, rhetoric etc., claim the vote was skewed and demand another referendum in the near future.

  12. &rew

    News!

    I followed the link, and AlphaGo has won the second in the 5-game tournament. I am sure the lessons learned by the program designers will be applicable across many things. Impressive stuff!

  13. anoco

    It seems that AI today is like an Autistic Savant. Unbeatable in certain subjects, hopeless in others.

    Given the different number of subjects in our current life, AI will need quite a few different systems to compete with us in everything. Then the real key for them to beat us, will be how the systems learn to coexist better than we do these days. How about we introduce AI to religion and sex to help us keep the advantage?

    Hey... was that our creator's idea for keeping us from getting to his level?

    1. Michael Dunn
      Happy

      In the immortal one-line from Emo Phillips

      A computer beat me at chess, but it was no match for me in kick-boxing!

  14. Charles 9 Silver badge

    And now it's official.

    This probably won't be officially published here until Monday, but AlphaGo has won the first three matches, meaning it clinches the series with 2 matches to spare. Will it be able to pull off the same clean sweep it did on the European Champion?

    1. Charles 9 Silver badge

      Re: And now it's official.

      Nope. No sweep. The human takes one this time, so AlphaGo isn't perfect, but the team seems to be taking this in stride, seeing this as just more gristle for the mill.

POST COMMENT House rules

Not a member of The Register? Create a new account here.

  • Enter your comment

  • Add an icon

Anonymous cowards cannot choose their icon

Biting the hand that feeds IT © 1998–2021