Stories
Slash Boxes
Comments

SoylentNews is people

posted by martyb on Monday November 07 2016, @01:57AM   Printer-friendly
from the Destroy-all-Human-[Player]s! dept.

Google's DeepMind division will attempt to make an AI that can play Starcraft II in real time without using the same unfair knowledge and capabilities (such as controlling units that are "off-screen") that Blizzard's own AI use. Blizzard and DeepMind are working on a build of the game that will be "open and available to all researchers next year".

Reported at The Washington Post and The Verge .


Original Submission

This discussion has been archived. No new comments can be posted.
Display Options Threshold/Breakthrough Mark All as Read Mark All as Unread
The Fine Print: The following comments are owned by whoever posted them. We are not responsible for them in any way.
  • (Score: 2) by julian on Monday November 07 2016, @02:53AM

    by julian (6003) on Monday November 07 2016, @02:53AM (#423370)

    It'll be interesting to see if they can get it to scale to more complex games. Atari games have at most one button for input and the joystick, and movement is sometimes not even 2-dimensional but merely linear.

    My limited understanding with how they've gotten it to work so far is that the AI knows what's on the screen, knows the score (or the win condition), and has access to the game's usual controller inputs. So something happens on the screen and the AI reacts randomly at first. Some reactions correlate with increasing the score (or winning) and those are "learned" to be appropriate moves to make when the screen is in a certain state. If you do this enough the obviously bad moves get weeded out and the good ones become increasingly refined until perfect play is achieved.

    This works well for Atari games because there's few choices to make. In Pong you can either make the paddle go up or go down, or stay in its current location. If the ball is coming at you high the only good move is to move the paddle upward. Bad moves are to move it downward, or not move the paddle. But in Starcraft or any other modern computer game there could be, at any moment, thousands of choices you could make which branch to thousands more at every node. The search space is vast--probably vaster than even Go and has to be done in real time. Looking at the video demonstrating the API [youtube.com], it appears they've graphically reduced what the AI "sees" to be very similar to an Atari game.

    If any group will be able to do it, it'll be Google and DeepMind. It's definitely something to watch. Better AI in games is just a toy example of what this is good for. There's no reason the input couldn't come from cameras and sensors in the real world instead of a virtual one. It might enable real-world tasks to be treated like games and learned the same way it learned Starcraft.

  • (Score: 0) by Anonymous Coward on Monday November 07 2016, @02:55AM

    by Anonymous Coward on Monday November 07 2016, @02:55AM (#423371)

    Some of you may remember the bnetd lawsuit. It let you run your own server, because someone reverse engineered the protocol. What better way to run bots vs humans? Researchers could have been doing for the past decade. Will Blizzard have another trantum and DMCA someone for whatever excuse, like running a bot against official servers to try and see if it passess as another human?

    • (Score: 3, Insightful) by aristarchus on Monday November 07 2016, @03:02AM

      by aristarchus (2645) Subscriber Badge on Monday November 07 2016, @03:02AM (#423376) Journal

      , like running a bot against official servers to try and see if it possesses another human?

      Beg your pardon, but did you mean this? If you did, wouldn't it be more of a daemon than a bot?

      --
      #freearistarchus!!!
      • (Score: 0) by Anonymous Coward on Monday November 07 2016, @03:23AM

        by Anonymous Coward on Monday November 07 2016, @03:23AM (#423382)

        pass could be also:
        To go through any inspection or test successfully; to be approved or accepted.
        To be suffered to go on; to be tolerated; hence, to continue; to live along.

      • (Score: 0) by Anonymous Coward on Monday November 07 2016, @03:33AM

        by Anonymous Coward on Monday November 07 2016, @03:33AM (#423385)

        , like running a bot against official servers to try and see if it possesses another human?

        Beg your pardon, but did you mean this? If you did, wouldn't it be more of a daemon than a bot?

        Is this like a comical strawman joke?

  • (Score: 2) by moondrake on Monday November 07 2016, @08:50AM

    by moondrake (2658) on Monday November 07 2016, @08:50AM (#423414)

    Let DeepMind play eu4 instead.

  • (Score: 2) by linkdude64 on Monday November 07 2016, @04:43PM

    by linkdude64 (5482) Subscriber Badge on Monday November 07 2016, @04:43PM (#423590)

    They are beginning to weaponize AI under the guise of "gaming."

    • (Score: 0) by Anonymous Coward on Monday November 07 2016, @06:05PM

      by Anonymous Coward on Monday November 07 2016, @06:05PM (#423649)

      Just don't tell it about the game "global thermonuclear war". At least not before you taught it Tic Tac Toe.

    • (Score: 2) by butthurt on Monday November 07 2016, @07:13PM

      by butthurt (6141) on Monday November 07 2016, @07:13PM (#423695) Journal

      IBM's Watson was designed to play Jeopardy!; now they're flogging its medical diagnoses.

      /article.pl?sid=14/03/23/0142202 [soylentnews.org]

      • (Score: 3, Funny) by maxwell demon on Monday November 07 2016, @09:30PM

        by maxwell demon (1608) Subscriber Badge on Monday November 07 2016, @09:30PM (#423764) Journal

        Does that mean if you get a diagnosis by Watson, you are jeopardized?

        --
        The Tao of math: The numbers you can count are not the real numbers.
  • (Score: 0) by Anonymous Coward on Monday November 07 2016, @06:36PM

    by Anonymous Coward on Monday November 07 2016, @06:36PM (#423671)

    This is interesting from a perspective of AI and computer science. This seems kind of silly in terms of "human vs AI."

    The metric Actions Per Minute has a major effect at the highest levels, and computers have a dramatic advantage there (effectively being able to enter one input per frame of video... or something like 1800 APM, rather than the ~300 for humans). This would an inferior computer to just brute-force beat superior strategy from humans. It's a lot like the Jeopardy gameshow a while ago, where once an AI has a basic amount of knowledge it will always win (because at the highest levels the game devolves into "who can push the buzzer at the correct moment")?

    That being said, it would be really fascinating in regard to AI and dealing with ambiguity. It would also be really interesting to see two AIs playing each other and the micro-fights they would have... much like watching a Tool Assisted Speedrun.

    Actually, overall, this strikes me very much like a Tool Assisted Speedrun versus a human Speedrun. Both are entertaining, but they are not competitive with each other in any real way.

    • (Score: 0) by Anonymous Coward on Monday November 07 2016, @08:03PM

      by Anonymous Coward on Monday November 07 2016, @08:03PM (#423724)

      They seem to be artificially limiting the APM to human-like levels for this AI, in order to make it compete on a level playing field.

      • (Score: 0) by Anonymous Coward on Monday November 07 2016, @10:37PM

        by Anonymous Coward on Monday November 07 2016, @10:37PM (#423806)

        "Let's have a human have a tug-of-war with a truck... but to make it fair we'll limit the truck to be able to pull at most 50 Newtons of force."

        I guess the idea is to compare strategy between a human and a computer? Still, this seems like an odd way to approach the problem. I guess it could be "fun," though?

        • (Score: 2) by takyon on Tuesday November 08 2016, @01:33AM

          by takyon (881) Subscriber Badge <reversethis-{gro ... s} {ta} {noykat}> on Tuesday November 08 2016, @01:33AM (#423885) Journal

          It's necessary to limit APM and perhaps force the computer to "scroll" the screen or "click" the minimap in order to make a fair comparison. The existing Starcraft AI's ability to move almost all of the units instantly isn't interesting.

          It is a little weirder than the AlphaGo challenge because of the real-time nature of an RTS. Both AlphaGo and Lee Sedol got the same amount of time to work with. But even with a limited APM, the Starcraft AI won't be blinking, sipping mountain dew, or "thinking" about anything else. You give Lee Sedol 60 seconds, and he will probably think up a good move. With the RTS, you're living second to second. Maybe Civilization would have been a better choice?

          --
          [SIG] 10/28/2017: Soylent Upgrade v14 [soylentnews.org]