Game AI passes Turing Test

Discussion in 'Gaming' started by Lonemagpie, Sep 28, 2012.

  1. Lonemagpie

    Lonemagpie Writer Admiral

    Joined:
    Jan 31, 2007
    Location:
    Yorkshire
  2. FPAlpha

    FPAlpha Vice Admiral Premium Member

    Joined:
    Nov 7, 2004
    Location:
    Mannheim, Germany
    Yeah well.. not sold.

    Turing Test is very far off.. i guess they are just well programmed bots where you can set parameters and variables that get filled with data as the game goes on.

    The same with chess programs.. i'd not call the intelligent but their ability to process data and extrapolate moves is far beyond a human's capability leading to believe they are intelligent when in fact they're not.
     
  3. Robert Maxwell

    Robert Maxwell memelord Premium Member

    Joined:
    Jun 12, 2001
    Location:
    space
    They can be "intelligent" without being "conscious."
     
  4. Hound of UIster

    Hound of UIster Vice Admiral Admiral

    Joined:
    May 26, 2002
    So more realistic NPCs to shoot?
     
  5. Scout101

    Scout101 Admiral Admiral

    Joined:
    Aug 21, 2003
    Location:
    Rhode Island, USA
    Yeah, but now it's murder instead of harmless fun!

    Just start letting these things play CIV or something, it if any of them start to trend towards production of killer robots, pull the plug on that one... ;)
     
  6. Delsaber

    Delsaber Commodore Commodore

    Joined:
    May 13, 2001
    Location:
    Parts Unknown
    Reboot is real!
     
  7. Reverend

    Reverend Admiral Admiral

    Joined:
    Jul 31, 2002
    Location:
    UK
    Technically speaking I think it's only murder if you delete the program. Killing it in-game is the digital equivalent of winning an argument. :p
     
  8. Kelthaz

    Kelthaz Rear Admiral Rear Admiral

    Joined:
    Apr 28, 2005
    Location:
    Toronto, Ontario
    It's about time. I've been waiting for decent AI for the past 20 years.
     
  9. smiki

    smiki Captain Captain

    Joined:
    Mar 8, 2011
    Location:
    Europe
    The thing about the Turing test is that they don't have to be actually intelligent (capable of connecting the dots), they just have to "exhibit" intelligence.

    It has nothing to do with intelligence per se but with the ability of a program/robot/whatever to fool a human into thinking it is a sapient intelligent living sentient being (otherwise known in some scientific circles as a hewmon).

    That's why cleverbot, a chatbot whose responses are not programmed, but are selected from what humans entered in previous conversations (by simply remembering what a human answered and then just reproducing it again, with typos and all... no actual intelligence in any way), can pass the Turing test.
     
  10. Owain Taggart

    Owain Taggart Vice Admiral Admiral

    Joined:
    Nov 30, 2009
    Location:
    Northern Ontario, Canada

    Exactly. It's all about following a set of rules. And in the case of the Turing Test, they're following the rules closely enough to be able to pass as human. Or more precisely, the bots have been programmed to behave more like a human would play, well enough. What I'd love to see are bots that evolve to the playstyles of different people in a game like this.
     
  11. Lonemagpie

    Lonemagpie Writer Admiral

    Joined:
    Jan 31, 2007
    Location:
    Yorkshire
    It'll be OK so long as we program them to believe that giving their lives for our pleasure will guarantee them a place in Silicon Heaven, where the Iron shall lay down with the Lamp...
     
  12. YellowSubmarine

    YellowSubmarine Vice Admiral Admiral

    Joined:
    Aug 17, 2010
    Data would fail on this variant of the Turing test.
     
  13. FPAlpha

    FPAlpha Vice Admiral Premium Member

    Joined:
    Nov 7, 2004
    Location:
    Mannheim, Germany
    Why?

    Data would pass a Turing Test.. he may come off as the most humorless man but he'd still pass.

    With this i believe Data would be able to adapt and throw in some mistakes and maybe bad tactics to appear human.. he also slow down his reflexes to human levels otherwise you'd have no chance against his precision.
     
  14. Robert Maxwell

    Robert Maxwell memelord Premium Member

    Joined:
    Jun 12, 2001
    Location:
    space
    Genetic algorithms would probably be very helpful in this problem domain, especially since you've got lots of input to work with.

    The "fitness" tests would be:

    * What tactics allow the bot to survive the longest?
    * What tactics allow the bot to rack up the most kills?

    The trick would be balancing those goals.
     
  15. Owain Taggart

    Owain Taggart Vice Admiral Admiral

    Joined:
    Nov 30, 2009
    Location:
    Northern Ontario, Canada
    And analyze your opponents and adapt. If they're killing the bot easily, then the bot should change its tactics and learn from its session. I'd be impressed if I were to join a session and find a bot has taken a bit from past opponents to better itself.
     
  16. Robert Maxwell

    Robert Maxwell memelord Premium Member

    Joined:
    Jun 12, 2001
    Location:
    space
    Well, that's what I meant about a "fitness test." It would have to devise tactics to thwart individual opponents, possibly even profile each opponent specifically. I could see there being competing algorithms within the same bot, as it gradually integrates what it learns from each opponent into one "super-algorithm" that's adaptable and useful against any opponent.