Turing test thoughts

Forum Forums Discussion Turing test thoughts

This topic contains 2 replies, has 3 voices, and was last updated by Carl Arbogast Carl Arbogast 5 years, 2 months ago.

Viewing 3 posts - 1 through 3 (of 3 total)
  • Author
    Posts
  • #17963
    quberewt
    quberewt
    Participant
    0

    Hi guys,

    I’m very new here, but I am singularly impressed with the work that is done in this format.

    This might be to meta, but I’d like your guys thoughts on the “Turing Test” and how it applies to the work you’re doing. At what point will a program like Miqobot become indistinquishable from a “live” player.

    At this point, I don’t know, and would like you guys thoughts on the matter. This interests me.

    #17964
    Miqobot
    Miqobot
    Keymaster
    4+

    Miqobot will be able to pass the Turing test when she starts doing all activities that a human player normally does.

    • Creating a character with unique visual features.
    • Completing Main Scenario without any cheats or internal quest databases.
    • Completing dungeons and trials with other players, without the necessity of a bot-premade party.
    • Defining and expressing unique social behavior.
    • Maintaining alternative character progression options, such as randomly playing other jobs and unlocking additional content.

    Implementing these features would defeat the purpose of Miqobot project, because there would be no reason for a human to play the game anymore.
    Miqobot is designed to help players skip through boring grind, but still enjoy the most interesting and rewarding aspects of the game.

    We sincerely hope that the day when Miqobot is able to pass the Turing test will never come.

    #17965
    Carl Arbogast
    Carl Arbogast
    Participant
    0

    Maybe then you will find useful this quote from Miqobot in 2016:

    The day when Miqobot will be able to clear end-game without supervision will probably be the day when machine becomes equal to a man πŸ™‚

    It was right after Miqobot announced combat features (DPS Solver, Enemies priority list and new 3D radar) in development along Raid support for this purpose:

    According to our preliminary research, this feature should tremendously reduce the amount of stress imposed by end-game raids.

    It was in a topic talking about the possibility of a Roll Solver where Miqobot presented an analysis of Square Enix systems, and how it was indeed possible to predict the roll, but that the cost of the implementation of such a feature was close to the price of a space station.
    Since no user agreed upon raising the Miqobot monthly fee to several millions dollars, Miqobot came up with an alternative solution, known today as the Combat Assist.

    There is a reason to why this myth exists.
    In theory, it is indeed possible to predict the outcome of a random roll.
    But in practice, it would take some tremendously expensive equipment which is comparable to spaceships in price.

    The root of this myth is that random numbers in games are not random.
    They are pseudorandom.

    It means that Random Number Generator uses a clearly defined pattern which is built in such a way that it looks random to an outsider. The simplest generators can be defined with a single formula:

    Xn+1 = (214013 * Xn + 2531011) mod 232

    This gives a sequence of numbers X1, X2, X3, …, which seems very random but in fact it is not. And yes, of course it is calculated server-side.

    Final Fantasy probably uses more complex algorithms, but it’s not the only obstacle here.
    The main problem is that random numbers are generated in enormous quantities.

    For battle, every single hit is randomized (+ or -).
    For monsters, their idling behaviour is randomized.
    For crafting, lots of abilities have a success rate, and it’s randomized.
    For gathering, every hit on the node is randomized.

    The list may go on. Considering that all the players are doing all these things together at the same time, the ingame RNG must be producing more than 1,000,000,000 numbers per second. And in order to be able to predict them, we’d have to calculate even faster.

    Basically, in order to implement the RNG solver, it is required to:

    Find out the exact formula that game server uses. At the very least, we could purchase the source code from game developers themselves.
    Connect to the server hardware and take readings of server processor cores – in order to determine the current state of RNG. This would probably require a direct channel and some sort of API provided by game developers.
    Perform calculations at a rate faster than RNG does. Preferrably around 1,000 times faster. Which brings us somewhere in the range of supercomputers.

    So as alluring as this idea sounds, unfortunately we have to set this request aside.
    But still, thank you for bringing up such an interesting topic πŸ™‚

    P.S. In order to solve the problem with grinding Lanner Mounts, we are planning to implement a Raid Grinding Bot.
    It will require a very complex AI, and probably a lot of testing.
    But this way, nobody will have to cheat πŸ™‚

    • This reply was modified 5 years, 2 months ago by Carl Arbogast Carl Arbogast.
Viewing 3 posts - 1 through 3 (of 3 total)

You must be logged in to reply to this topic.