OpenAI Bot Remains Undefeated against World’s Greatest Dota 2 Players

Discussion in '[H]ard|OCP Front Page News' started by Megalith, Aug 12, 2017.

  1. Megalith

    Megalith 24-bit/48kHz Staff Member

    Messages:
    10,150
    Joined:
    Aug 20, 2006
    The machines have won again: last night, OpenAI’s Dota 2 bot beat Danil “Dendi” Ishutin at The International, one of the biggest eSports events in the world, and remains undefeated against the world’s top Dota 2 players. The bot was trained by simply copying the AI and letting the two play each other for weeks on end. While this was a less complex, 1v1 battle, OpenAI is working on another bot that could play against and alongside humans in a larger 5v5 battle.

    “We’ve coached it to learn just from playing against itself,” said OpenAI researcher Jakub Pachoki. “So we didn’t hard-code in any strategy, we didn’t have it learn from human experts, just from the very beginning, it just keeps playing against a copy of itself. It starts from complete randomness and then it makes very small improvements, and eventually it’s just pro level.” Not shockingly, Elon Musk was watching along and had some thoughts of his own, calling unregulated AI vastly more dangerous than North Korea.
     
    Last edited: Aug 12, 2017
  2. Ultima99

    Ultima99 [H]ardness Supreme

    Messages:
    4,248
    Joined:
    Jul 31, 2004
    At some point we'll all become worthless.

    Robots and AI will be able to do everything a human can (and much more) faster, stronger, better, and smarter.
     
  3. Seventyfive

    Seventyfive Gawd

    Messages:
    874
    Joined:
    Jul 14, 2004
    I feel like a MOBA is a perfect setting for an AI bot since it can tell if a spell is within range to within 1 pixel, it can track all ally and opponent cool downs to the millisecond, it should be able to last hit basically every single mob since it will know to the exact number whether their auto attack will do enough damage to finish the minion. There's a lot of simple ways that an AI could easily outplay a human.
     
    Nytegard and MrGuvernment like this.
  4. Wolf-R1

    Wolf-R1 [H]ard|Gawd

    Messages:
    1,893
    Joined:
    Aug 30, 2004
    LOL Like this is a shock?? Computers can calculate moves, act and react far faster than people can. Once you teach an AI how to do something how is it still a shock that it does it better than people??
     
    Makaveli@BETA likes this.
  5. lolfail9001

    lolfail9001 [H]ard|Gawd

    Messages:
    1,546
    Joined:
    May 31, 2016
    Well, trick here is that it learned to do it on it's own.

    Update: people figured out that you can just confuse it. I wonder how long until people understand that even with deep learning, quick adaptation is not AI's strong point.
     
  6. nomu

    nomu Gawd

    Messages:
    770
    Joined:
    Jul 30, 2006
    Well the reference point was Valve's built-in DOTA AI which can't do anything competently, so the players weren't prepared for how thoroughly they got trashed.
     
  7. Maxx

    Maxx Gawd

    Messages:
    996
    Joined:
    Mar 31, 2003
    "There are some things that can beat smartness and foresight? Awkwardness and stupidity can. The best swordsman in the world doesn't need to fear the second best swordsman in the world; no, the person for him to be afraid of is some ignorant antagonist who has never had a sword in his hand before; he doesn't do the thing he ought to do, and so the expert isn't prepared for him; he does the thing he ought not to do; and often it catches the expert out and ends him on the spot." -Mark Twain

    To be fair, when I play against an AI I play differently than I do against human players. I wonder if some of these guys don't remember what AI is like...
     
  8. Nytegard

    Nytegard 2[H]4U

    Messages:
    2,566
    Joined:
    Jan 8, 2004
    "Vastly more complex than traditional board games like chess & Go." - Elon Musk

    I'm sorry, I just disagree here. Yes, in a technical sense, he's correct. There are infinite possibilities of what you can do in Dota. That doesn't mean though that most of them are useful. There are only so many valid strategies. I went ahead and watched the video, and it becomes crystal clear that the computer is making split second decisions that Dendi just wasn't humanly capable of matching. (E.g. The AI knows precisely when to attack him, whereas you see him often click, but not be within range).

    A player has to learn, and once he learns, he still has to execute. The execution is what determines a good player from a bad. The computer just has to learn. The execution is the easy part for it.
     
  9. w4ffles

    w4ffles 2[H]4U

    Messages:
    2,073
    Joined:
    Mar 13, 2008
    It reminds me of the scene in the Robocop reboot where they talk about how humans are inferior because they part of the brain needs to weigh decisions before making a judgement. There's no way a human can beat the reaction speed (nor precision) of a machine.
     
  10. blkt

    blkt Limp Gawd

    Messages:
    474
    Joined:
    Oct 9, 2009
    idbuythatforadollar.gif
     
  11. katanaD

    katanaD Limp Gawd

    Messages:
    185
    Joined:
    Nov 15, 2016
    whats up with the robe the kid is wearing in that pic? like he is some boxer or something

    LOL
     
  12. nomu

    nomu Gawd

    Messages:
    770
    Joined:
    Jul 30, 2006
    They did entrances like it was a boxing match. They brought out a mid-tower on a rolling cart under a robe too.
     
  13. M76

    M76 [H]ardness Supreme

    Messages:
    4,933
    Joined:
    Jun 12, 2012
    In AI terms this is infinitely more complex than a chess game. A chess game has simple rules and a board where every chess piece has an easily determinable position and function and threat level. A map of an action game is like a chess board with billions of fields that is not even arranged by any order.
    The rules and goals of chess is relatively easy to define in mathematical terms. Now try that with dota.
     
  14. w4ffles

    w4ffles 2[H]4U

    Messages:
    2,073
    Joined:
    Mar 13, 2008
    Here's a simplified version of how the AI learned.