Forgot your password?
typodupeerror
AI Games

Two Teams Win the BotPrize 56

Posted by Soulskill
from the just-make-them-swear-a-lot-over-xbox-live dept.
An anonymous reader writes "For the past five years, the 2K BotPrize has challenged artificial intelligence researchers and programmers to create a computer-game-playing bot that plays like a person. It's one thing to make bots that play computer games very well — computers are faster and more accurate than a person can ever be — but it's a different thing to make bots that are fun to play against. In a breakthrough result, after years of striving and improvement from 14 different international teams from nine countries, two teams have crossed the humanness barrier! The teams share $7000 in prize money and a trip to games company 2K's Canberra studio. The winners are the UT^2 team from the University of Texas at Austin, and Mihai Polceanu, a doctoral student from Romania, currently studying Artificial Intelligence at ENIB CERV — Centre de Réalité Virtuelle, Brest, France. The UT^2 team is Professor Risto Miikulainen, and doctoral students Jacob Schrum and Igor Karpov. The bots created by the two teams both achieved a humanness rating of 52%, easily exceeding the average humanness rating of the human players, at 40%. It is especially fitting that the prize has been won in the 2012 Alan Turing Centenary Year. The famous Turing test — where a computer has to have a conversation with a human, and pretends to be another human — was the inspiration for the BotPrize competition. Where to now for human-like bots? Next year we hope to propose a new and exciting challenge for game playing bot creators to push their technologies to the next level of human-like performance."
This discussion has been archived. No new comments can be posted.

Two Teams Win the BotPrize

Comments Filter:
  • by SuperKendall (25149) on Saturday September 15, 2012 @11:22PM (#41349855)

    Can't have been too hard to make a bot that spews random obscenities and anti-gay/misogonistic comments, while randomly firing bullets into team members and tea-bagging you when you die.

    That gets you up to what, 75%?

  • by khasim (1285) <brandioch.conner@gmail.com> on Saturday September 15, 2012 @11:30PM (#41349899)

    The UT^2 team is Professor Risto Miikulainen, and doctoral students Jacob Schrum and Igor Karpov.

    IGOR! I need a new brain for the robot!

    • Master? Can I change your mind? I like abynormal so much better.

      No Igor, you can't change my mind. I just changed it and want you to bend over and take it like a bot

  • While it may lack in variety, I often prefer to spend what little time I do have to play games in skirmish offline or against bots. I find that better than dealing with most of the bullshit online. Yeah I know your counterargument, if I spent more time finding good players or servers it isn't really a problem. I just don't want to. Bots don't go berserk or disconnect or lack (albeit often predictable once you've played a game enough) skills, or any of the other usual multiplayer complaints.
    • I often prefer to spend what little time I do have to play games in skirmish offline or against bots. I find that better than dealing with most of the bullshit online.

      I used to play Unreal Tournament a lot. I liked to play online, there was a lot of bullshit but in the end I just felt like playing against bots was too fake.

      I don't know why that is, because I love many single player games where essentially I'm playing a whole game against bots and carefully tuned algorithms meant to fool me into thinking I

      • Some of the old classic arcades had this option. I was a passable midline Mortal Kombat II-III player before I retired from most video games. A couple of fun local shops had the setting on medium. On a good day you could beat the game with a couple of bucks. Then on road trips some of the other shop owners were greedy and cranked the difficulty setting, and that became NO FUN AT ALL. I don't recall the specifics, but a couple of the characters just got way too fast with the Throw maneuvers, Goro and Motaro

  • by martin-boundary (547041) on Saturday September 15, 2012 @11:52PM (#41349969)

    easily exceeding the average humanness rating of the human players, at 40%.

    Whoa! Who woulda thunk that playing computer games makes people less humann?

    • by SeaFox (739806)

      Finally, a story I can tag "morehumanthanhuman" and have the tag be a literal statement.

  • I wonder if part of how they did this used human game play to train a neural network. A quick google of the idea makes it seem like that is plausible and has been done in at least one study, but I don't know if it is a widely used AI technique or not.
    • It looks like someone else has taken tech like that and done something geared toward finding biomedical applications for it. The ideas and experiments in this abstract [ieee.org] could perhaps be useful as a building block for some sort of basic tricorder. Reading the body's reaction to drugs and stress, screening neuropsych deficits, and possibly lots more by just simply scanning them with one single machine? That's Star Trek stuff.
    • It's hard to imagine that naively training a neural network with input from human players could produce play that's more characteristically human than those players.
      I'm not saying that these teams were limited to simple techniques, just that using a neural network isn't the "secret sauce/"

  • If a bot can be "more human" than actual humans, what's next?

    Robot Love!

  • by Baloroth (2370816) on Sunday September 16, 2012 @01:28AM (#41350229)

    ...or the judges are very bad at distinguishing human from bot. One interesting thing to note is that lower skill default bots were rated quite highly on the "humanness" rating (higher than the average for humans), which might suggest the judges thought human players are worse than bots. The default bots "humanness" average was only slightly below the average for the actual human players (~37% vs ~41%), which suggests the methodology is a little questionable. If you can't distinguish the default, "non-humanized" bots from actual humans, how would you expect to distinguish bots that have been intended to be "humanized"?

    • by Hentes (2461350)

      UT bots were always designed with the goal to feel human, altough playing a lot against them would teach you the way they work. But it's not nearly as easy as distinguishing between a chatbot and a person. For example, sometimes even grandmasters get fooled by amateurs cheating with the help of a chessbot, because it's very hard to differentiate between a computer and a human playing a game.

    • by CastrTroy (595695)
      I think the problem is that some of the better players I've played against in my day play quite a lot like robots. They know the map very well, they follow a defined path making sure to hit every power-up. They know the hiding spots and always check them. They make seemingly impossible impossibly jumps and never miss a step. They aren't really fun to play against either.
  • http://www.nytimes.com/2012/09/16/opinion/sunday/kristof-the-foreign-relations-fumbler.html [nytimes.com]

    "Mitt Romney spent the last week blowing up his foreign policy credentials to be president"

    Is Mitt a twit or actually a candidate for the BotPrize? Would Mitt stand a Turing test?

    • Nice, Obama brags about killing Bin Laden and kills four embassy workers as a result (including a renowned EVE player), and YOU complain about ROMNEY's foreign policy?

      Romney is a genius compared to "Watch the Middle East Crumble" Obama.

      Here's a thought, I'll bet Romney could figure out you just might want to guard a U.S. Embassy in Libya with more than a single EVE player.

      I'm also pretty sure Romney would not be arresting film makers no matter what subject matter they used. But if you hate free speech, by

  • by Dean Edmonds (189342) on Sunday September 16, 2012 @03:02AM (#41350499)

    I find it interesting that the ordering of judges on the "Most human humans" list is the exact opposite of those on the "Best human judges" list. So the more robotic a judge appeared to others, the better they were able to recognize the true bots in the games. A great example of "it takes one to know one".

  • by JimboFBX (1097277) on Sunday September 16, 2012 @03:15AM (#41350527)

    Ut2004 is a chaotic game. Very bot friendly especially since you never see a player's perspective (judging was done in game using a tag gun). I'd be more impressed if a bot could recreate the muscle memory twitch and intelligence of a counterstrike player via the first person perspective.

  • by geoffaus (623283) on Sunday September 16, 2012 @03:57AM (#41350639) Homepage
    Dear winners, Please dont judge Australia by what you see in Canberra. Its a soul sucking dreadful place. And its also full of our federal politicians. Please escape and go somewhere good - there are lots of great places in Australia. Canberra is not one of them.
    • by jamesh (87723)

      As an online discussion grows longer, the probability of a reference to Godwin's Law approaches 1

      Didn't Hitler say that?

  • The one defining characteristic of human players in most multiplayer games is that 70% or so are completely clueless about tactics, strategy and teamwork. Add some insulting behavior to that, and you have the perfect emulation of a human player...

  • by Anonymous Coward

    You can download the source for the UT^2 bot entry (written in Java using the Pogamut framework) from http://nn.cs.utexas.edu/?ut2 . The bot has a modular behavior architecture and uses evolved neural network controllers for some of its behaviors and a playback of human game traces (available as a separate download) for others.

While money can't buy happiness, it certainly lets you choose your own form of misery.

Working...