Published on September 30th, 2012 | by Steven Hodson0
The “Turing Test” Gets Passed By – AI Game Bots
For those not familiar with the “Turing Test” it is the test used to determine if machines, or computers, can exhibit intelligent behavior. This is decided by a human judge engaging in conversation using specific question with both an unseen human and unseen “machine” and the supposition being that if the human judge can not tell from the answer being given which is the human and which is the machine then at that point the machine can be considered to be intelligent.
This is an exceptionally simplistic, and probably wrong in some parts, explanation of the test but it would seem that this moment has been reached by some computer scientists at The University of Texas at Austin. In a contest that took place within Unreal Tournament 2004 the computer scientists were able to create an in-game bot that the other players in the game really believed was a human being.
When the human players were asked to tag other characters in the game; which included the bot-controlled characters, it was found that the winning bot – UT^2 – got a score of 52%.
Professor Risto Miikkulainen said his goal was to achieve a score of at least 50%. UT^2 actually shared the victory with the equally impressive MirrorBot, which was created by Romanian programmer Mihai Polceanu. Both winning teams split a $7,000 grand prize.
Interestingly enough, the average “humanness” rating that human players received was just 40%, meaning UT^2 played more like a human — or at least, other players’ perceptions of what a human should play like — than actual flesh-and-blood gamers.
Now about that whole Skynet thing……