He played like a man

September 30, 2012 0:15

Turing Test in Unreal Tournament 2004.  Photo: dailytechinfo.orgArtificial intelligence system that is implemented as a virtual player, so-called "bots", the researchers programmed the University of Texas at Austin, has won the prestigious award «BotPrize» tournament game Unreal Tournament 2004, sponsored by 2K Games. But this is not the most important thing, the AI-bot made an indelible impression on the judges with their abilities and actions indistinguishable from the actions of real people experienced players.
"The idea of the competition« BotPrize »is to create a non-human characters in a computer game (nonplayer characters, NPC), which are controlled by artificial intelligence algorithms, which should mimic the real people and their actions as best as possible" — says Professor Risto Miikkuleynen (Risto Miikkulainen) who, along with their graduate Jacob Shrumom and Igor Karpov has focused on building the AI bot UT ^ 2 for the game Unreal Tournament.

As you can see from the video below, the AI bots are measured by each other and with an equal number of people on the players. As in the original game, UT ^ 2 is only one goal — to survive and defeat as many enemies. But every man, player, in addition to the primary virtual weapons, has also so-called "judicial weapon", using which they can mark any other player as a person or as an AI bot.

Naturally, the winner of the competition «BotPrize» is that the AI-bot, who received the maximum number of "human" evaluations from people-players, who are also the judges of this competition. Texas AI bot UT ^ 2, who participated in the preliminary events in the last month, tied for first place with an AI bot MirrorBot, the author of which is a Romanian programmer Polkianu Mihai (Mihai Polceanu). But the most incredible is the fact that both the AI-bot got the "assessment of humanity" at 52 percent, while the real human players have not received an average rating above 40 percent.

In fact, the gameplay is quite complex and three-dimensional virtual environments require the perfection of artificial intelligence bots, whose task is complete imitation of the actions of people, including the movement in three-dimensional space, a chaotic battle against a variety of opponents and choose the best strategy in any given game.

"Artificial intelligence can even simulate unreasonable and far from optimal behavior at some point in the game, because sometimes people got mad at any of the opponents continue his stubborn pursuit, despite the more appropriate tactics, more profitable purposes and for their own interests" — Jacob says Shrum. — "Some aspects of the behavior of AI-bot modeled us with samples of human behavior, but most of them is the product of a process called us neyroevolyutsiey. This process is based on the self-learning neural networks, artificial intelligence, evolving through the usual procedure of natural selection in a game environment "

"When this competition, this kind of" game "the Turing test was conducted for the first time, our goal was to obtain the 50 percent estimate humanity bot. It took us five years to create the perfect artificial intelligence systems, and finally, we have reached the goal. And, I'll tell you, it was not a fluke, "- says Professor Miikkuleynen.

And, in conclusion, remember that the so-called Turing test is the underlying technology, which is used to determine the level of artificial intelligence. Its purpose is to mislead the person who interacts with another person at the same time and with a system of artificial intelligence.

Like this post? Please share to your friends: