[rael-science] Killer Bots Are Getting Human

 

~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
The Raelian Movement
for those who are not afraid of the future : http://www.rael.org
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Source: Science
http://www.sciencemag.org/cgi/content/full/330/6000/30-b

IEEE International Conference On Computational Intelligence And Games:

Killer Bots Are Getting Human

John Bohannon

The bot that won the third annual 2K BotPrize, a competition to create artificially intelligent game-playing agents that can fool a judge into believing they are human, represents a leap forward for game AI because the team used machine consciousness rather than just mimicking human behaviors.

It was standing room only in the computer lab as intense violence played out on a giant screen. The game was Ultimate Tournament 2004, the classic multiplayer first-person shooter. But not all of the avatars blasting at each other were controlled by humans. Half of them were bots programmed by scientists in the room, nervously monitoring their programs for crashes. This was the third annual 2K BotPrize, a competition to create artificially intelligent game-playing agents that can fool a judge into believing they are human.

The humans are dead! A Spanish team (right) won this year's 2K BotPrize. CREDIT: COURTESY JORGE MUÑOZ AND RAÚL ARRABALES MORENO [Larger version of this image]

The contest is a variation on a classic test, first proposed in 1950 by computing pioneer Alan Turing, in which a judge has a conversation with a human and a computer and must decide which is which. The Turing test still defeats artificial intelligence (AI) 60 years later; machines largely remain terrible conversation partners.

Action-based video games can offer an alternative Turing test. "They don't require speech, they provide a highly constrained environment but are still a challenge for AI," says Philip Hingston, the computer scientist at Edith Cowan University in Perth, Australia, who organized the contest. The rules are simple: Avatars controlled by humans and bots are dropped in a complex environment littered with weapons. It's kill or be killed. Each round, some of the human players—the judges—must decide which of their opponents are machine-controlled, based solely on their behavior. The team that designed the bot best at fooling the judges wins the $5000 prize and a trip to Australia, funded by the game company 2K.

This year's prize was scooped by Conscious Robots, a team of Spanish computer scientists. Its bot represents a leap forward for game AI, says Hingston, because the team used machine consciousness, a technique rarely applied because of its complexity. Rather than just mimicking human behaviors—such as using imperfect aim or introducing randomness into running routes—the team's bot was designed to think like a human. "In our approach, we try to effectively integrate several cognitive skills, like attention and learning," says Raúl Arrabales Moreno, a computer scientist at the University Carlos III of Madrid. The bot has a set of innate behaviors that are regulated by a higher control system, similar to the role of a conscious mind. It was incorrectly identified as human by the judges 32% of the time. By comparison, one human player was incorrectly identified as a machine 35% of the time. "There is only a slender gap between the humans and bots now," says Hingston.

"There has been significant progress since the 2009 competition," says Simon Lucas, a computer scientist at the University of Essex in the United Kingdom and one of the human players in the contest. Besides creating more engaging computer-controlled opponents for mass-market video games, the goal is to create better AI agents for "serious games" that simulate natural disasters and other complex problems (see above). Lucas predicts that a bot will be fully indistinguishable from human players "within the next 2 years."

---