The Turing Test is a rubric proposed by computing pioneer Alan Turing: Any computer that can fool reasonable people into thinking they're dealing with another person is, for all practical purposes, artificially intelligent.
While many academics invest years creating machines that can meet Turing's criterion, Kenneth Newby, a programmer at Electronic Arts, has a different take.
"I prefer to loosen that characterization somewhat in a borrowing from the world of art," said Newby. "If you can't tell the difference, or if the differences become subtle or insignificant enough, you're willing to suspend your disbelief temporarily."
Computer-game designers face that challenge every day: how to imbue their work with enough intelligence so players will involve themselves in fantasy worlds. From the outside, it may appear that gaming companies are giving academic theorists a run for their money, at least in terms of practical AI applications.
At least one gaming engineer thinks so. "I don't think I've seen anything in academia that really comes close to Half-Life for creating a populated interacting environment," said Gabe Newell, founder and managing director of Valve. His company is releasing Half-Life, which won the Best PC Game award at this year's E3 convention. The industry buzz is that the game could herald a new generation of "first-person shooters," a characteristically violent action genre heralded by games like Quake and Doom.
Networked play has long been the main attraction for first-person shooters. Legions of devoted warriors spend their free time chasing others in underworlds of fantasy, blasting away enemy players or teams. Solo play pales when it's set against the range of action and smart unpredictability of real opponents and allies. Valve and other game companies are working to fix that.
Newell says that the characters in Half-Life will talk to each other, respond to the world around them, and do things like healing an injured player or opening doors for which the player doesn't have the right security clearance. "If told to, they will follow the player around the world," Newell said.
Another gaming company, Bungie, experienced early success from first-person shooters. But its recent Myth series has moved the company into the realm of strategy games.
A top priority for programmers of this genre, like WarCraft or Age of Empires, is increasing the intelligence of units under a player's control. The developers of the soon-to-be-released Myth II plan to go one better by raising the intelligence of the game's creatures. Balancing players' demands with technical limitations is the name of the game.
"Academics have fluffy ideas, but we're here in the trenches dealing with the real world," said Quinn Dunki, a senior engineer at Bungie and a strong advocate for AI. Nevertheless, game designers borrow from academic research.
Newell said that specific algorithmic techniques used extensively in games -- such as neural networks and genetic algorithms -- were originally developed in academia, but others are coming from commercial gaming. "Any [techniques] that are developed in the gaming world tend to be kept fairly quiet for competitive reasons and are usually a lot more ad hoc," he said.
That competitive silence worries those who want to ensure the progress of the field as a whole. While research within a company might prove valuable for that firm, that secrecy can impede progress. Henry Lieberman, a professor at the Massachusetts Institute of Technology who specializes in AI and autonomous agents, emphasized the problems for scientific advancement if all research and techniques were the province of the corporate world.
"I have nothing but respect for the programmers of the gaming industry, who do some amazing things," said Lieberman. "However, it would be hard for science to progress solely that way, because gamers don't typically publish their techniques or even share knowledge with each other."
Bungie's Dunki said that gamers are only extrapolating from what academics have already discovered. Her experience has been that given a new project, engineers immediately hit the Web to peruse academic resources.
"This is a really great time for the sharing of information," said Dunki, who points to researchers' penchant for publishing their efforts on the Web. Because of the ease of finding material, Dunki usually begins a major project by surveying up to 100 academic papers on the topic in question.
She believes that strong AI could improve certain aspects of the games, like creating more challenging opponents and interesting sidekicks. It might also enhance the game's camera or point of view, as well as the graphics engine making smart rendering decisions. Newby, at Electronic Arts, already uses algorithms to generate crowd noises in simulated ballparks.
For now, most of these improvements are on hold while game companies focus on developing strong visuals, Dunki says. "Slowly, the novelty is wearing off with the eye candy, then we'll get back to focusing on gameplay."
Newell described the excitement that AI can generate.
"When we were playtesting Half-Life, we had one young woman who discovered that when she fired off her machine gun, she could scare the other scientists," he said. "The scientists would startle and start running away. She'd chase after them, cackling, and fire off her machine gun again. Then she noticed that when a scared scientist would encounter another scientist, the new scientist would learn what had scared the first and start running as well. From that point on, she pretty much ignored what she was 'supposed' to be doing and spent her time creating scientist stampedes.
"This was all great fun until she accidentally shot one of the scientists, and a nearby scientist started yelling at her 'Stop it! What are you doing? He's a friend!' She recoiled and actually pushed her chair away from the game and had this horrified and guilty expression on her face, which was pretty much the end of her playtest."
AI in gaming, Newell said, is just reaching the point where these kinds of spontaneously generated, emotionally significant events can happen. If gamers are as excited as he is, virtual beings in games may soon become as confusing, surprising, and interesting as other humans. At least more interesting than those in chat rooms.
