WHAT DO THE U.S. ARMY, Hollywood animators, medical schools, and artificial intelligence researchers have in common with gawky teenage boys huddling over their PlayStations or Xboxes?

They are all fascinated by computer games. Indeed, the three-dimensional fantasy worlds and the behavior of the animated characters in them have become so realistic—and computers have become 
so capable—that games are now being used to train personnel ranging from Army troops to surgeons, as well as to advance the development of artificial intelligence (AI). It’s a field becoming known by the oxymoron of “serious games,” and it’s not just for kids.

Sixty serious engineers met in May to exchange the latest developments at the second IEEE Symposium on Computational Intelligence and Games held at the University of Nevada in Reno.

This learning program, which runs 
on the Internet, helps soldiers 
practice their perceptual attention while on patrol.

ARTIFICIALLY INTELLIGENT The basis for most serious games these days is artificial intelligence, a slippery concept at best, according to IEEE Associate Member Sushil Louis, director of the University of Nevada computer science and engineering department’s Evolutionary Computing Systems Lab. “Intelligence is in the eye of the beholder,” Louis says. “It’s hard to come up with an objective definition of intelligence, because subjective perceptions are what’s important. According to computer pioneer Alan Turing, if a computer program can fool a human into thinking that it’s human, then you would consider that program intelligent.”

AI has gained the edge in recent years over the so-called expert systems that were a leading area of research in the 1970s and 1980s for such potential applications as diagnosing diseases. Expert systems were supposed to embody all knowledge about a specific field, but their design proved to be what developers called “brittle,” Louis explains: “Their knowledge was out of date almost as soon as it was programmed into the system, and they were difficult to update 
because the systems wouldn’t learn.”

Those drawbacks sparked interest in AI systems, which were designed to learn from imitation, mistakes, experience, and trial and error—more the way people acquire knowledge rather than the way computers are programmed.

That’s useful in training, because the AI system can learn even as a player learns. Thus, the game itself can get tougher to play. “According to learning theory, you learn most from an opponent who is just slightly better than you are,” Louis says. “In serious games, where the goal is to have the player learn, AI allows the opponent to adapt as the player learns skills and strategies, so the game grows along with the player.”

TRAINING BEHAVIOR The U.S. Army is using AI in computer games not only for teaching military strategy and tactics but also for training soldiers to interact with others, says IEEE Member Michael van Lent, professor of computer science at the University of Southern California, in Los Angeles, and one of the keynote speakers at the IEEE conference.

“The Army is already good at training tactics. What’s difficult is teaching soldiers leadership skills and how to conduct negotiations in a stressful environment,” says van Lent, who is also a research scientist at USC’s Institute of Creative Technologies (ICT). The institute was initiated six years ago by the Army to blend Hollywood movie graphics with USC’s expertise in AI to 
create wholly realistic simulations. “Role-
playing games can immerse a soldier in the stress and emotion of the moment,” he added.

In one research demonstration ICT developed for the Army, “you’re a lieutenant trainee traveling with a platoon in Bosnia, and one of its vehicles injures a child,” van Lent recounts. “The animated AI characters in the game have their own goals, whose reactions depend on the trainee’s choices. For example, there’s a TV cameraman who adds to the stress by filming the episode, and there’s the child’s mother, who models more or less emotion as the trainee acts to get the kid to a hospital or to send the rest of his platoon ahead.”

How realistic do soldiers find the games? Van Lent says realism is achieved by collecting stories from soldiers returning from Iraq. “Stories more effectively communicate information to people than a list of dos and don’ts,” he says. And it’s the stories that ICT turns into game-playing training scenarios.

This soldier is playing a game developed for U.S. Army captains to teach them 
leadership skills.

In addition, developers are devising what they call “explainable AI,” where at the end of a training session the trainee can quiz each of the AI characters to see why it responded the way it did to each of the trainee’s decisions. This technique has the potential to be valuable in helping soldiers learn the Army’s detailed procedures for the “right” way to carry out every kind of order, from loading a rifle to building a bridge.

Other games funded by the U.S. Navy have the potential for treating soldiers with post-traumatic stress disorder.