They have been able to generate predictive targeting system, that shoot where the target is heading

Andrew Vardy, AI researcher

But the contest is not really equal because the artificial intelligence behind the computer-controlled enemies has access to lots more information than is available to human players and can react much faster.

"AI techniques in games often cheat," said Andrew Vardy, an AI researcher from Carleton University, Ottawa, Canada. "Current methods of putting AI in games tend to be predictable and inflexible."

By contrast, the agents Mr Vardy is working on use genetic algorithms that mimic evolution to develop control systems that help them react to other elements, such as obstacles, ammunition or enemies in the game world.

Battle hardy

The control programs evolved by this method have been tried out in a simple world that pits six computer-controlled agents against one human player.

The current generation of smart agents did a good job of defending themselves against foes, said Mr Vardy.

"They have been able to generate predictive targeting systems that shoot where the target is heading," he said.

Mr Vardy said that the techniques could be used to create artificial players that learn the best way to defeat human opponents. The best artificial players could then be posted on the net for everyone to play.