Case study: Games students play

October 13, 2006

As paradoxes go, the concept of artificial intelligence must be one of the most unsettling for its sheer brilliance and potential. In theory, AI involves scientists trying to program a computer to instil intelligent behaviour and, at the other end of the scale, to create artificial consciousness through formal logic systems. The irony lies in the fact that human intelligence aims to build a thinking machine that outwits its creator.

Indeed, this is precisely what happened in 1997, when the then world chess champion, Garry Kasparov, was beaten by Deep Blue, a chess-playing computer. More recently, AI cars in the DARPA Grand Challenge drove themselves across several hundred miles of challenging terrain without any communication with humans to win a $2 million prize.

Such achievements are astonishing, but take Deep Blue and the prizewinning cars out of their environment and they are like a fish out of water. So the next step for computer scientists such as Bradford University's Peter Cowling is the generalisation of AI programmes.

"Deep Blue may be one of the strongest chess players in the world, but it can't tie its own shoelaces or hold a decent conversation. I think methods that can solve a broadening range of problems and learn from previous successes and failures are an important next step," he explains.

AI is already applied in scheduling and planning, but Cowling believes the major advances will be born out of creativity in games programming.

"Much of my own work is on general architectures for playing games. Once you have a decent player for, say, draughts, you can teach the program to play chess using some of the same principles in a different context."

One principle played out in game programming is usually more at home in the biological sciences: genetic algorithms that work by the Darwinian principles of evolution are fast becoming the tool of choice for games development. According to the principles of genetic algorithms, solutions to a programming problem are in effect artificial organisms with their own DNA. A number of these artificial organisms are allowed to reproduce and exchange DNA in the same way that a biological organism would create an offspring with a different complement of genetic material. The well-known survival-of-the-fittest principle kicks into action, with the most suitable artificial organism providing the best solution to the problem.

Despite sitting at the cutting edge of research, advances in computer science swiftly translate into commercial use. "We use genetic algorithms a lot, for creating schedules such as for mobile engineer workforces like BT repairmen, for finding strategies in a game-playing program and for evolving 'neural networks' that mimic the action of the brain in solving some problem. Others have used them to create new patents for electronics design," Cowling says.

Back in the classroom, Cowling is a firm believer in teaching AI through the games medium. "The latest challenge is to have students write programs that use a different approach - we encourage them to write games with tools such as Game Maker ( ) so that they can control game agents by writing AI rather than controlling the agent themselves."

Cowling's enthusiasm appears to be rubbing off on the students. Bradford's AI for gaming module is by far the most popular subject among undergraduates, and the university is hosting a new MSc in AI for games from 2007.

Back to Microsoft index page

Register to continue

Why register?

  • Registration is free and only takes a moment
  • Once registered, you can read 3 articles a month
  • Sign up for our newsletter
Please Login or Register to read this article.