There Is Research To Be Done

As researchers go, Robin Baumgarten isn’t especially unusual. He sits down at his desk, does his calculations and runs his experiments. Tucked away behind the leafy trees and aging museums of Imperial College, London, Robin’s duties have a rather unnatural side-effect. When Robin begins a simulation, millions of people die. And that’s just if he’s successful.

Robin’s not a virologist, he’s not working with Ministry of Defence funding and he doesn’t have a secret underground lair. In fact, ask him what he works on during the day and you’ll probably drift off after the first few syllables. Robin is a researcher in the Department of Computing at Imperial College. And the Petri dish for his most recent work? The cult multiplayer strategy game DEFCON. For him, nuclear holocaust is just part of the equation.

Robin’s work represents a new thrust in academic institutions the world over to use videogames as a test bed for cutting-edge research, and for the game industry to reap the rewards in return. For a medium that began in the supercomputers tucked away at the Massachusetts Institute of Technology, the thought of gaming once again gracing the halls of science and engineering is an exciting one, and Robin represents a key part of building the relationships that will facilitate this exchange of ideas.

The project initially only aimed to replace the standard DEFCON A.I. with a more capable model, but this simple idea led to significant results, both for Robin, who composed a master’s thesis from the work, and for Introversion, who will soon have tools that they can release to players to let them write their own A.I. for the game. “When we have more people writing A.I. for this game,” Robin enthuses, “we’re hoping to run competitions at conferences and other events.”

The origins of the project are more humble than you might think. Introversion approached Robin’s project supervisor, Dr. Simon Colton, some time ago at the Games and Media Event, a one-of-a-kind conference held annually at the university.

“Mark Morris [of Introversion] came up to me over a beer and said, ‘We’ve been thinking of getting research going with universities,'” Dr. Colton explains. “One thing led to another. The major breakthrough was when Introversion said they would allow us full access to their code for DEFCON.” With funding and resources secured, Robin’s project culminated in an improved A.I. and scope for further work. Since then, Dr. Colton has undertaken research projects worth hundreds of thousands of pounds with companies such as Rebellion, the developers behind Alien vs. Predator: Requiem and Rogue Trooper.

Meetings such as those at the Games and Media Event are crucial networking opportunities that open up relationships for funding and inspire work like Robin’s. Now four years old, GaME has played host to a spectrum of gaming luminaries, including David Braben, Peter Molyneux and of course the Introversion team. But just as importantly, it’s also had software engineers from Rare discuss the complexities of Xbox 360 programming, the founder of Codeplay chatting about multi-core games creation and a host of academics demonstrating the cutting-edge results of their research projects. GaME has science and technology at its heart – something that not only separates it from many similar networking events, but also makes it a perfect breeding ground for this very specific sort of cooperation.

One of the academics who has demonstrated at GaME in the past is Maja Pantic, a reader in human-computer interaction who presented her work in automated facial expression analysis, something that she’s hopeful will one day be used in games to assess the player’s mood. It might seem blue-sky to think that a game like Valve’s zombie fright-fest Left 4 Dead could one day regulate enemy spawn patterns based on how scared the player looked, but this is very much the beauty of an event driven by research rather than profit.

Paul Kelly is the Professor of Software Technology at Imperial and one of the chief organizers of the project. He’s a believer that relationships between industry and academia can really benefit both sides. “What [universities] are really doing is looking for compelling instances of general problems where our research ideas can make a difference. By developing a good long-term relationship with interesting university research groups, a company can both tap into the research community and sometimes also guide it towards interesting problems.”

However, as with any collaboration where large sums of money are involved – and note that some grants being issued for research work relating to the videogames industry are quickly approaching seven figures – there’s a fine line between taking part in academic investigation and outsourcing development work to students. “The key thing is to recognize that they’re not buying results,” Professor Kelly emphasizes. “They’re investing in a relationship.”

Of course, these collaborations don’t always result in any tangible benefits for gamers. Leave the leafy confines of London and travel across the sea, and you’ll find two academics that used their research to deliver an indie gem.

Talk to the average gamer about Façade, and they’ll probably be able to tell you that there’s some unusual artificial intelligence at work behind this sparse dinner party sim. They’ll also probably tell you it’s not all that good of a game. Yet despite this it’s won awards at the Independent Games Conference and the Slamdance Games Festival. Since its launch in July 2005 over half a million gamers have downloaded this extremely unusual interactive experience.

Façade sits you down outside the house of your two close friends, Grace and Trip. They’ve invited you around for a dinner party, and how the one-act drama pans out is largely determined by how you act. To speak, you simply type, and whoever is in earshot will react to what you say as best as they can.

Inevitably, their best is often not good enough, but what’s hard to appreciate when playing the game is just how much work the program is doing to generate an interesting experience on the fly. The relationships, events and, ultimately, outcomes are all a result of your own actions within the game world. Procedural generation of a game’s plot is something that, up until recently, has never been seen outside of Façade. And while it might not always work, in the same way text adventures can’t cover every option you choose to hammer in, the papers and publications that came out of Andrew Stern and Michael Matheas’ attempt to create a “highly procedural, A.I.-based” interactive experience are genuinely innovative, laying a foundation for future work in narration, story structuring and dramatic pacing – all possible with almost no human input whatsoever. Their work marks the beginning of the end for linearity in gaming.

Back in London with his mass-murdering A.I. simulation, Robin’s still talking games. “Recently I played Halo 3, and the A.I. is still so stupid. The friendly A.I. is driving around and crashing into you, or into a wall or something,” he says with a grin that most Xbox owners would recognize. “Often it’s easier to let the A.I. cheat than making it really embodied and restricting its senses. But it’s getting harder and harder, with these complex worlds. Now’s the time, I think, where research comes more and more into play.”

His supervisor, Dr. Colton, is also the head of the A.I. Games Network, a cartel of researchers and academics who are interested in testing the limits of artificial intelligence techniques by applying them to some of the biggest problems in gaming today. Their conferences tackle issues in non-player character realism, automatic avatar control for players who might want a break from their MMO duties and even game design itself. But he’s keen to draw the line between his struggle for progress and the developer’s desire for an “x-factor” for their game.

“Remember that we’re A.I. researchers, and we’re not videogames people,” Dr. Colton says. “We have to keep telling ourselves that. If we don’t find problems in videogame design or NPCs or avatar control which really stretch A.I. techniques, then all we’re really doing is acting as A.I. consultants.

“How do we tell whether people are enjoying something more or less? We don’t know the answer to that yet. We don’t know how to dynamically alter a game to make it better. These are problems that will stretch A.I. as much as improve games. If they make millions out of the games, that’s great. And if we get dozens of papers out of it, then it’s win-win.”

Michael Cook has experiments to run.