Using Interactive Play to Explore How We Think


December 6 2001

Artificial intelligence, the saying goes, is no match for natural stupidity.

Fortunately, this hasn't stopped researchers and programmers from trying to create computerized human-level intelligence.

Without advancements in artificial intelligence, or AI, enemies in action games couldn't dodge or shoot back. Opponent teams in football games would call the same play over and over. Populating games with realistic computer-controlled characters is a critical component of fun. And with increasingly powerful processors, game developers have enough horsepower to craft convincing AI characters that can react instantaneously to the player, negotiate a physical environment, create a plan and even learn a player's habits and adjust the game accordingly.

Game AI programs have become so sophisticated in recent years that a few university researchers have taken an interest in the field, including John E. Laird, a professor of electrical engineering and computer science at the University of Michigan.

Laird recently argued in a paper titled "Human-Level AI's Killer Application: Interactive Computer Games" that computer and video games are perfect laboratories for artificial intelligence research.

Question: What's your research goal?

Answer: I am fascinated with what makes us tick. I also like to build things that surprise me. For me, one of the big unknowns is how we think and how we build systems that think like we do. Games provides an environment for us to explore that issue. Not too many people explore that goal, but it's fun to do. My secret plan is to try to get more people in the AI community to use games as a research tool and, in turn, elevate the level of AI in games. This is an interesting marriage that can lead us to understand what intelligence is all about.

Q: What's the hardest issue in building a good game AI?

A: One of the issues is that there is a tension between realism, the idea of believability, and fun. A simple example is in car racing games. Originally, cars would go around the track in a pre-designed pattern. If you take the best driver in the world as your AI, that's not going to be fun, because the player will never win. So in some driving games, the computer slows down the other drivers if you fall too far behind so you have a chance to catch up. So what you see is a foil for the player that's competitive, but not so hard that they'll never be able to beat it.

Q: Different genres require different AI programs. A shooter game requires enemies. But other games require social intelligence as well. Do you see any games out there that do a good job with this aspect of AI?

A: One thing we haven't seen very much of is social challenges. "The Sims" is unique in that way. The popularity of "The Sims" suggests that there's a demand for this. But that kind of social interaction is very simple. It's a big dollhouse. You tell your Sims what to do, and they go off and do it. It would be nice to have a first-person game where you can interact with your environment, as opposed to directing the characters. But we have a ways to go before we get to that point. We have to have better ways of communicating from human to AI, for example. But we'll eventually get there, and we'll see an explosion of games involving social interaction. AI is going to be a big part of that.

Q: Any dos and don'ts?

A: Don't make your AI too predictable. The whole point of adding AI is adding dynamic challenge to the game. That gets boring very fast if they do the same thing. That doesn't mean they're completely arbitrary either. The real potential is for characters to have their own goals and their own reasons for being in the world.

Don't build an AI to be super-complex if people aren't going to appreciate it. Say the player is in a room and the enemies are outside. You can build an incredibly complex AI that directs how the enemies plan their attack and position themselves outside of the room. But the player sitting inside the room never sees any of that. So you should forget about the planning and just make them pop inside the room. Personally, this rubs me the wrong way in terms of making an AI that is as human as possible, but it lets you build things efficiently.

Q: When should you use cheats and shortcuts?

A: In general, the mantra of games for the programmer is: Cheat as long as you can't get caught. There's no reason you have to play fair. In "Age of Empires," the AI will just give itself extra resources. As long as the player doesn't know it's cheating, it's OK. Do it in such a way that's not obvious.

Q: What did IBM's chess-playing Deep Blue computer contribute to the world of AI?

A: It gives us this notion of planning, of anticipating a human player's next move and seeing how that changes the world. We've been trying to build that idea of anticipation so the system is able to predict player behavior and react.

Q: Can you build an AI engine that can be reused in more than one game?

A: Maybe for action games like "Quake" you can. But most games are so different that it's difficult to come up with generic AI architecture. For different games, you need different AI.


Alex Pham covers the video game industry. She can be reached at

For information about reprinting this article, go to