The “drama manager” Mateas and Stern created sounds very complex. It can read subtle cues from typed-in words and figure out how to use them to move the story along. Can you tell me a bit more about the technology behind this? How, for instance, does a character know you’re angry? Do certain words tip her off?
Façade has several layers of new technology going on. It has language recognition. It doesn’t have true language understanding yet, but if you type in a sentence, the computer will try to get a sense of whether it’s a hostile statement or a friendly statement and which character it’s directed at. Then it tries to use that statement. The way it does that is through the drama manager, a kind of artificial intelligence that stands back and looks at what’s going on at any given moment. Then it says, “All right. Given what’s happened so far, how can I shape this into a drama? Now’s the time to ratchet up the tension, now’s the time for the revelation, now’s the time for the conclusion.”
Have Mateas and Stern programmed in every possible situation they can imagine? Does the drama manager just say, “Okay, he’s getting angry, and I know where to go from here”?
That turns out to be very difficult; there’s no way you can script every possible scenario in advance. You’d have to write an enormous game with terabytes of data to encompass everything a player might say or everything a character might do.
This gets to the third level of artificial intelligence they’ve got going on. They’ve preprogrammed some basic dramatic possibilities. But these possibilities are like beads, and the drama manager sequences the beads. It works with a language called ABL, which stands for “A Behavior Language.” This gives the drama manager a way to control the character without knowing everything in advance. It will let the character say something while also walking and mixing or holding a drink at the same time.
So ABL and the drama manager, working together, can animate the characters on the fly to react to what you’ve said in real time, and then sequence the dramatic possibilities—which have been supplied in the game—to create a compelling response.
Mateas and Stern chose to model their game after dramatic works like Who’s Afraid of Virginia Woolf? and Scenes from a Marriage. What is it about these stories that lends itself to the technology they had in mind?
First of all, they both liked all of those works of art. Second, they needed a contained dramatic environment. There’s no way you could start a project like this with multiple scenes and many characters. An intense marital argument is a very good, natural way to create an intense drama in a short space of time with only two characters in one room. For technical reasons, that was a very good place to begin.
There’s also the fact that you’re walking into a drama that’s already taking place without you, so your role is secondary. Would it be harder to create a game where the player was one of the two people having the marital argument?
I’m sure it would be. That’s a very astute observation. The game comes pre-supplied with dramatic tension. It’s not like you’re walking into any old room where there are two people there and they say, “Hi, how are you?” and you say, “Hi, how are you?” and you just have to hope something interesting happens. The moment you walk in the door—in fact, before you walk in the door, from outside—you can already hear them arguing. The whole world of Façade is pregnant with dramatic possibility. And the player’s role is to catalyze that.
One of the main criticisms of Façade is that it’s intriguing, it’s sometimes touching, but it’s not really fun. And that’s true. But people have to remember that this is a research project. If somebody figures out how to teach a computer to create a dramatic world, create realistic characters, and allow people to interact with them, the limits of using that in ways that are fun are mind-boggling.
Stern feels that video games should span a wider range of genres—that there should be not only action games but also drama games and comedy games. Do Stern and Mateas plan to create a “comedy manager” in the near future?
Their next game, The Party, is meant to be a comic melodrama. The goal is to set up complicated, farcical situations in which improbable things can happen. It doesn’t involve a comedy manager, but it does involve a lunatic, Desperate Housewives kind of story.
But humor is much subtler than drama. You can speak a language fluently and still not fully understand the nuances of its jokes. It’s hard to imagine a computer spinning out comedy in any effective way.
That could be true. A great deal of the interest in this project lies in exactly that. How difficult will it be to teach a computer to write a comedy or a mystery or a tragedy or a love story? If Mateas and Stern succeed, we’ll learn a lot about people and about stories. Even if they fail, we’ll learn something.
Could this technology be applied in other ways? For instance, could an online dating service use it to screen a user’s tastes and personality? Could a therapist use it to teach clients how to interact with different kinds of people?
If I recall, Mateas told me that before he went back to grad school, what got him involved in games again was working on training games for corporations—games that simulated environments employees would find themselves in, dealing with each other in a company. So there are obvious applications for games that can create and animate powerful characters.