The latest issue of Dr. Dobb’s Journal arrived today, and contained an interview by Michael Swaine with game design legend Chris Crawford on the subject of his new venture. You can find the piece online at DDJ.com. If games and game design interest you, I advise reading the interview before continuing. I have admired Mr. Crawford since the mid-1980s, around the time of Balance of Power. This is a guy who wrote his first commercial computer game for Atari in 1979. Balance of Power was a certified 250k copy smash when it was released in 1984. Somewhere along the way he veered off from the game business and went in a different direction. His new gig is a company he cofounded called StoryTron. After reading the interview I went to the site and read everything there, including some of the message board content. I think I now better understand where and why he parted ways with the industry he helped to create. I’m not sure that, in the end, it will be a journey with a successful ending, or at least it might not be the kind of success he envisions. But I understand the attraction of what he is trying to do.
Mr. Crawford wants to build worlds. You can see that in the DDJ interview, and in an interview he gave to Gamasutra. You can see it even more clearly in this abstract from his book Chris Crawford on Game Design. In it he discusses the motivation and design behind Balance of Power. He was perhaps the first designer to catch the Creator bug, and begin to envision his simulations as believable alternates for reality. BoP began with a highly-detailed model of the world of international diplomacy, and although he had to pare it down to make it work, it was still one of the most complex simulations of its time. The urge is a powerful one. Virtually every designer of mechanisms dreams at one point or another of the concept of automata. Game designers tend to take the idea further than anyone else, because they can. An industrial engineer has to be satsified with creating the best individual robot. A game designer can create an entire virtual world and render it for the viewer.
If you’re inclined toward this kind of thing it is an incredibly enticing vision. And if, like most programmers, software designers, and logical thinkers, you are both a pattern seeker and a rule writer, a miner of fundamental formulae, then it’s very easy to be siezed by the notion that if you… can… just… get… the rules right you can simulate life! What a virtual world that would be! It would be a world populated by amazingly intelligent and interactive characters all going about their daily lives, and in which a million individual stories bloom and wither moment by moment, any one of which your character can dive into or not, as you desire. Yep, that is one hell of an idea for a simulation designer. It’s the Holy Grail of simulations.
Unfortunately, there’s always the chance it would be about as exciting as the Motor Vehicle department for the average consumer. The market for Mr. Crawford’s worlds was different back in the 1980s. Just about anyone who qualified to buy and play with one was an enthusiast, and might be easily taken by the same fascination that consumes the designer. And the designer in this case is very much consumed. There is a messianic quality about Mr. Crawford’s enthusiasm for what he has always called interactive storytelling, but which could now be more fittingly termed social simulation, that often takes him past where a less motivated person would rein in. For example, he refers to the StoryTron “actors” as “intelligent,” “emotional,” and “perceptive.” I’m going to go out on a limb here and say they are none of those things. We don’t know how to program those things. Nobody does.
What StoryTron has done is to spend a lot of time thinking about a model for social interaction, which they believe to be the basis of all “stories.” They have developed a type system (Actors, Roles, Verbs, Principles, Plans, Stages), operators to interact with the types, a scripting language called Deikto to handle input and output, and an engine which they call StoryTronics to run the whole thing. The basic model is that Actors execute Verbs in response to other Verbs, or in response to their own Plans. They choose possible Verbs based on their Roles, and the game matches interacting Actors and the Verbs they have executed according to Principles. I’m impressed by the neat elegance of the model and framework. It’s obviously been very well thought out. In the end, though, it is primarily a framework for walking a behavior tree based on heuristics, such as the evaluation of a given Actor’s response to a Verb based on the Actor’s role, and variables capturing attitude, standing, etc.
If the complexity of the tree, i.e. the number of possible stimuli and responses, is great enough, then you could certainly expect to see interesting emergent behaviors in such a system. If the global concepts of Principles and Roles actually give you enough high level control to allow the world to be defined and nudged into the appropriate behavior in a reasonable amount of time, then you might even be able to fashion interesting worlds within such a framework. Crawford is clearly aiming in this direction, given that all the world-building and programming takes place in a drag-n-drop paradigm where you connect Verbs, Actors, Roles, etc. I find the whole thing completely fascinating, and will definitely follow its development, and experiment with it. However, I find the concept itself more enticing than Crawford’s vision of it, and I suspect that vision might ultimately cause him to fail to get the most from the idea.
Ninety percent of what they are doing is creating a heuristic model of social interaction between characters. But Crawford doesn’t see it that way. For him the model is just a necessary component of an environment in which you can tell interactive stories. But of course he doesn’t mean “tell” at all. He expects designers to fashion worlds in which interesting stories can emerge. If you are scripting events and moving the reader (player?) along a plotline, you would seem to be off the reservation philosophically. The reader is supposed to create the stories by interacting with the world. This is far too tall an order for any current technology to deliver on. It is, in fact, not philosophically distinct from the idea that a billion monkeys typing would randomly produce Shakespeare. It’s a beautiful, wonderful notion that is just not happening any time soon. It requires algorithms that are not only intelligent, but are capable of being inspired to beauty. The best you can achieve is to embed many stories in the world, have the character stumble on them, and have them progress somewhat randomly to some set of possible conclusions. That much simulated disorder is manageable.
Not that there is no market for interesting simulations that run themselves. Look at Wil Wright’s success with The Sims. But that is exactly the point at which the real problem occurs. According to Crawford The Sims is a toy (and to be fair he says that Wright originated this view), and not interactive storytelling at all. StoryTron does not see itself as creating an interaction engine. That is not big enough. They are talking about creating an entire platform for these emergent stories. So I did not see any mention of a serious idea for separating the model from presentation. If I want to use StoryTronics to drive a population of NPCs in an RPG it doesn’t seem I will be able to do that. Similarly I did not see any idea for hooks to custom evaluators. If I want to code some more specialized AI for a given character, say a General who needs to evaluate global strategic options, it doesn’t seem I can do that either. That’s because this isn’t an engine. It’s a world in which writers, not programmers, can… well, not do what they usually do, which is to tell stories. Rather “story builders” will fashion worlds with rules that cause stories to emerge from the model.
Crawford specifically disavows the genre that he calls “branching narrative,” but I think this is as much a stretch as the idea that his Actors are intelligent or emotional. What he is creating is an engine that can handle a branching narrative with orders of magnitude more branching, and much subtler means of evaluating decision points. Such a fine-grained system is unlikely to produce anything that looks like or can be experienced as a story, unless the high level authoring tools are truly breakthrough technologies that allow the author to layer the story on top of the world. I don’t think that is where they are going, simply because of Crawford’s disdain for plot. He sees plot as the author telling the reader where to go and what to do. That’s absolutely true, but then, that is the whole nature of a story. Given a blank canvas, a brush, and paints the average viewer of fine art is unlikely to be able to create a Van Gogh, or even something he or she would appreciate. The average reader does not want to write their own story. The average gamer does not want a simulation constrained into a “story telling” interface. So where does StoryTronics fit in? I think it is flexible enough to be used for some really interesting things, and it may well find a profitable niche. But it will not be telling stories. Artists tell stories. Until we can make a program that is capable of being inspired as opposed to evaluating simple rules, computers do not.