Chris Crawford on Interactive Storytelling

Recently I picked up and read Chris Crawford on Interactive Storytelling with an eye to what it might have to teach about IF.

Now, before I say any more than that, I should admit to certain biases. First of all, while I haven’t ever played with an Erasmatron game, I am familiar with descriptions of Laura Mixon’s “Shattertown Sky,” and they did not leave me with the impression that the Erasmatron was the next best thing in interactive storytelling. (See for instance Neil deMause’s article “All That Erasmatazz”, though that was some time ago.)

I should also point out that in reading it this way, I was going against the grain of the book itself. Chris Crawford doesn’t see his advice as something that could enliven the craft of interactive fiction; on the contrary, he considers interactive fiction a dead end as far as storytelling is concerned:

Interactive fiction is certainly interactive, and it’s fictional in the sense of being made up, but it’s certainly not storytelling. Some practitioners of the field write eloquently of the glorious narrative possibilities, but the actual creations remain elaborate puzzles. (337)

And then he states flatly that interactive fiction will never lead to interactive story-telling, citing Floyd’s death scene as an example of why not: “Players were overwhelmed with the moment’s emotional power. There was nothing players could do to avert it.” And therefore, says Crawford, it was not interactive story-telling; the implication being that therefore no legitimately interactive-yet-moving moments could ever conceivably occur in the medium. Generalizations of this type abound: Crawford has strong opinions about what type of thing interactive storytelling is, how it might be achieved, and why most of the current efforts are sad failures. They are sometimes aggravatingly unsubstantiated.

Nonetheless, as I read the book, I was struck by ways in which interactive fiction does offer solutions to problems that Crawford outlines, and ways in which Crawford’s analysis could be applied to IF.

Interactivity and Choice, in Theory

The book comes in five parts: “From Story to Interactive Storytelling”, “Styles of Thinking”, “Strategies for Interactive Storytelling”, “Core Technologies for Interactive Storytelling”, and “Applications”.

Much of Crawford’s argument about what interactive storytelling is or should be occurs in that first part. His analysis of Choice is worth pointing out:

…the absolute number of choices isn’t important; it’s the number of choices offered, compared to the number of possibilities the user can imagine. If the user has reached the climax of the story and must choose between leaving his girlfriend for the war and shirking his duty, having only two choices doesn’t detract from the power of the interaction; it’s difficult to imagine any other reasonable possibilities. (41)

One of the challenges facing would-be authors of branching IF is determining which options to offer, and how many. While Crawford doesn’t exactly offer a simple solution for this problem, he does draw attention to its nature, and remind authors to present only the dilemmas that are interesting in the context of the tale.

On the other hand, Crawford also concludes that he must rule out certain kinds of option from interactive storytelling — and this I found interesting, because in fact I believe these are the places where IF shows its strength. In a section labeled “What Can’t Be Part of Interactive Storytelling” (61), Crawford argues that the computer cannot offer the player what he calls “the Third Option”: “the protagonist’s stroke of genius that gives him a third option for escaping the sharp horns of a dilemma”. The reason this is impossible (Crawford explains) is that if you presented the player with a list of options, the correct one would be obvious, and the decision would become uninteresting. This strikes me as an excellent argument for using a text parser: IF can present a hidden third option, and let the player find it when he’s clever enough or determined enough to do so. (Consider “Slouching Towards Bedlam”.)

In the same section, Crawford explains that interactive storytelling cannot include the “Creative Option” — things that the player can do that the author had not anticipated. This is a bit stickier, and IF is not long on creative options. The trick lies, I think, in providing a simulation for whatever aspect of the world the player uses to express his choices. This is impossible if the player is expressing choice via an option list. It is possible with a world model and parser, though, to give the player several ways to achieve the same outcome, and even (with a sufficient simulation under the surface) for that list of ways to include some unexpected by the author. It’s true that this does not give the player the freedom to invent entirely new outcomes, but it does expand the possibilities somewhat.

Part II, “Styles of Thinking”, is a bit of a mixed bag. It covers the scientist/artist divide, criticizing both sides freely; while there’s some merit to Crawford’s criticisms, this mostly comes off as a rant. More interesting are the chapters on “Abstraction” and “Verb Thinking”, in which Crawford argues that interactive storytellers need to think generally and about activities (rather than objects). Even if these chapters had nothing else going for them, they would be worthwhile for the emphasis on one thing:

Crawford’s First Rule of Software Design: Ask “What does the user DO?” (99)

This is a fundamental question — for IF as well as for any other software — and Crawford demonstrates what can happen to an interactive story built by someone who hasn’t been asking this question.

Implementation Detail

In Part III, Crawford begins to get down to specifics. “Simple Strategies That Don’t Work” addresses an assortment of plot shapes authors have used to build interactive stories, especially linear plots, plots that branch, and plots that branch and then meet up again. He offers a good explanation of the problems that attend these formats: that branching plots rapidly require more content than anyone wants to write, for instance. I found this chapter to be a solid introduction to something I already knew. It’s most likely to inform readers who have not yet given much thought to branching narrative structures.

The other three chapters are a little more mystifying, because Crawford outlines three approaches (“Environmental Strategies”, “Data-Driven Strategies”, and “Language-Based Strategies”) to creating stories, but he mostly presents background information without explaining what we might reasonably hope to do with these strategies. “Data-Driven Strategies” is largely an overview of the work of Propp et al.; “Language-Based Strategies”, a discussion of constructed languages for communicating with the computer. “Environmental Strategies” is probably the most interesting from the point of view of the IF author, since it touches on games designed around a specific setting; on the other hand, it’s also fairly brief and dismissive.

Part IV tackles the question of how one might build personality models and characters who communicate with each other. This is in some ways the most grounded section of the book: Crawford frequently refers to specific learning experiences in his own research. The first chapter in this section, “Personality”, might be especially useful to someone who is interested in numerically modeling a complex interactive character, with its analysis of how to choose which characteristics to track and how to define conditions (should I add the NPC’s strength to his determination to find out whether he climbs the wall, or should I multiply?).

I also felt, unfortunately, as though some fundamental disconnect had occurred between Parts II and IV. IV contains ever-more-detailed techniques, but the further Crawford got into the simulationist aspects, the less I could see how the result would be an actual story. Given what I’ve heard about the Erasmatron, I’m not sure Crawford has solved this problem, and it strikes me as an important one. In my own experience (though it is, obviously, rather briefer than his), no amount of stringing together automated behavior produces narrative moments as powerful as those designed by an author. Does that mean interactive storytelling is impossible? I don’t think so, but I do suspect that, at least for now, the core of the story needs to be written, not computed.

For instance: “Drama Managers” offers some suggestions for stringing together plot events flexibly using a computer story-teller: this is something that has seen discussion in the IF community from time to time, though no one has produced much by way of demonstration. (I suspect the problem is harder than it looks.) Crawford’s analysis here mostly covers some ways to shepherd a wandering player, by trying to change his goals, adding environmental blocks to send him the right direction, or providing NPCs with strong urges of their own. (I confess, I found it difficult to take seriously his suggestion that a hell-bent Frodo should be distracted by making Sam badly need to urinate, but I suppose this was meant to stand in for a range of possible alternatives.) This all dodges the much more difficult question of teaching the computer how to construct a good story, and when it would be appropriate to invent these diversions.

Subsequent chapters in Part IV become even more detailed and even more specific to the design of the Erasmatron. Here and there one can extract some useful points about NPC AI — for instance, that NPCs capable of social goal-planning should bear in mind the probable feelings of the other NPCs. All the same, there’s a good deal to wade through that may not be useful for any other model world or system than Crawford’s own.

IF’s Relations

Part V, “Applications”, covers some past and current attempts at interactive storytelling (as well as interactive fiction), and makes an interesting introduction to IF’s cousins and more distant relatives, from Crawford’s own Erasmatron to hypertext and RPGs. Some of the projects outlined are already dead in the water, such as CMU’s Oz project (which, however, provided the underlying concept for Nate Cull’s Reactive Agent Planner, as well as a number of spinoff projects less directly related to interactive fiction). Others, like Andrew Stern and Michael Mateas’ Façade, are in more general release. It’s not worth buying the book just to find out about Façade, though — there’s far more information on the authors’ website. Indeed, Crawford doesn’t go into much detail about any one of these storytelling mechanisms other than his own Erasmatron; this may not be very surprising, given that there’s a lot of territory to cover, but the “Applications” section really does function as introduction; readers who want to explore a specific medium should look elsewhere for depth.

…and So?

Taken in whole, Chris Crawford on Interactive Storytelling presents a vision of an interactive experience that consists of making emotionally-charged choices from a fixed selection (do a? or do b?). The choices are important, and he has some useful things to say about choice and storytelling. But the simplicity of the interface lets his vision down. At one point Crawford remarks that it is fine to present long chunks of un-interactive narrative to bridge the gap between the interesting choices (55). Personally, I tend to find such chunks a turn-off in interactive fiction, especially if they involve forcing the player character to do an assortment of things I didn’t particularly intend.

This may just mean that I need to reset my expectations as a player, but I think there’s actually a different problem here. To wit: a story may be about choices, but the choices aren’t all there is to a story. It must also provide a reason for the player to care about the outcomes: background exposition, in short. Yes, it’s possible to present the player with wodges of exposition and then simple choices at the junctures, but I think it’s more interesting to let the player participate in the discovery of that background exposition. The decision points, when they come, will have greater power.