The Wrong Tree

[An augmented version of a comment posted on Boing Boing concerning the public release of Chris Crawford’s ‘Storytron’ interactive storytelling engine. I should write something longer and more thoughtful about this endeavour, because I’ve never believed that there was anything to be gained from the approach, and yet Crawford’s passion is real. I presented a paper years ago at an academic conference at which he was one of the star speakers, and he’s the mad prophet of the gaming/interactive fiction world — Howard Beale without the rage.]

I also tend to think that Crawford is barking up the wrong tree. Worse, I have a nagging feeling that there isn’t a tree there at all. Notwithstanding his talent for grandiose self-promotion — not that that’s unusual in AI circles — his insistence that what he’s doing has anything much to do with stories is making a rod for his own back. What he’s doing is (still) interactive game design. Nothing wrong with that, obviously, except when it pretends to be something it isn’t.

This sort of story-world as pinball machine — build all of this machinery, give the user some small element of input, analogous to flipper control, and then expect what comes out of the other end to be a story — is a problem in all sorts of ways. Firstly, building the worlds is extraordinarily difficult. It’s basically the AI world-knowledge problem. That means almost no-one can/will take the trouble, as I think Crawford has discovered. Secondly, a sequence of events — even a perfectly coherent sequence of events — is not at all the same as a story. Stories need an authorial hand, not just the laws of physics.

I’m also far from convinced that ‘interactive fiction’ is something people even want. Maybe I’m lacking imagination, but the most profound value in fiction is that we don’t interact with it. It’s a time when we get to enjoy the skill of the author, into whose hands we willingly place ourselves.

9 Comments

  • A couple of notes:
    Building storyworld using Storytron’s tool, SWAT, does not involve true AI. Storyworld scripting is based on mathematical tools, not AI techniques. Those same scripts allow the storybuilder to lend an authorial hand to the player, allowing storytelling to take place.

  • Louis:

    Without getting into that sort of philosophical discussion, I’m not sure there’s a lot to be gained from trying to divide stuff into ‘true AI’ and stuff that isn’t ‘true’. From what I’ve read, there are aspects of what you’re doing that ought to feel happy calling themselves AI, whatever the underlying methods used.

    The presence – or absence – of the ‘authorial hand’ here is definitely one of the key things about the project. Whether the authorial hand comes from intrinsic qualities of SWAT, or the constructs built by human authors using SWAT, the crucial thing is the experience of the human processing the ‘story’. And that’s the point where I fail to grok the whole ‘interactive fiction’ concept. I can see the appeal of a traditional, human-authored story. It’s a roller-coaster, which we can’t control. I can also see the appeal of games which allow some sandbox-y control within a human-authored, overarching narrative – the Grand Theft Auto games being the best examples.

    But fiction which is genuinely, truly interactive – in which the very structure is determined on the fly, based on user-input – seems to me to be quite incompatible with the authorial hand. The stronger the authorial hand, the less genuine interaction there must be. And the stronger and more real the interaction, the weaker the authorial hand. I suppose the goal is a sweet-spot combination of both, but I’m not convinced that sweet spot actually exists.

    A crucial question to ask here is what, for you, makes a story a story. If it’s just the cashing out of goals and plans, with the whole affair steered in some way by user input, then it’s possible to see the product of Storytron as stories. I tend to think it takes a bit more than that – and absolutely requires the strong authorial hand, by necessity pushing the scope for interactivity back.

  • Paul,
    My clumsy mention of ‘true AI’ was meant to address the AI world knowledge problem that you cited as an impediment to building storyworlds. My point is: storyworlds do not need to incorporate much real world knowledge, so the world knowledge problem does not apply. However, they must incorporate knowledge of the fictional world, and whatever storytelling rules that apply.
    I want to be clear that Storytron’s focus is not ‘Interactive Fiction’ but ‘Interactive Storytelling’. The experience of interacting with characters in a world of interpersonal drama is the goal.
    I have not built any storyworlds myself, so I am not really qualified to debate the mix between interaction and the authorial hand. However, it seems to me that the amount of interaction allowed should match the storybuilder’s ability to apply the appropriate rules of storytelling to those interactions.
    Also, to directly answer your last question, following the unwritten rules of storytelling to achieve drama is what turns events into stories.

  • Louis:

    My clumsy mention of ‘true AI’ was meant to address the AI world knowledge problem that you cited as an impediment to building storyworlds. My point is: storyworlds do not need to incorporate much real world knowledge, so the world knowledge problem does not apply. However, they must incorporate knowledge of the fictional world, and whatever storytelling rules that apply.

    I’m not sure it’s the case that the world knowledge problem does only apply to the ‘real’ world. If you want to create a rich user-experience with characters behaving in complex and believable ways, you hit the world knowledge problem, whether what you’re doing is intended to represent the ‘real’ world or not. The only way you really get away from the world knowledge problem is to create toy ‘worlds’. And that’s fine, as a place to start, but if you have genuine ambition for Storytron, or CYC, or any other project which seeks to capture the things in a world, and how they behave, sooner or later things get very hard.

    I want to be clear that Storytron’s focus is not ‘Interactive Fiction’ but ‘Interactive Storytelling’.

    I’m not clear on the distinction between those, as you define them.

    Also, to directly answer your last question, following the unwritten rules of storytelling to achieve drama is what turns events into stories.

    Forgive me for being facetious, but a formal system can’t follow rules if they’re unwritten. But in general I agree that there are some such rules – heuristics might be a better word. My problem, in a nutshell, is seeing how those rules can be followed well enough to turn sequences of events into structures people genuinely perceive as *stories*, without narrowing the interactivity to almost nothing. It could be that I need to lower my standards for what I think of as a story. 🙂

  • Paul,
    Storytron’s software really is designed for creating and running “toy worlds”. See item #4 at this link: http://www.watercoolergames.org/archives/000916.shtml
    Concerning the distinction between ‘Interactive Fiction’ and ‘Interactive Storytelling’, I’ll quote the Wikipedia article on ‘Interactive Storytelling’:
    “Interactive storytelling and interactive fiction are distinct in that interactive storytelling focuses on drama and dynamic circumstances, where interactive fiction games, traditionally (but not necessarily) focus on puzzle-solving and navigating through pre-conceived circumstances. They are similar, however, in that well-written forms of both are nonlinear.”
    I’ll respond to the issue of rules in another post soon.

  • Paul,
    Interactivity is possible if the rules programmed in the storyworld are distilled from an abstract story ideas. A trivial example: Suppose I want my storyworld to show that bad things happen to those who do wrong. I can implement a rule that increases a “bad event” factor as the character uses the “wrong” verbs. This would allow the player to make moral choices, and experience the results of her choices and those of other characters. A resulting story may not be as profound as a written story, but since it would follow a human principle, it would not be a random chain of events.
    Note: The case for abstraction is presented in the book Chris Crawford on Interactive Storytelling.

  • Louis:

    [Rearranging the order slightly:]

    Concerning the distinction between ‘Interactive Fiction’ and ‘Interactive Storytelling’, I’ll quote the Wikipedia article on ‘Interactive Storytelling’:

    Got it. I’m not sure those phrases are the best way to make that distinction, but it is a useful distinction, and worth making.

    Storytron’s software really is designed for creating and running “toy worlds”.

    Yes. I get that too. I’ve expressed some concerns before about whether this sort of interactive storytelling is something people really want/need, but I also have concerns about the feasibility of the concept on its own terms, and scalability is one of the big problems.

    Chris referred to his latest interactive work as the “Pong of storyplay”, and it’s clear what he means by that, but there are some problems with the analogy. While clearly being a portent of much greater things to come, Pong was a sensation on its own terms. Its success wasn’t dependent upon a scaling up of the technology. Chris’s [by which I lazily mean his whole team’s] work wants to be seen as the first step towards something great – a proof of concept, rather than something which is necessarily playable as it is. That’s fine, but it therefore at some point needs to address the problems of scalability, and I think they’re significant. I can think of at least three:

    1. The knowledge problem. As a storyworld gets bigger, the explicit knowledge required to make things work in a way that’s both coherent and, crucially, which feels like a real story, grows exponentially. The problem is finessed away if one accepts that one is dealing only with toy worlds, but toy worlds aren’t that interesting.
    2. The author problem. Following from the above, as a storyworld gets bigger, managing the knowledge involved becomes exponentially more difficult, and the human effort required balloons. Finding people willing and able to make this effort will get harder and harder. Perhaps I’m wrong about this, but I get the feeling that gathering authors willing and able to work with Storytron as it stands now isn’t all that easy.
    3. The story problem. As a storyworld gets bigger, the information wrangling necessary to keep things feeling like a story also gets much harder. Far more options can be – and therefore perhaps *should* be – available to the user/player, creating an extremely profuse story tree. Very few branches of such a tree would look like stories, simply because very few information structures in any context look like stories. Keeping things storywise isn’t impossible, but – again – it’s a significant challenge.

    Interactivity is possible if the rules programmed in the storyworld are distilled from an abstract story ideas.

    Right, and your example is clear. My concern is that I see the storyness of a story as a structure/shape which exists at a much larger scale, and I’m not sure to what extent Storytron can cause that structure to fall out of the back-and-forth. Story is more than a sequence of coherent actions and reactions. It’s about rising action, and character development, and plot complexity, and flashbacks, and flashforwards, and foreshadowing, and all of the other informal tricks you’d see in a typical screenwriting book.

    If the user/player doesn’t intuitively feel that they’re genuinely getting into a compelling storyworld – if the system they’re interacting with feels like a toy – then I’m not sure that, for example, ‘moral’ choices really come into it. What people do with toys is to play with them, see what they’re capable of. “Let’s see what happens if I do *this,” or “Let’s see what happens when I do *that*.” My own reaction to BOP21 was to make a few choices, and then – slightly bored with what was happening – to try nuking everyone to see what happened. Needless to say, it didn’t result in anything like a story.

    To be clear, I don’t want to seem to be down on your efforts – although I’m sure that’s how it comes across. I know from experience how much ingenuity and work is involved. I wish you the best of luck and success with it. Perhaps I’m too fixated on my own sense of what a story is, and what I’d want from an interactive experience.

  • Louis linked me to this discussion; it’s interesting. A few comments:
    1. On the “self-promotion” thing, that press release was written by another member of our team. We have an ongoing battle in which I attempt to get them to scale back on the hype and they counter with the point that I’m the world’s worst salesman and that good business requires that we *promote* ourselves. Inasmuch as I have taken other people’s money to advance this project, I grudgingly acquiesce to the imperative. But hell if I’m going to write that pap myself.
    2. You note that there will always be a difference between conventional stories and the results of interactive storytelling, and you worry that this difference will be too great to overcome. I agree that the difference will be huge, and that the results of interactive storytelling will never be as rich as traditional stories. I maintain that this doesn’t matter. You noted that Pong was a big success in its own right. What, precisely, do the players do in Pong? They bounce a little ball back and forth. This is entertainment?!?!?! Well, yes, it is, because the interactivity lends enormous appeal to what is otherwise an insipid activity. We don’t need to compete with conventional storytelling because we have interactivity on our side, and interactivity is a huge bonus.
    3. You’re right to worry about scalability problems — but I see these problems as addressable by ever-improving tools. Already we have made substantial improvements in SWAT in response to difficulties that I faced in building BoP2K. Pong was written in assembly language, and it would have been impossible to scale Pong up to modern first person shooters with the same tools used to create Pong. BoP2K was the first serious project undertaken with SWAT. We learned a lot and we’ll continue to improve SWAT as we learn more. While the scalability problem will always be with us, I think we can make its slope shallower.
    4. You are right on the nose with the problem of recruiting authors. Lots of people have been intimidated by the complexity of the technology. We believe that this complexity is, to some extent, intrinsic to storytelling, because storytelling is an immensely complicated intellectual task. We believe that our bacon will be saved by the impetus people get when they see what can be done with the technology. Every new technology has problems with early adapters trying to get a grip on how to use it. When the Mac came out, there were damn few people who “got” the technology. I was one of the first group of early adopters to figure it out, and that was definitely a factor in the success of the original BoP.
    We’re working on it.

  • Chris:

    You note that there will always be a difference between conventional stories and the results of interactive storytelling, and you worry that this difference will be too great to overcome. I agree that the difference will be huge, and that the results of interactive storytelling will never be as rich as traditional stories. I maintain that this doesn’t matter. You noted that Pong was a big success in its own right. What, precisely, do the players do in Pong? They bounce a little ball back and forth. This is entertainment?!?!?! Well, yes, it is, because the interactivity lends enormous appeal to what is otherwise an insipid activity. We don’t need to compete with conventional storytelling because we have interactivity on our side, and interactivity is a huge bonus.

    Maybe, but I’m sceptical. With respect to interactivity, I can see at least two important questions to ask.

    The first is: in order to achieve interactivity, what’s *lost*, and does/can the interactivity provide a net gain. It’s a complex calculus, and I don’t know the answer, but I don’t think it’s *necessarily* true in the case of stories that there’s a net gain. As I said in my original Boing Boing comment, I think a profound strength of stories is that they’re *not* interactive. Reading fiction, or watching a film, is a time when we can give ourselves over to someone else’s creativity, and enjoy the ride. It’s a particular pleasure to pick up a novel by an author we trust to tell good stories, and know that the ride will be a good one; it’s perhaps an even greater pleasure to discover on reading an author new to us that they really know what they’re doing, and that wonderful things lie ahead. Processing a story isn’t a passive exercise, exactly, because our brains are continuously processing the information we have, asking questions, revising assumptions, projecting ideas forwards, and such. But the strength of the process is to a large extent that the story is a structure we trust to be a good one, and wish to be *told*, rather than to guide ourselves.

    Pong provided interactivity with no loss. We played it without seeking to compare it to another medium that worked without interactivity. The interactivity was, for most people, new, and exciting, and didn’t feel like a compromise. At least at the moment, Storytron feels like a huge compromise. At one end of the spectrum there’s a novel, which is entirely authored, and not at all interactive (in the sense of ‘interactive’ we’re using here); at the other end of the spectrum there’s Pong, which is entirely interactive, and not at all authored (in the sense of ‘authored’ that’s relevant here). Both work very nicely. The middle of that spectrum is a very tricky place to inhabit. If you give up the concept of an authored artefact – which strikes me as a big loss – in order to buy some space to be interactive, then the interactivity has to be truly compelling, in order to make up for what’s lost.

    The second question is whether the interactivity works on its own terms. ‘Interactivity’ can mean all sorts of different processes, and they’re not equal in power and worth – and certainly also work well to different degrees with different people. It’s probably the case that Pong had the *novelty* of its interactivity on its side, but there were other things too. It could be picked up and learned incredibly quickly, played for a few minutes or a few hours, and by pretty much anyone. Feedback from the interactive process was immediate, and simple, yet also challenging. There was a single, clear goal to the exercise, which pushed people to explore. It was also a social activity, much the way that Wii games are these days. All of these things meant that the interactivity worked, from day one, for a huge population.

    Storytron’s interactivity has few of these advantages: there’s a learning curve which excludes many potential users; feedback is complex – even with a story-world as simple as that of BOP21; it’s not a social activity, so there isn’t the inherent incentive of competing against a friend or loved-one; and there’s no clear goal to guide our exploration – other than whatever we come up with ourselves. There’s an argument that the lack of explicit goals within Storytron worlds (or the presence of many goals) is the *point* – and that’s fair enough – but, as with the loss of the strong authorial hand, the loss of goal-directed ‘play’ is not a small one, and its absence leaves a big hole for the interactivity to attempt to fill. Pure exploration can be fun, but the narrative Macguffin can be a powerful motivation for the exploration. Exploration of the worlds of the GTA games – Liberty City, San Andreas, etc. – is great fun, but is given depth and longevity by the motivation and navigation provided by a sequence of explicit missions. In the same manner, the ostensibly thin veneer of championships in the Gran Turismo games serves to brilliantly guide our path through the wide range of cars, and tracks, and constitutes a quasi-narrative of its own.

    You’re right to worry about scalability problems — but I see these problems as addressable by ever-improving tools. Already we have made substantial improvements in SWAT in response to difficulties that I faced in building BoP2K. Pong was written in assembly language, and it would have been impossible to scale Pong up to modern first person shooters with the same tools used to create Pong. BoP2K was the first serious project undertaken with SWAT. We learned a lot and we’ll continue to improve SWAT as we learn more. While the scalability problem will always be with us, I think we can make its slope shallower.

    I can see that more advanced tools would help, yes, at least with respect to the authoring process. It’s not clear how much they’ll help with the fact that the user/player/whatever will be faced with navigating an ever-more complex world, with an increasingly bushy story-tree. It seems to me that the authorial hand would by necessity need to become stronger to hack out even a minimally-manageable path through the thicket. The more complex the world, the more authorial guidance it would require.

    But, again, I think the analogy with Pong can be misleading. What you’re building are essentially knowledge-based systems, and, even with more elaborate authoring tools, scaling up such systems is a problem of exponentially-increasing complexity. To reduce the issue to an adage, there’s no Moore’s Law for knowledge-based systems. As complex as they are these days, 1st-person shooters aren’t nearly as onerous as the sorts of knowledge-based systems I think you’re aiming for. Their complexity mostly resides in scenery design, animation, etc.; the story and world design is still relatively simple, linear, and authored.

    Video games had the advantage of being able to evolve from Pong onwards in small increments, which at each stage could produce something commercially viable. The developers could therefore (potentially) invest in the tools they needed, with some confidence that the investment would pay for itself. I’m not sure you have that luxury. Without the chicken of major investment in development tools and human world/story design, it’s hard to see where the egg of wide use and potential financial return might come from. And without the egg of wide use and financial return, it’s hard to see where the investment in development and people might come from.

    You are right on the nose with the problem of recruiting authors. Lots of people have been intimidated by the complexity of the technology. We believe that this complexity is, to some extent, intrinsic to storytelling, because storytelling is an immensely complicated intellectual task. We believe that our bacon will be saved by the impetus people get when they see what can be done with the technology.

    I hope so. I do think that, even more than getting used to the technology, it’s the complexity of *stories* themselves that’s the biggest issue here. It’s yet another problem with the Pong analogy. Back then – and even today, to some degree – it was quite possible for a single, bright programmer to create a successful video game in a reasonable time. The amount of heavy-lifting involved in creating a rich, interesting story-world might easily pass beyond the scope of a single programmer, making the ratcheting forward of the whole process even more difficult.

    Okay, enough. I really appeciate you taking the time to comment, Chris, especially as I’m not exactly being very positive about the whole thing. It’s less about hoping that you don’t succeed, than fearing that you won’t, and not quite *getting* the whole approach. I’d love to be proved wrong.

Leave a Reply to Louis Dargin Cancel reply

Your email address will not be published. Required fields are marked *