July 30, 2004
On Improving the Form
Via our neighborhood ludology blogs, here are links to two articles with ideas on how to improve interactive narrative experiences. First, a new essay by Timothy Burke in which he strongly advocates agency in MMOG virtual worlds.
MMOGs can never be virtual worlds until they abandon the character as the primary unit of persistence. To be virtual worlds, they have to make the gameworld itself the major unit of persistence. … This is the dream of many MMOG players: they beg for gameworlds in which their actions matter, in which there are events of consequence. Developers promise to pursue this chimera, but rarely implement anything even approaching the most modest dreams of players.
Second, an older essay (1989) espousing the concept of game-stories, by Ron Gilbert, veteran developer of adventure games (Monkey Island) and its technology (SCUMM), posted on his new blog. In the essay, which holds up quite well 15 years later (perhaps suggesting how little progress has been made in interactive narrative since then), Gilbert discusses his “rules of thumb that will minimize the loss of suspension of disbelief” in game-stories. Particularly interesting to me, in light of our current experiment in real-time interactive drama, is Gilbert’s rule that “Real time is bad drama”:
One of the most important keys to drama is timing. Anyone who has designed a story game knows that the player rarely does anything at the right time or in the right order. If we let the game run on a clock that is independent from the player’s actions, we are going to be guaranteed that few things will happen with dramatic timing. … The key is to use Hollywood time, not real time. Give the player some slack when doing time-based puzzles. Try to watch for intent. If the player is working towards the solution and almost ready to complete it, wait.
I agree with this principle, that a strictly real-time interactive experience can be problematic. It’s usually a good thing for the system to slow down or speed up as needed to adapt to what the player is doing and best support a satisfying experience. However, unlike most puzzles you encounter in adventure games and IF, I much prefer interactive experiences that are always moving forward, that keep you engaged at all times, that never just sit and wait for you to act. Perhaps “real-time” isn’t the perfect word to describe this, since always moving forward can still be adaptive; maybe “continuous-time” is better. Continuous-time interactive drama.
Burke offers a solution for how to offer true agency in virtual worlds: interactive narrative. He labels it the “narrative-nudge” model of gameplay. I couldn’t agree more, except with how he would implement it:
The gameworld is constructed around a massively branching tree of preset “narratives” which are associated with present implementations of gameworld conditions. … Players use the actions of their characters to “nudge” the gameworld towards different branches of the narrative tree. Once a set threshold point is reached, the gameworld will at a random moment “launch” a new narrative branch and the distinctive gameworld conditions associated with it. Once the gameworld has moved along a new branch, it cannot return to past narrative junctures.
Burke later describes the fatal flaw with this implementation approach:
An NN-MMOG would have some very serious design difficulties to overcome. First, the effort of designing and programming what would have to be a very, very extensive and wide-ranging series of narrative branches would be enormous, easily equalling or surpassing the challenge that game content poses for a conventional MMOG. (On the other hand, eliminating the endless managerial nightmares posed by conventionally accumulative character-based persistence like levelling or loot would save labor, as well).
Unfortunately, a branching narrative approach just won’t work for solving this conundrum. I wasn’t as sure about this as I am now, until we spent years banging our heads against it, trying to invent a new approach to it.
The real solution — sorry folks, there’s no getting around it — is to create far more procedurally generative architectures. Systems where there are no explicit branches. Story nodes, of as small granularity as possible, that are dynamically generated from very small hand-made chunks of content / knowledge, and are dynamically sequenced. Put simply, a system that is to some extent actually creative.
In Facade, the system is a creative editor, dynamically sequencing a large collection of small grain-size story pieces (individual sentences of dialog), hand-authored by us; the system is able to evaluate which pieces are better than others to choose from, to keep the drama interesting and responsive. That’s not good enough. Even after years of labor there’s nowhere close to enough content in the system, even for a small dramatic scenario.
Such a system needs to be much more of a creative writer. The system needs to be constructing the sentences themselves, not just sequencing them. This doesn’t mean building AI that can creative narrative from whole cloth; even human writers don’t do that. This does mean building AI that has a lot of knowledge about how characters, stories and emotions work, and the rules and procedures for constructing new ones. That’s what people do when they write; they draw upon their experience and knowledge of stories they’ve experienced, to know how to create new ones.
Generativity is what we need to turn our time and energy towards. Once Michael and I finally get Facade out the door, and we each get a chance to regather our wits, I’m hoping to push in that direction. Exactly how, exactly what aspect to push on, I’m still thinking about. (First, a thorough re-reading of the research done to date in generative narrative will be important.)
Over time (years) I’m planning to regularly write about this. I’m sure Michael and others will too. It will be very, very challenging work. Progress almost surely won’t be quick.
(This is not to say generative architectures are the only way to do interesting things in the space of interactive narrative, of course. I’m simply saying that to achieve the level of agency Burke is wishing for, generativity is the only viable solution.)
Burke goes on:
The challenge isn’t just a labor-time issue: it’s a theoretical one. The evident comparison here is to the authorial problems and issues that hyperfiction has faced. I don’t think I’m alone in finding that virtually all works of hyperfiction disappoint both in the number and type of branching explorations they allow. Every branch poses the question of its own limitations. The more expansive and imaginatively branching a work of hyperfiction or ergodic literature is, the more the reader asks why he should be limited to the choices provided.
Also worth reading is Burke’s writeup of his experience at the recent Autonomous Agents conference.
July 31st, 2004 at 10:50 am
Timothy Burke, who teaches at idyllic Swarthmore College, comments here from time to time, and had some interesting criticisms of Star Wars Galaxies a while back, has just posted a pair of papers on his weblog about design directions for
July 31st, 2004 at 6:09 pm
You’re describing what I would essentially regard as the only viable way to go as well, save for one problem: I don’t think it’s technically feasible yet. So far generative architectures used with most artistic or narrative work, ones calling upon emergence, complex-systems designs or autonomous agents, solve the problem of creating sufficient content on the fly for readers or consumers of texts, but at the cost of making narratives that has the linearity that I think narrative (or many other art forms) require. The best “emergent” improvisational jazz or music I’ve heard, for example, seems to me to be good enough if one was trying to create a Phillip Glass simulator, but otherwise is missing some elusive quality. Most experiments in narrative are even farther from that–we can create the building blocks of narrative generatively, but so far we can’t get close to the kinds of editing and compression functions that good narrative requires, the kind of selectivity needed. (Facade may be the breakthrough, I agree–I’ve long thought that you were one of the few people out there who has some sense of what’s needed both technically and aesthetically…I am looking forward to it with some anticipation.)
So with a certain skepticism, I guess I think that a narrative-centered, world-centered virtual environment is going to take huge amounts of handcrafting for the forseeable future. I will be delighted–ecstatic–to be proved wrong, because what you’re working on is really the only realistic, viable possibility of something like this happening.
August 2nd, 2004 at 10:58 pm
You’re describing what I would essentially regard as the only viable way to go as well, save for one problem: I don’t think it’s technically feasible yet.
Yes, it’s going to require years (say, 20+) of research and experimentation to get to the level of sophistication we’re talking about here. (Note that video games have already been around for about 30 years, so another 20 sort of seems reasonable!) I’d guess that several people, including myself, are willing to predict it will actually be possible to achieve these kind of results we’re talking about here, with enough effort. Hopefully there will be interesting milestones along the way.
but at the cost of making narratives that has the linearity that I think narrative (or many other art forms) require
The descriptor I like to use for what you’re talking about is “cohesiveness”, or perhaps “cause and effect”; “linearity” I think implies inflexibility.
Facade may be the breakthrough
No, no, far from it… We’re hoping Facade makes varying degress of progress in several directions, but in terms of actual generativity, which is what I’m advocating here, Facade is only marginally generative. Like I mentioned earlier, such a system needs to be writing the sentences themselves, not just sequencing them.
But, if you shoot for the stars, you might just hit the moon. With Facade, maybe we’ve made it into orbit.
Thanks for the kind words though.
So with a certain skepticism, I guess I think that a narrative-centered, world-centered virtual environment is going to take huge amounts of handcrafting for the forseeable future.
I guess what I’m trying to say is, the kind of results you’re hoping for are not going to be feasible with handcrafting. It’ll just be too labor intensive and cumbersome and messy to brute-force / hand-author a solution; it’s too huge of a problem. I thought and hoped differently in the past, but now I am sure about that (as is Michael, who says he pretty much always knew :-).
There is currently no solution to deeply interactive cohesive narrative. The community need to invent new technology (-gies) to solve it. We’ll be experimenting and writing about it for the forseeable future. Kind of like the genome project or something, it’s a problem we could spend our lives on if we choose, it’s that much work. But I think crackable within our lifetimes.
August 3rd, 2004 at 9:55 am
Yes, it’s going to require years (say, 20+) of research and experimentation to get to the level of sophistication we’re talking about here.
I may be more of an optimist about that, Andrew. I’ve be following the paper trail of Facade for more than three years now, and I was always awed about how ambitious a project it is. It might just have been too early to try developing a three-actor drama, whilst the dynamics of an interactive two-actor drama have not yet been explored.
I think we need to figure out two-actor drama first, and I think that a good medium to use for this exploration is chatbot technology of some sort. I think that we need to develop systems that can generate narrative for just one PC and one NPC character before more PCs/NPCs can get involved, like you tried to do it in Facade. You were just a bit early and a bit brave :-)
August 3rd, 2004 at 11:23 am
Dirk, thanks again for your support and faith in the project. The project goals we described in the papers are definitely ambitious, and Michael and I are pleased we did manage to set up an architecture with the open-ended-ness required to achieve the goals. However, the big obstacle to fully achieving the goals became exactly what I described in my post — we couldn’t handcraft anywhere close to enough content as we hoped, about which we’re less pleased :-). So the various project goals were achieved to varying degrees. True generativity is the only viable solution to the fully realize all the goals we laid out.
Once the project is finished and others can play it, I’m sure we’ll be writing up various post mortems about what we believe did and didn’t work, and why. We’ll be really interested to hear everyone else’s take on it, of course.
And I very much agree that it makes sense to solve the relatively simpler problem of just a few characters on screen, before tackling a massively multiplayer narrative.
August 3rd, 2004 at 5:20 pm
We purposely chose three characters (two NPCs and a player) because we wanted to be able to move the story forward, even for shy first-time players. With a chatterbot experience, nothing really moves forward unless the player interacts. With two NPCs, the NPCs can conspire to set up dramatic situations that encourage the player to interact, and give the player something to talk about. Also, chatterbot technology doesn’t generally provide means for creating experiences with story structure. Since chatterbots are relatively stateless, there are no computational structures like goals, dramatic beats, behaviors and so forth that give the author hooks for structuring the experience.
August 3rd, 2004 at 9:40 pm
With a chatterbot experience, nothing really moves forward unless the player interacts. With two NPCs, the NPCs can conspire to set up dramatic situations that encourage the player to interact, and give the player something to talk about.
Usually, chatbots get their clients into short interaction sequences quite nicely – most people will try and write something into that interface box. The question is how to keep them coming, how to take whatever it is they throw at me and wrap a story around it so that they get curious enough to ask: “Why?”. I don’t need outside characters to provide drama, because the communicative situation between a human and a bot is inherently dramatic in itself. Man and machine are trying to understand each other – that’s drama right there.
Also, chatterbot technology doesn’t generally provide means for creating experiences with story structure. Since chatterbots are relatively stateless, there are no computational structures like goals, dramatic beats, behaviors and so forth that give the author hooks for structuring the experience.
True, but it’s relatively easy to extend chatbot technology, using the programming language of one’s choice, and incorporate all kinds of structural models and devices. The problem is not how to give chatbots structure, but which structure to give them. I just think that a three-actor structure is too hard to implement at this moment, and that concentrating on two actors might be easier.