Re: Interactive storytelling and me; and a challenge

[WFreitag@aol.com]

Brandon and Benja, since your discussion appears to have wound down a bit (at 
least temporarily), let me say that I've been following it with interest and 
appreciation (which goes for Mark and David and other contributors too). Where 
I've been with my own interactive storytelling efforts is, if explained at 
length, would come across as another boring and vaguely depressing account, 
similar enough to those already posted here recently to be unnecessary. In short, 
though, where I've been is at the Forge (www.indie-rpgs.com), which I've 
already recommended (on this list, some two years ago, more or less) that anyone 
interested in interactive storytelling must visit. It makes no sense any more to 
talk about metastory generation, either positively or negatively, without 
being aware of such shipped products as My Life With Master by Paul Czege. It's 
not that this game or others like it settle the issue, either by succeeding 
unambiguously in metastory generation or by proving it unworkable. But 
collaborative on-the-fly story generation by people, using framework mechanics that may 
include the use of randomizers, has become far more sophisticated than the 
"take turns adding a paragraph" methods Benja and Brandon recently discussed. 
This certainly sheds new light on the question.

Combine My Life With Master with Erasmatron (an enormous undertaking, to be 
sure) and interactive metastory generation would be a done deal.

My Life With Master is a tabletop role playing game. But those whose idea of 
role playing games is Dungeons & Dragons (that is, most people) would barely 
recognize MLWM as such. The game's mechanics don't tell the players what 
happens specificially (someone gets injured, someone kisses someone, or whatever). 
Rather, it provides a series of appropriate constraints on the story the 
players collectively narrate, in the course of a completely playable competitive 
game. For instance, a die roll might determine whether a "minion" character 
succeeds or fails at "making a connection" with a "connection" character, and also 
determines who gets final say in narrating how and why the success or failure 
came about. But the narrator is free to (and is required to) instantiate what 
actually happened in the scene. This is just about the exact inverse of 
traditional tabletop role playing (such as D&D, again) in which the rules focus on 
internal cause and effect and leave it up to the gamemaster to force, cajole, 
or fudge the resulting sequence of events into a semblance of a story.

Take a look at one sample rule from the game: If the result of a die roll 
would increase a minion character's Self-loathing score, and the current 
Self-loathing score would thereby be set to (or is already at) a value greater than 
Love (another character score) plus Reason (a global score), the minion's 
Self-loathing score doesn't increase. Instead, the minion loses his next scene, and 
on that player's next turn a type of scene called The Horror Revealed occurs 
instead. The Horror Revealed is a scene in which something horrible occurs, or 
the pre-existence of something horrible not previously established is 
revealed, affecting the environment (rather than any of the focal characters 
specifically). 

In one actual play example, in a game where the Master character had been 
established to be a clockwork-mechanical genius, and it had previously been 
established that his minions had been abducting children, a Horror Revealed scene 
depicted a deadly attack on some villagers by monsters consisting of the heads 
of previously abducted children attached to mechanical spider bodies. (Please 
don't be confused by the word "revealed" into thinking that this was something 
prepared in advance by the gamemaster or in a scenario text. It was invented 
on the fly by the player the system designated to narrate that particular The 
Horror Revealed scene at that particular time.)

Note that the rule that brings about The Horror Revealed has nothing to do 
with causality within the game world. The minion's self-loathing doesn't cause 
the horror to occur or be revealed. The horror that's revealed doesn't have to 
have anything to do with that particular minion at all. Rather, a minion 
having high Self-loathing relative to Love and Reason is a handy and (apparently) 
reliable indicator that, on an authorial level, the story has reached a point 
of dramatic tension where a Horror Revealed scene is likely to be aesthetically 
appropriate. Since die rolls are involved, there's no guarantee that a Horror 
Revealed will occur at any given time, and no predicting e.g. how many will 
occur in the course of a given game. The rule only guarantees that when a 
Horror Revealed DOES occur, it "fits" with the pacing and style of the emerging 
narrative. Many of MLWM's rules work like that.

Here's a link to a discussion thread containing a complete account of a 
recent MKWN game from start to finish: 
http://www.indie-rpgs.com/viewtopic.php?t=15555&postdays=0&postorder=asc&start=0. It's seventeen posts long, with the 
posts ranging from a paragraph to 2 pages or so. This should give you a better 
idea of how the game works and the kinds of outcomes it produces. (It's still an 
active thread, so it might be longer than 17 posts when you get there, but the 
first 17 are a pretty complete account, if you stop there.) If you want to 
know the rules in detail, you can find many examples mentioned on other Actual 
Play threads at the Forge (also look for the MLWM forum in the Inactive Forums 
section), or buy the game. (here: http://www.halfmeme.com/howtobuy.html)

Then, we can start talking about Dogs in the Vineyard, The Mountain Witch, 
and a dozen or so other recently developed tabletop role playing games that are 
just (or almost) as interesting from an interactive storytelling point of view.

Although I agree that producing narrative that meets the needs of the players 
in the monent is sufficient for a role playing game or other interactive 
storytelling exercise, these games show that that's far from the best that can be 
expected. Every Actual Play account of MLWM play I've read has compared 
favorably, in terms of narrative quality, with Hollywood's average and some have 
rivaled Hollywood's best.

So, here we have a fully working example of something that interactive 
storytelling people (including me) have been blathering about the "possibility" of 
for years: a complete set of "plot structure rules" for creating a story 
interactively. Granted, this particular set of rules applies only to one very 
specific subgenre of story. (Hey, what more do you want for $8.95?) But the relative 
simplicity of the rules, and their effectiveness in application, makes me 
wonder why similar rule systems weren't invented a long time ago and why they 
aren't being churned out and tested at media labs now. The reason might be that 
folks bent on interactive storytelling as a digital medium haven't been able to 
separate the problem of plot structure rules from the problem of 
instantiating those rules -- that is, for instance, actually being a worthy player in a 
game like MLWM -- by means of a computer program. The latter problem is a bitch, 
to say the least.

But perhaps it's a problem that AI techniques can now begin to chip away at, 
describable by a series of necessary abilities that collectively are 
guaranteed to add up to interactive metastory generation when all are in place and 
plugged into the overarching MLWM-like rule structure. For instance, a system that 
answers the question "given world state W, and character personality model C, 
what does character B do when character A attempts to declare his love for 
her?" is clearly not useful or necessary to play MLWM. We can see that to play 
MLWM, a system instead has to answer questions like "given world state W and 
character personality model C, why might character B plausibly fail to deepen 
his connection to character A by declaring his love for her?" With the possible 
answer-space ranging from "because character A finds character B repulsive" to 
"because just as he's about to declare his love, a monster comes along and 
carries her away." Possible? Maybe; it would appear to be able to draw on a l
arge body of existing research in AI planning. Can a computer program invent 
Horror Revealed scenes from scratch? Clearly not. But selecting from a hundred 
pre-formed Horror Revealed scenes and modifying it based on current context... 
that seems more reasonably achievable, though certainly not easy.

And that's where I stand today. Understanding gained. Progress made.

And, I must say, a different perspective on the debate about the possibility 
or impossibility of interactive metastory generation. One side argues, 
"airplanes are proved impossible (and probably worthless anyway) because all efforts 
to train gerbils to build them have failed." The other: "the existence of 
airplanes proves that gerbils can build them if sufficiently trained." (Airplane = 
interactive metastory generation; gerbils = computer; airplane built by 
trained gerbils = solo-player digital interactive metastory generation.) The 
problem won't be solved by better tools for gerbil-trainers; what might actually 
help is better tools for the gerbils.

- Walt