Game design musings

[alexxkay]

Last night, I was reading an article in Game Developer Magazine, which was a roundup of various gaming luminaries putting in their two cents as to what revolutions the next ten years will bring in gaming. Quite a lot of them mentioned some form or another of "non-linear storytelling" or "interactive storytelling", which has long been an interest of mine.

It occurs to me that the term "non-linear storytelling" in the gaming community shares a lot of similarity with the term "artificial intelligence" in the academic community. Both of them describe semi-chimerical moving targets. In academic AI, once a technique is well understood, it's often not considered to be "really intelligent" any more. "Sure, Eliza fools a lot of people, but we understand the simple way in which it's faking it, so it isn't really intelligent." In the same way, game developers keep inventing new techniques for interactive storytelling, but as soon as we understand them, we decide that they are not the "holy grail" we've been looking for.

Of course, this similarity is no coincidence. After all, truly non-linear storytelling is in the set of what I refer to as "AI-complete" problems. For a full solution to be implemented, you need something that is equivalent to (and perhaps indistinguishable from) an actual AI. Many such problems are linguistic in nature; for a program to process language (written or spoken) in a complete manner requires (at least) human-level intelligence.

Comments

Re: Competition

[jducoeur]

And, yeah, the only good storytellers around are humans (Pan narrans, to quote The Science of Discworld II); the task is pretty well certain to be AI-complete--even most humans don't manage it very well, after all.

I don't know if this logically follows: essentially the same statement has been made in lots of fields, until people found the right "cheats" to be able to produce decent algorithmic approximations.

I mean, think about chess. Once upon a time, this was viewed as the canonical AI problem: a computer couldn't "think" abstractly enough to beat a human. Then clever algorithms and brute force came along, and lo -- master chess players started to be defeated by them. The computer doesn't go about it in anything like the same *way* a person does, but it still accomplishes the goal.

Another good example is the Thief engine. This produces in-game characters that move and sound astonishingly real sometimes. But it's all smoke and mirrors: carefully-constructed scenarios and well-designed stimulus-response engines, with just enough degree of freedom to appear real while still being deterministic under the hood.

Personally, I think it's mainly a matter of analysis. Truth to tell, I've given this question at least some lightweight thought -- back when I was at Looking Glass, we did a proposal for Ultima Underworld Online that was fundamentally grounded in this. I don't remember all of the handwaves, but the high concept was that we would use a combination of semi-randomness, control of world events and observation of user actions to create a concept of "destiny" for users: we would have oracles that would steer users in fluidly-evolving campaigns with as much thematic unity as we could accomplish. I have no illusions that this would have been perfect, nor that it would be easy to do at all, but I'm fairly confident that we could have come up with "cheats" that gave a good measure of story-like flavor without any real intelligence being involved at all.

Computer games are entirely the art of illusion. The most successful ones are usually the ones that understand that fact, analyzing the surface aspects of their problem space to simplify out the key facets. The resulting illusion isn't perfect, but can be plenty good enough if you constrain the problem space carefully...

Re: Competition

[metageek.livejournal.com]

And, yeah, the only good storytellers around are humans (Pan narrans, to quote The Science of Discworld II); the task is pretty well certain to be AI-complete--even most humans don't manage it very well, after all.

I don't know if this logically follows: essentially the same statement has been made in lots of fields, until people found the right "cheats" to be able to produce decent algorithmic approximations.

I mean, think about chess. Once upon a time, this was viewed as the canonical AI problem

True...but storytelling is much harder than chess; and true interactive storytelling is harder still. I mean, once you have a program that can play the role of any character in the game world, and react in a human-like way to whatever the player does or says, how far is that from being able to play the role of a real human? That's what AI-complete means, after all: a problem is AI-complete if any solution qualifies as artificial human mind. (Not necessarily a smart one, I suppose...)

And yet...your idea of oracles makes me think again. It
doesn't strike me as a full solution to the problem, but I suppose it would be interactive storytelling; you'd be exploring the story and interacting with the storyteller from time to time. It's just that it would be hard making it a good storyteller.

Re: Competition

[alexx-kay.livejournal.com]

storytelling is much harder than chess

Unproven (and possibly unprovable). They are very different problems, to be sure. But it is not clear to me that one is necessarily harder than the other. We may simply have not found the right "cheats" yet.

To come back around to my original post, Eliza is totally built out of cheats, but it's good enough to pass at least a limited sort of Turing test. It is at least conceivable that good enough cheats (or at least, much better than we have now) may be found in the realm of game storytelling. Of course, as soon as some significant cheats are understood, they get redefined as "not really interactive"...