Saturday 17 October 2009

The Canon of Procedural Games

While updating the PCG wiki today, I've come across the difficult notion of canon in procedural content generated games. I've flirted with this concept before, by defining games which are prototypically procedural, but in general I've tried to be inclusive rather than exclusive when it comes to including games in the PCG wiki.

I've hit a stumbling point writing up an article on adaptive difficulty - always a controversial point in games. I'll quote the whole article to saving you having to go to the original link:

Adaptive difficulty is the process of adjusting the game in reaction to the player. By spawning new enemies or powering up existing enemies if the player is progressing quickly through the game, or by decreasing the frequency and/or difficulty of existing enemies if the player appears to be having problems progressing, adaptive difficulty techniques attempt to create the 'optimal' game experience.

Classically, adaptive difficulty has been seen as a hard problem, requiring a level of artificial intelligence in the game to attempt to model the player to attempt to determine if they are finding the game easy or difficult.

However simpler RPG style mechanisms can also be seen as adaptive difficulty techniques. Allowing the player to level up by playing through additional easier content can ensure the player is able to grind their way through parts of the game in order to decrease the difficulty of sections of the game where the difficulty level increases. Paradoxically, adaptive difficulty techniques which increase the difficulty of the game by scaling up enemy strength have been fiercely resisted by RPG players, as can be seen by the negative reactions to the difficulty scaling in Oblivion.

Adaptive difficulty is not usually seen as a procedural content generation technique, but it has most of the features of such techniques. It could be seen as decreasing a game's randomness instead of increasing it which would make games which feature it without other PCG features to fall outside the 'canon' of PCG games.

Should I include games which have adaptive difficulty in the PCG wiki? There are plenty of examples of games which have adaptive difficulty and are procedural (Oblivion I've already mentioned, Left4Dead) but there are plenty of games which are not (SiN: Episodes). And I don't want to include every RPG, based on the argument I've made above.

The real question is not whether I should include these games, and the answer to that is probably not, but why? What good reason can I give to not include SiN: Episodes, for instance, as a procedural game?

Again, the randomness argument is the most plausible, but it is not completely convincing.

Convince me.


kaypy said...

"However simpler RPG style mechanisms can also be seen as adaptive difficulty techniques."

Since these can be summarized as 'the player levels up but the rest of the game *doesn't* change', I'm not sure they can be described as procedural difficulty adjustment though. If anything quite the opposite- is there an unprocedural* category available out the other side of merely not procedural?


* ie: where the lack of changing content is a critical element rather than a mere absence. Another example might be the old 2d memorize-the-attack-patterns shoot-em-ups

Paul said...

I'd say a rough version of the incompleteness theorem applies: No genre definition can be both complete and consistent. The Wikipedia article on procedural generation mentions Doom 3 arguably counts because the lightmaps aren't precomputed. However you chose to define PCG, something is going to feel misclassified.

Andrew Doull said...

Paul: For my view on how definitions work, see

While it's nice try to trot out the incompleteness theorem once in a while, it doesn't apply to systems which aren't formal - like human language.

Further thinking about this makes me wonder if adaptive difficulty is not procedural because it relies on input which doesn't vary randomly: the player's skill.

Paul said...

I'd argue it still applies, it's just a question of violating the consistency principle instead of the completeness principal. If you used each phrase in language completely consistently I suspect you would have a formal language (and indeed, formal systems are just a subset of general language used consistently).

But that's all tangential. Your latest definition was what started me towards posting a comment, but wouldn't that rule out creature generation in Spore? If the input is human action instead of random numbers you can still get emergent elements. I think a satisfactory definition is going to need to take into account 1)combinatorial interactions (so random items are of higher order than random names), 2)scope (if entire levels are generated based on user skill that's more PCG then just enemy health), 3)predictability (an expert in a game with adaptive difficulty could know the damage/health of an enemy based on a players performance so far, but couldn't predict a map in rouge...but at the same time being able to predict subelements (e.g. vaults) is more PCG then just random noise).

But it seems to me that any rules I can come up with have exceptions

The Mad Tinkerer said...

Well, I'd say that Sin Episode 1 (because it just seems silly to call it Sin Episodes when there's just the one)does count as an example of procedurally determined adaptive difficulty.

However, it's probably not the most notable example. Left4Dead, which was done by some of the same team, takes a few of the ideas first seen in Sin Episode 1 and takes them to a whole new level as well as adding on a ton more. Like how Mario Bros. is a platform game but Super Mario Bros. is a much better example of a platform game by the same folks.

rabidcat said...

Interesting question. Procedural Content Generation is about content not behavior or experience. AI is about experience or behavior. Content adds to experience but experience does not normally contribute to content (unless it is a parameter into the grammar. different topic). It’s a nit. Who really cares? It’s a topic that tweaks the same brain cells as the other procedural topics. Personally while I find AI interesting, it is in my mind a separate topic complete with its own forums and web sites. AI and behavior is by definition procedural based on premise that it uses grammars, decision trees, fuzzy logic, or Bayesian rules for current generations of implementation.

Didn't mean to hone in on your topic. gives me something to ponder the rest of the afternoon.