Wednesday, 13 August 2008

Game Design Sketchbook

There's a great article by Jason Rohrer in the Escapist, called Testing the Limits of Single Player. In it, he explores possible games that have deep game play, setting himself the following limits:

Can you make an AI-free, randomness-free, physical-challenge-free, single-player game with gameplay depth akin to that of Go?
Interestingly enough, his solution i45hg is modelled in part on Conway's Game of Life, which is a no-player game that has the above properties. This suggests at least one way around the problem, which is to design a Turing complete interactive game system and let the player explore that.

I can describe another such game, which involves preventing the player explore any previous path. At least, I can't describe the game, but assume that at each stage, there are enough different choices so that the combination of possible sequential choices grows big very quickly. You could design a game this way, which has victory conditions that can be met in a single game, but the player is restricted from ever repeating those victory conditions again. Provided the ratio of victory conditions to non-winning moves is sufficiently low, you'd end up with a game that could remain infinitely replayable without being random.

5 comments:

JohnH said...

Cellular automata in general (including Life and other types) are excellent sources of gameplay depth. SimCity Classic is basically such a specialized automata.

As for Roher's challenge... well, I should say that what follows is of a highly muse-ish nature, I'm basically streaming my thoughts directly into the comment box. Please excuse any gross stupidies.

First, most video games aren't games in the same sense Go is a game. Mario Galaxy is more about exploration and discovery than game-like resource management and decision-making. There is no way to map the elemental joy that comes from making a planet-orbiting long jump onto any aspect of Go.

Forgiving that, we then have to figure out what "depth" means. It's a very vague, squirmy term. In particular, it means something specific for two-player games, the degree to which strategy prevails against an opponent, which is meaningless for single-player, non-AI play. From one way of thinking the answer is "no," because the question is invalid: two-player game depth is a different thing than single-player game depth.

But let's define "depth" as meaning the degree to which the game engages the intellect. Then we come across this fact: it is easier to make a two-player game of a given depth than a one-player game of the same depth. A turn-based two-player game bears at least a minimum depth so long as the first player to take a turn doesn't have an overriding (dis)advantage, and game success is mostly dependent on strategy. If both sides have similar advantage, and both's opportunity to win is determined by his strategy, then the game can conceivably be as deep as the other player's skill at playing. (Note: The requirement that one side not have an advantage rules out games that have been "solved," since it is then known that one side or the other will always win with perfect play. This means that my premise is more complicated than it may seem at first, but it's still not difficult to invent an unsolvable game: Rock Paper Scissors is such a game.)

Further, in the great majority of two-player games, all of the player's actions are aimed at winning, but "winning" means a lot less in a one-player game, especially one which is algorithmically constructed. Perhaps the three most popular goals for such a game: the player might play to see if he can achieve a given win state (such as with Freecell), or to optimize some value (most commonly a score), or just to explore the space of game states (such as with Life).

Of these three:

The first would probably require a designer whose capacity is great enough to win the game, which means the game obviously won't be as deep as Go, which does not easily fit into anyone's mind.

The second, maximizing a score, has potential, but simply making a number as large as possible is not really interesting in itself. It also tends to be unnecessarily reductionist; a strategy that could earn 1,000 points is intrinsically better than one that could best earn 999.

The third, open-ended, exploratory play for its own sake, has the most potential. It is the motive of playing Life itself, after all, and the Sim games too. But it could be argued that these things are still further from Go -- they don't even provide a measure of player skill, after all.

Andrew Doull said...

Feel free to stream your thoughts into the comments box any time - although if you're going to apologise for gross stupidities, you might want to spell it correctly first :)

I've just realised Desktop Tower Defense arguably fills the requirements that Jason Rohrer gives. It has ultimately a reductionist, score-maximising style of game play (Have a look at the high score boards to be depressed how simple some of the maximum score solutions are), while being intensely addictive.

For any exploration based game-play, or for that matter most complex game times, you are probably going to reach a complexity that is equivalent to random. For instance, I could generate a 4 dimensional game space built of a three dimensional universe plus time, which is fixed every time the game is played, but due to minute time differences in your input speed, plays out differently, without necessarily requiring reflex based game play.

Andrew Doull said...

Which leaves us with the question of what constitutes 'deep' gameplay.

According to David Sirlin, whose point of view I tend to believe about this sort of thing, in a two player game you can get away with rock-paper-scissors with an asymetric payoff for each victory type (e.g. rock gives $10, but paper gives $100 and scissors $20).

That's enough to be forced to construct a mental model of your opponent in order to achieve yomi.

In single player, there's no mental model to construct - so to achieve depth, you'd have to be forced to construct a model of something equally complex.

Brog said...

I tried to come up with a definition of what a "game" is a while back. It is difficult to create a definition which is both precise and useful, but one thing I concluded was that there is no such thing as a single-player game.
Any so-called "singleplayer game" is either a toy, a puzzle, or an asymmetric multiplayer game (where, for example, one player builds levels and another player tries to beat them).

So given my definition I can trivially show that his task is impossible. This doesn't mean it's not worthwhile to try making as game-like a puzzle as possible (or that others would even agree with my definition).

I'm going to go off on a tangent now, because once I started thinking of designing levels as playing one side of an asymmetric game, some ideas arose that with your procedural-content fetish you might be interested in.
It seems at first that as a player, the level designer has a ridiculous advantage. This is not the case; if they are playing a game there are rules that they must abide by. Making there be no path to the exit or placing 400 dragons on level 1 would be invalid moves - their boss would not allow them, or else nobody would be the game. When designing a game, you could try to formalise these rules and make this feel more like playing a game. It's easy enough to come up with ideas for how a level-building game could work: dealing cards representing monsters and room types, placing nice items to earn points to spend on nasty monsters. Perhaps the level starts as a simple corridor from start to finish, and valid moves can add branches and kinks, but never cut it off, thus trivially ensuring it stays connected.

This could give a useful approach to generating levels procedurally, which I don't think has been tried. Think of this as writing an AI to play the level-design game. If you have designed it as a game, you can try using traditional AI methods like searching the game tree and optimising heuristics to generate levels. Deal out some rooms, monsters, items, and then let the AI try to find a maximally challenging arrangement of them, within the rules. Shiny items will end up being hidden out of the way behind rooms full of monsters, because it has to put them somewhere but it doesn't want the player to get hold of them.

Obviously this would be more work than the usual roguelike dungeon generation method where stuff is just placed basically at random, but it might give the level some kind of sense to it, and it would lend truth to the story that the dungeons were built by a malevolent wizard.

I haven't tried any of this yet, so I don't know how well it will work in practise. Let me know if you have any thoughts on it.

Tom Hudson said...

Roher's challenge felt almost straw-mannish: if we exclude lots of standard parts of (computer) game designs, yes, it's hard to devise something to fill the gap - that's why those other things are standard parts of games.