header image

How to design games without designing them

A rather thought-tickling post over at Togelius on what the author describes as “automatic game design”. There’s even a paper on the concept that not only goes into further detail, but maps the output of a game (or games, I should say) built on the principle.

Automatic game design is exactly what it sounds like: a game built by an intelligent system iterating through entities and assigning rules to them. Entities can be anything from a power-up, enemy or something inert regardless of their appearance – no assumptions are made. The resultant games are then marked on how easy they are to complete by tracking scores, player/enemy deaths, time played, etc. The system then produces a new game based on what it has learned.

Essentially, it attempts to quantify fun.

Some limitations are immediately apparent. You can’t have an entirely random system, constants must be set. Secondly, even with these limitations, the odds of making the next Halo or Oblivion are remote. If you could build a system that could create these types of games at the same quality, I would argue the time would have been better spent creating them the old-fashioned way. I can’t begin to imagine the calibre of game you could create with all the resources you’d pour into this thing.

But these limitations are way besides the point. The author doesn’t claim AGD to be the second coming, but a proof-of-concept and a thought-provoker (what I believe is the most rewarding takeaway). Further to this, it can produce permutations – as simple as they may be – that a human designer might never conjure themselves. Those permutations could then evolve into actual, working concepts. The paper suggests the system could be used to fine tune a near-completed game as all the rules and entites have been clearly defined, though I’d label this “automatic balancing”.

I think AGD’s biggest weakness can be summed up with the adage: what looks good on paper doesn’t always work out. Designing often requires a degree of “feeling” out what’s fun and right. It’s a quality no amount of rules or parameters can replicate. But AGD could get reasonably close and, for that, it could be an excellent tool in the designer’s arsenal.

Anyway, it’s a chunky concept to digest all at once, so I may have further to say later on. In the meantime, if you’re looking for a deep read with a sprinkling of maths to pass the time, I can recommend it.

~ by Logan on January 15, 2009.

2 Responses to “How to design games without designing them”

  1. These are my first impressions having not read the actual paper thoroughly yet, so this might not be quite possible with what they’re presenting.

    Instead of applying this to actually designing entire games, the concepts could be applied to automatically balancing the gameplay. Simple things like tweaking the parameters of enemy entities, powerup types, scores (if that’s relevant) and so on. You set up the rules of the game as a designer, you define some mutable parameters, maybe set a minimum and maxium, seed them with some ‘best guess’ values and then tell the game what sort of difficulty curve you’re after and it goes in and simulates everything and optimises things so that you can avoid things like wonky difficulty spikes and so on. At the very least it could be a useful first-pass utility to give you an initial base that you could manually tweak and so on.

  2. An interesting concept, I agree that it could be a great tool for balancing a game, even post release.
    I’m probably super imposing my own ideas onto this a little, but the idea of entities mutating based on their interaction with the player and other entities could be useful for entity AI – In an ideal gaming world, entity relationships should define entity AI/actions.