Deep Space Exploration: Generating Design Challenges

This post is about my work at The Metamakers Institute! It originally appeared here.

Over the past few years game design has become a bigger and bigger part of my everyday research focus, and the more I find out about it and try it out for myself, the more fun I find it. Thinking about systems, communicating ideas, encouraging or discouraging behaviour, balancing strategies, leaving surprises for people to find—all of these things make design games a really rewarding creative activity. The last five years have seen some really important steps towards making game design more accessible to people, like the creation of Twine and PuzzleScript that have helped people cut straight to game-making and bypass a lot of technical barriers. The popularity of tools like Game Maker and Construct is also a sign that more people are able to start creating games without a deep programming background, which is fantastic news.

I'm currently working at the Metamakers Institute, a research group based at Falmouth University in Cornwall. We're interested in using generative and creative artificial intelligence in new and interesting ways, and right now we're looking at how generative software might help people design games. As part of our work we're making an app called No Second Chance, or N2C for short—you can use it to play games and design games, but there's also a bit of special AI secret sauce dabbed on top. We're actually interested in bringing on some new testers, so if your interest is piqued by this blog post do contact us to get on our TestFlight (there'll be another link at the bottom of the page).

No Second Chance

NSC is about designing and playing small physics-based games. Each game is different in some way, but they all have a few things in common: there's always a target in the center of the screen which you move with your finger; and you always need five points to win (circles can be captured depending on the game rules, and some circles are worth points - again, depending on the game rules). Besides these two simple facts, there's a huge set of variable factors that can change in a game, from how big the circles are, how fast they move, what colour they are, where they move towards, whether they're worth points to catch, what happens when they touch other circles and more (in fact, there are over 70 parameters in an N2C game at the moment, and we're adding cool new ones all the time). This flexibility means that games might demand all kinds of skills, like quick reflexes, a steady hand, patient movement, a careful eye, or all sorts of other things. For example, if the circles take several seconds to latch on to the target, but are also gently repelled by your finger, you have to carefully shepherd them around in circles to avoid them fleeing off the screen. If the circles latch quite quickly but are destroyed by other circle types flying across the screen like asteroids then you might need to quickly zip around to protect your scored points.

N2C has two main modes of use right now: a mode for playing games made by you or other N2C users, and a mode for designing games using our simple design interface made up of buttons and sliders. But really, N2C isn't a design app or a game. I've begun to think about it a bit like a pocket-sized possibility space, with different modes for exploring it. Sometimes you explore that space by playing, because NSC games don't have instructions so each game is a discovery process of poking and testing and (hopefully) mastery. Sometimes you explore that space by designing, trying out ideas for games, testing rule combinations and making small tweaks. And most importantly, sometimes you let the app itself help you explore it—we have an AI built into the app that can generate, test and suggest starting points for a new game design. It's still in an experimental stage right now, but we're hoping that a lot of N2C users will ask the generative system for design starting points, as a way of becoming inspired by being given a partially-finished game design.

Partial Generation As Co-Creation

Using generated games as a starting point for human game design is a really interesting kind of human-computer interaction. We don't want the system to randomly produce bad or impossible games, because that doesn't really help the user much. The design space of N2C might seem constrained from the description I gave earlier, but combinatorially it's huge and a lot of the random points in the space are really bad. So we definitely want our system to be clever enough to find 'okay' games. But equally, we don't want the system to produce polished, perfect, unimprovable games, because that might discourage users from changing things or making the design their own—it turns out this is just as well, because generating perfectly designed games is pretty hard. One way of looking at this is that we want our system to generate games that are 'flawed gems'—they have the seed of a good idea, but their design is broken or incomplete in some way. Maybe it needs a little tweak to not be frustrating, or a slider changed just a little to be challenging. Maybe one person would think it was too easy and make it harder, and another user do the exact opposite. Maybe it sparks off an idea that's totally different and you go and make that instead. The point is that it's an inspiration, a creative starting point for you to go and do something.

Making a flawed game is almost as hard as making a perfect game though, especially if you want to know why it's broken or whether it's possible to fix it. It's even harder for real-time, physics-based games like N2C which are computationally intensive to simulate on a desktop machine, let alone a small mobile phone. A lot of automated game design research can playtest thousands of games to try out ideas and get information, but the devices that we're working with (and the attention spans of our users) wouldn't last nearly long enough to do that. So we're exploring new ideas, like using deep learning to analyse the design space for hidden structure about the types of games that can be produced, or analysing existing game designs to uncover relationships between design parameters and their impact on game designs. Hopefully we'll be able to build a game generator that excites and inspires people such that making games with N2C will be as much fun as playing them.

Join The Beta!

Right now we're opening up an early beta version of the app to people for testing and feedback. We can't accommodate everyone, but if you're interested in trying the app out and giving us some feedback, and you have an iOS device with TestFlight installed, you can take part by emailing testflight -at- metamakersinstitute.com with "MeMeMe" in the subject line.