In my first #notGDC post I reported on the first day of work at the Shonan Meeting I was at – a research retreat in Japan, themed around investigating the future of AI in games. That first day I ran a workgroup called AI as Critics, Curators and Best Friends, which was a great opportunity to plan out new directions for game AI that might challenge us to think differently about our work. On the second day, in the same spirit, I joined a group run by Gillian Smith, titled AI for Playground Games. Here’s a little summary of some of the most interesting things we talked about, and what we might try to do next.

Why look at playgrounds? Gillian was interested in finding a space of games that challenge the way we think about building AI players and designers, and exploring spaces where play is more freeform and less constrained. Nowadays a lot of effort is put into building AI to play a very particular kind of digital game. Despite this we use the word “general” a lot, and talk about things like playing Atari games in terms that make it sound more important and all-encompassing than it really is. The term “General Game Playing” is rarely applied to playing games like Proteus, Mafia, Bounden or Tag. Gillian wanted to see what would happen if we shifted our focus a bit, and started from a new place – a playground, as it happens.

We had a lot of fun putting the group together. We shared our own personal experiences of playground games, finding links between Korea, Thailand, Egypt, America, Japan and England, and then at the coffee break invited other attendees to share more games with us – Sweden, Spain, France, Portugal, Australia, the Netherlands and many more countries joined the fray. We also learned about the culture of the playground outside of the games themselves, like ways of picking teams or assigning roles, which include playing quick knockout minigames (in Egypt, Korea or Thailand for example) and just letting the popular kids pick their friends (common in England and America).

Two whiteboard pictures from the group. Left: games contributed by Shonan participants during the break. Right: a partial sketch for a description language. Click to enlarge.

As we discussed these games, we began to identify interesting properties that made them especially challenging or unusual for AI to play. Many of the games have no ending for instance – when does a game of Tag end? When people get tired? When they run out of time? Rules for these games can change during play, they can change based on the weather, or if a ball gets lost (or found). Questions like this make the games feel more structureless than rules-based digital games, which makes it harder to think of ways to format and represent the world for AI in a way that works for the kind of games we normally study. This process was really inspiring, and helped us break out of the traditional ways of approaching AI game playing problems, hypothesising new kinds of physical, game-playing agents that have creative desires, curiosity, and their own ideas of fun.

We spent the afternoon trying to create a new language to describe these games formally, and capture these unusual features – and it was hard. Creating languages is fun, but it’s difficult to get right, and not something that can be polished off in an afternoon. Despite this, the process of trying to come up with a new language to describe something can really help solidify thinking and highlight unseen problems. When does a local rule variation stop being a variation and become its own game, for instance? Does tag have teams, or is everyone playing for themselves? These aren’t easy questions to answer, because the games live in communities and exist only in human language, rather than computer code.

As usual, I can’t cover everything we discussed as there isn’t enough space, but hopefully this gave you a flavour of the problem! We’re hoping to look more at these questions in the future, and maybe begin building some simple simulations to try out some ideas. You can also follow the group’s participants to see how this maybe influences the things we do in the future: Gillian, Ahmed, Kyoong-Joon, Sila and Mike.

On Wednesday we went on an excursion to visit two Buddhist temples and a Shinto shrine, which was a great break from work and just generally pretty beautiful.