This week I’ve been at a Shonan Meeting – a research retreat organised by the Shonan Institute of Technology and Japan’s National Institute of Informatics (NII). It’s designed to bring researchers together to meet without having to present papers, letting them discuss new ideas and plan out research collaborations for the future (and to see Mt. Fuji, which is beautiful). This is my first time in Japan, but previous meetings like this – at Canada’s Banff and Germany’s Dagstuhl – have always been really fruitful and exciting. This is a short summary of what we talked about yesterday in the workgroup I was in, titled AI as Critics, Curators and Best Friends.

As the title suggests, we were interested in building AI that could take on new or unusual roles in games communities, and especially in response to playing games. When we think about AI designed to play games, we usually think of systems that are designed to win, and the observations they make about games only really matter if they help achieve that goal. But performing tasks like critique, curation or recommendation require different things: appreciation; awareness of cultural, social or historical contexts; unique preferences or biases; the ability to form an argument and lots more besides.

The reason “best friends” appears in the title is to draw the line between the recommendations we get from AI today – like Steam’s suggested games – and the kind of recommendations we get from friends. These recommendations are far more special and varied – they can be tailored to someone specific, or they can be a general endorsement for a game; they can have personal, emotional, intellectual or many other kinds of motivation. We actually spent half an hour recommending games to each other to explore this – each of us recommended very different games, and used very different reasons and arguments to do so. Some of us explained how the game worked, some of us told stories about our experience of playing the game, some of us talked about the qualities the game’s design had. There was a huge breadth of approaches.

Another important discussion we had was about the implications for the future, the ways these bots could impact the world in both good and bad ways. Bots that can curate exhibitions with thoughtful themes and justifications based on their experience of playing games is a really promising idea – we imagined a “Museum of itch.io” which changed every day with new curated exhibits, shining spotlights on creators and helping put together a history and context for the tens of thousands of amazing works uploaded to the site every year. But this brings new issues with it as well. Having hundreds of exhibitions run every day would mean everyone got a chance to see their work in special contexts, but it might also make it less special as well. How can we build bots that provide critique, that write recommendations or that curate the big wide history of games, without making it feel hollow and cold?

One of the most promising suggestions we discussed was to build AI with personality. Instead of trying to build a perfect, neutral, balanced AI that can judge us and curate our works from a great position of superiority, instead we should have communities of bots with flaws, quirks, biases and preferences. Yes, it means the curator who only likes games which feature the colour red isn’t going to host the most nuanced exhibitions in the world, but that’s okay, because that bot is just one of many, and this texture and variety might help defuse some of the problems these bots bring, and also add interest and complexity to the contributions they make. Recommendations become more important than Steam’s tag-based suggestions not just because the bots have played the game, but because they have a flavour, a context, a meaning. That way, we can worry less about whether these bots are producing the perfect objective assessment of a game (which doesn’t exist) and more about whether they’re helping people enjoy playing, making or exploring games more (which is what we should be worrying about in the first place).

I don’t have time to summarise every single part of the meeting, but I hope this gave you a flavour of what we discussed! If you want to read more, we’re hoping to write a full paper about our plans and discussions this year, which I’ll link to. You can also follow the group’s participants to see how this maybe influences the things we do in the future: Ana, Gillian, Mirjam, Ahmed, Hiro, Matthew and Mike.

In the evening I played Kanagawa (thanks to Antonios Liapis for bring it and teaching us!) - this is the canvas I finished the game with!

Posted March 19th, 2019.

Back to the blog index