Dagstuhl Reports - AI for Accessibility
Posted Dec 17th, 2019
This week I’m at Dagstuhl, a research center in Germany that hosts computer science seminars where people come together to discuss the future of fields like artificial intelligence. I’m meeting with games researchers, designers, mathematicians and more to think about what the future of games and AI will look like. This is a writeup of the first working group I was in, a half-day session titled Accessibility and AI, led by Tommy Thompson who you may know from his YouTube series, AI and Games.
Tommy was interested in discussing how AI might be used to build new kinds of tools and technology that would help people with a wide range of disabilities and accessibility needs. Due to the shorter time window on the first day, and the open-ended nature of the group, we spent most of our time discussing existing work and thinking about areas that were possibly (to our knowledge, at least) underserved by AI research.
This is a space that AI could do a lot of good in - millions of people have some kind of accessibility needs in order to play games (including many of us in the work group), and many of them play games despite these needs not being met, because games are an important part of their lives. AI also hopefully might lower costs for providing access to some kinds of game. Specialist controllers can be expensive, but AI could provide software-based solutions using existing (or simpler) hardware - or open up new avenues to accessibility that can only be provided via software.
Our afternoon whiteboard notes. Left: categories and impact areas. Right: working areas and challenges. Click to enlarge.
Went spent a while looking over existing resources like the Game Accessibility Guidelines which suggests many areas where games can provide assistance to players. One of the interesting things about the list is that a lot of the more advanced topics in particular require a total design overhaul or a lot of pre-planning. The GAG website itself acknowledges that most developers will only manage to target a subset of these features. We hypothesised that existing AI research might be reapplied in this space, such as automated game design. Could we have AI game designers rework and restructure games based on the needs of an individual player, stripping out features or UI elements that the player doesn’t need or want, perhaps even replacing it with new mechanics or elements better suited to that player? This raised questions about what it means to play a particular game - how far can we adapt a game using AI before the player feels like they aren’t getting the experience they want? There are many unanswered questions here, but the connection between the two spaces was interesting to talk about.
Emily Short spoke about the extensive use of screen readers in the interactive fiction community, and how accessibility standards have emerged for interactive fiction platforms and players. We researcher other spaces in which screenreaders are used, like roguelikes, and considered how tools like this might allow games to be transformed from one space to another. How far can we take screen reading? We couldn’t find evidence of screen reader usage for games like Civilisation, for example, but in theory a good AI system could summarise the game and answer player queries about the world and what options are available to them.
It was only a short afternoon discussing things, and it’s a space in which so much work has been done, often by charities and individuals without much support. All of our next steps require us engaging with experts: those with disabilities themselves, most importantly; charities like Able Gamers; and researchers in other spaces who have looked at adapting or designing games for spaces like this. The existing work in this space is inspiring and uplifting, and I’m sure we missed a lot of work that’s been done already (as is often the case with these working groups) but I really gained a lot from discussing these ideas with people, especially when it recontextualised my existing work on automating game design and content generation.
Posted Dec 17th, 2019.