First off, let’s talk about the Silicon Minds competition. Where did the idea come from?
The Applied Games group, which is part of Microsoft Research Cambridge, have pretty much driven the whole thing. We came up with the idea of an artificial intelligence games contest, and then we talked to the XNA people. Then we talked to our colleagues at Rare and Lionhead and they were also excited about the idea too. So they offered internships for the finalists, most likely in collaboration with us – we’ve had joint internships with them before and it’s worked really well. We’re looking forward to seeing what people are coming up with.
Who will be judging the entries?
The judging panel is going to be us at the Applied Games group, people from Rare and Lionhead, and an academic panel, which is made up of ten AI specialist academics. It’ll be both Microsoft people and external academics reviewing the game design documents and the finished articles.
Do you have any advice on the sort of things you’re looking for in the entries?
The first thing we’re looking for is finished games. It’s a really tight deadline – entries have to be in by January 27th. Originally we wanted people to demonstrate machine learning techniques, as that’s what we work on here at Microsoft Research Cambridge, but then we decided to widen it to the whole AI field. The games can be anything – board games, strategy games, first person shooters, whatever. It’s a great opportunity to see what people understand about machine learning and AI.
What would an internship at Microsoft Cambridge offer in comparison to Lionhead or Rare?
I think what we can offer is cutting-edge research – working on stuff that isn’t even out there yet. For example, we’ve been working on reinforcement learning, where you teach an AI robot how to learn instead of just giving it orders. So picture a racing game that we have, where you tell the car that it has to drive fast, but let it figure out itself which actions are good. It’s very impressive to see the demo – for the first lap, the car doesn’t drive very well, but it learns from its actions, and it’s this adaptive intelligence we’re interested in. So imagine, for example, that it started to rain in this game – just like a human would do, the AI has to adapt for driving in rain. So MSR Cambirdge is a good choice for people who are interested in research and who are keen to look into the future.
So you guys have a lot of involvement with Lionhead and Rare?
Yeah, we’re working closely with the both of them on future applications of machine learning to games. For example, Lionhead are working on the idea of one-button combat in Fable 2, and that’s a fairly ambitious plan. We’re investigating whether it would be possible to apply the same sort of techniques as those I just described in the racing example, where ideally it’d be fun if you could have your character learn and adapt its fighting technique from experience, or even switch between attack and defence mode and realise that the objective now is to minimise damage received.
Aside from the work we do with Lionhead and Rare, we also do research on intelligence for the game of Go, and we also developed the TrueSkill ranking system that is used by most Xbox Live games to matchmake players of equal skill as best as possible. Also, we worked on Forza Motorsport on the Drivatars, where players can train an AI car by showing it how to drive. We are now working on more advanced AI for racing based on reinforcement learning where the car learns to drive on its own.
Finally, what’s Microsoft’s aim with the competition – to get talented interns through the door of its facilities?
That’s one of the reasons, sure, but one of the original reasons is that we know many people doing game AI research and we wanted to sell the XNA platform to them, because we think it’s a great platform. Also, we’re deeply convinced that game AI needs to get further developed, so as researchers we’re concerned by that. So in addition to just getting interns, we’re really interested in pushing game AI research forward. Advancing AI is one of our big aims here at Microsoft Research Cambridge.