I’m interested in how learning, thinking, understanding, reasoning happens. Machine learning is a place where people are trying to figure out how to get it to happen in computers, which was kind of inherently interesting to me. So, in the case of something like a self-driving car, what you need is complex environments where there are lots of realistic physical interactions, and that’s why we need richer training environments. Games are becoming more and more realistic, so Unity as a game engine is kind of perfectly positioned to provide what people in the research field need. There’s a lot that machine learning can offer to games themselves, if you start to think about personalizing an experience. Again, this is something that machine learning could make possible. Now, kind of with all of these more-and-more complex algorithms, we’re starting to think about ways in which these algorithms can then be applied back to the games themselves. So, the two problems that game developers face when they’re trying to design a game is that the game either becomes too easy and players get bored with it and stop playing, or that in later levels, or even from the beginning perhaps, the game is too difficult and players simply quit because of that. Praneet is an intern that we had work with us in the Machine Learning team over the summer. What he’s been working on is taking the platform that we’re developing, and trying to use it in more specific ways that can actually help provide more interesting games for players. So, when I started my internship here, Arthur and me discussed what game we could actually work on, and we both came to the conclusion that TANKS, which is a unity environment, would be suitable for our task to test out the algorithms. What if we have a superhuman AI agent, and actually allow him to adaptively change his level of skill based on the way that you play? It’s a much more personalized experience for them, that you keep on matching their difficulty at every point of time, so you can get a more personalized experience. I feel that’s really cool. Unlike perhaps older fields from a few decades ago, the machine learning community is very open in terms of sharing ideas, sharing code, sharing work, so we think it’s a natural fit to continue this trend of releasing tools that are open, that anyone can use and inviting open collaboration. That’s why we’re so excited to be releasing an initial preview of this product.