My motivation for getting into machine learning was the hope that I might be able to contribute to the development of robust intelligences like our own. Of course, it's difficult to determine where to go in order to best fulfill that. Currently I've been exploring deep learning, and I'm intrigued by spiking neural nets.

A topic I find particularly interesting is how to bootstrap learning through multiple modes. For example, humans are very good at taking knowledge from previous areas and applying them to new ones (which has been studied to some extent in machine learning as transfer learning). But we also do this in quite sophisticated ways, for example you might play some game with a friend, learning the basics, and then go read about more advanced strategies, then play some more and gain a better understanding of those strategies. These two activities synergize and "bootstrap" each other; the more you play, the more you understand when you read about the strategies, and the more you read, the more effectively you can learn while playing. I'm interested in how to get computers to do this type of synergistic learning.