FLI has done a great job of kick-starting a discussion on the research challenges in ensuring the safety of AI systems, by putting forth a proposed program of research in their research priorities document.

To ensure that our efforts are directed effectively, it is important for other AI researchers to engage with the question of what topics are important for AI safety. Towards this end, I've put forward my own ideas about the potential risks from AI systems and what research questions they motivate. There is overlap with the FLI document but also some important differences. I would welcome commentary on the proposed research questions, as well as efforts by others to offer their own proposals. You can find the document here.