​Let's actually try to solve the hard part of alignment.

​Choose your focus:

​The Successor Problem

​How do we make sure that the AIs made by future AIs also have the goals that are good for humanity?

​Get Values Into the AI

​How do we make certain that the values we want are held by the AI?

​Alignment Targeting

​How do formally specify, in maths and code, the right preferences to optimize for?

​Dodging the Morality Question

​How do we sidestep the need to solve Morality? We don't know what is the perfect moral action in all cases and it's unlikely we will in the next 10 years.

​How do we build a Moral Agent of superhuman competence, without needing to solve this?

​Corrigibility

​How do we ensure the AGI will allow us to change it's preferences?

​Automating Alignment

​How would we have AI systems solve the above problems, without the risk of introducing new problems? How do we reduce the added ambiguity from it being a team of AIs rather than a team of Humans?

​Plans to be submitted here: https://ai-plans.com/create