I struggled with that myself, but then figured out a rather nice quantitative solution.
Eliezer’s stuff doesn’t say much about that topic, but that doesn’t mean it fails at it.
I don’t think your solution actually resolves things since you still need to figure out what weights to assign to each of your biases/values.
You mean that it’s not something that I could use to write an explicit utility function? Of course.
Beyond that, whatever weight all my various concerns have is handled by built-in algorithms. I just have to do the right thing.
I struggled with that myself, but then figured out a rather nice quantitative solution.
Eliezer’s stuff doesn’t say much about that topic, but that doesn’t mean it fails at it.
I don’t think your solution actually resolves things since you still need to figure out what weights to assign to each of your biases/values.
You mean that it’s not something that I could use to write an explicit utility function? Of course.
Beyond that, whatever weight all my various concerns have is handled by built-in algorithms. I just have to do the right thing.