How would you tell when you are mitigating these far-future risks, instead of taking irrelevant actions or making them worse?

What makes you think that AI risk is a tractable problem, or even a well-defined one?