Comment by hexane360
3 years ago
At least for the orthogonality thesis, it is a base assumption. It's a claim that cov(intelligence, goodness) = 0. For the instrumental convergence thesis, it assumes rational agentic behavior, which assumes AI behaves like an agent. While this may be reasonable, it's certainly an unjustified assumption.
These seem very reasonable assumptions to me. Of course we can say that even the most evil humans are non-zero human-aligned.