(I read CFAI once 1.5 years ago, and didn’t reread it since obtaining the current outlook on the problem, so some mistakes may be present.)
“Challenges of Friendly AI” and “Beyond anthropomorphism” seem to be still relevant, but were mostly made obsolete by some of the posts on Overcoming Bias. “An Introduction to Goal Systems” is hand-made expected utility maximisation, “Design of Friendship systems” is mostly premature nontechnical speculation that doesn’t seem to carry over to how this thing could be actually constructed (but at the time could be seen as intermediate step towards a more rigorous design). “Policy implications” is mostly wrong.
(I read CFAI once 1.5 years ago, and didn’t reread it since obtaining the current outlook on the problem, so some mistakes may be present.)
“Challenges of Friendly AI” and “Beyond anthropomorphism” seem to be still relevant, but were mostly made obsolete by some of the posts on Overcoming Bias. “An Introduction to Goal Systems” is hand-made expected utility maximisation, “Design of Friendship systems” is mostly premature nontechnical speculation that doesn’t seem to carry over to how this thing could be actually constructed (but at the time could be seen as intermediate step towards a more rigorous design). “Policy implications” is mostly wrong.