AI welfare as a demotivator for takeover. — LessWrong
kept by eddie
Superhuman AI might choose not to attempt takeover if it perceives the risks as high and the non-takeover alternative as sufficiently good, which we can improve by rewarding honest AI behavior.