I think this is being presented because a treacherous turn requires deception.
Right; my claim is that deception learned in this way will not lead to a treacherous turn, because the agent here is learning a deceptive policy, as opposed to learning the concept of deception, which is what you would typically need for a treacherous turn.
I agree that these stories won’t (naturally) lead to a treacherous turn. Continuously learning to deceive (a ML failure in this case, as you mentioned) is a different result. The story/learning should be substantially different to lead to “learning the concept of deception” (for reaching an AGI-level ability to reason about such abstract concepts), but maybe there’s a way to learn those concepts with only narrow AI.
They’re giving examples of deception being learned which don’t meet their starting assumptions:
I think this is being presented because a treacherous turn requires deception. (This may be a necessary condition, but not a sufficient one.)
Right; my claim is that deception learned in this way will not lead to a treacherous turn, because the agent here is learning a deceptive policy, as opposed to learning the concept of deception, which is what you would typically need for a treacherous turn.
I agree that these stories won’t (naturally) lead to a treacherous turn. Continuously learning to deceive (a ML failure in this case, as you mentioned) is a different result. The story/learning should be substantially different to lead to “learning the concept of deception” (for reaching an AGI-level ability to reason about such abstract concepts), but maybe there’s a way to learn those concepts with only narrow AI.
As I’ve mentioned before, that is technically false (unless you want a gerrymandered definition).