Human-wise this is an easy question, human will isn’t perfect, but what about an AI? It seems to me that “true precommitment” would require the AI to come up with a probability 100% when it arrives at the decision to precommit, which means at least one prior was 100% and that in turn means no update is possible for this prior.
I think wwa means 100% certainty that you’ll stick to the precommitted course of action. But that isn’t what people mean when they say “precommitment”, they mean deliberately restricting your own future actions in a way that your future self will regret or would have regretted had you not precommitted, or something like that. The restriction clearly can’t be 100% airtight, but it’s usually pretty close; it’s a fuzzy category.
Is true precommitment possible at all?
Human-wise this is an easy question, human will isn’t perfect, but what about an AI? It seems to me that “true precommitment” would require the AI to come up with a probability 100% when it arrives at the decision to precommit, which means at least one prior was 100% and that in turn means no update is possible for this prior.
Why? Of what?
I think wwa means 100% certainty that you’ll stick to the precommitted course of action. But that isn’t what people mean when they say “precommitment”, they mean deliberately restricting your own future actions in a way that your future self will regret or would have regretted had you not precommitted, or something like that. The restriction clearly can’t be 100% airtight, but it’s usually pretty close; it’s a fuzzy category.