carado.moe

tammy's blog about
AI alignment, utopia, anthropics, and more;

also on twitter, lesswrong, rss, and at

predictablizing ethic deduplication

does the machinery of the cosmos deduplicate identical moral patients and/or their experiences ?

this seems like a very difficult question to even address; but the good news is that for the general future we might not have to care about it. we can simply make it that the superintelligence that runs everything, does deduplicate (memoize) identical computations and data structures, which guarantees that the ethics we build on top of that (for superintelligence to implement) can know about deduplication.

why choose deduplication over no-deduplication? because if we add deduplication on top of any machinery of the cosmos, then we can know for sure deduplication happens, but if we don't implement deduplication, then whether computation is deduplicated depends on the machinery of the cosmos.

"but doesn't this require looking inside arbitrarily encoded computations, such as homomorphic encryption ?"

that is true, but for an aligned superintelligence we require this anyways. otherwise, it could just let unseen pockets of arbitrary suffering happen.