This is a quick summary of a few potential problems, flaws, questions, objections to the idea of Volitional Friendliness. Thinking about it, some questions assume a Sysop or similar type of situation.
Not all of these are guaranteed to make sense. Much of this could be worded in a clearer way, and organization could be much better, so I'm not completely happy with it. I'll post it anyway.
I obviously don't expect anyone to answer all these questions (it's likely that many questions here will have the same answer) - I think that a framework in which they can be answered should be given, though.
I would also like to know what problems Quicken sees that have to do with free will.
Alright, here it goes.
---
Why does the absence of an objective morality justify Volitional Friendliness? What is the rationale behind it? This is not intuitively obvious. Maybe it's answered in CFAI, in that case I should go read it again; but I don't recall anything convincing.
Is it a utilitarian sort of thing - minimize the extent to which volitions are violated integrated over the future of the Universe - or a duty-ethics sort of thing - it's always wrong to violate someone's volition?
If the first, how do you define and measure the degree to which someone's volition has been violated?
If the first, couldn't it be advantageous to do a one-time big assimilation event so that many volitions are violated now but none will have to be violated in the future?
If the second, is every deed that causes anyone's volition in your future light cone to be violated ethically forbidden? If so, isn't every deed already forbidden? If not, what deeds are ethically forbidden and what deeds aren't? Is a distinction between "causing" something and "not preventing" something needed?
If neither the first nor the second, but if making violation of volition impossible is merely a heuristic that is useful to some other goal instead of something to be pursued for itself, what is that other goal?
Does volition "count" across time - do your past volitions count, or do only present volitions count? (Comments to the quote from "Diaspora" in CFAI suggest the first, other information suggests the second)
If past volitions count, what notion of "diachronic identity" is used to identify mind states at different times as being the same person? What evidence is there that such a notion could (or would have to) exist without being completely arbitrary? (*)
If the definition of identity will be chosen by the mind it applies to, does that mean I can say "future-me" is the whole Universe?
If only present volitions count, does that mean it would not be unethical for someone to control someone as a "brain puppet", provided some sort of assimilation has already been done?
If past volitions don't count, does that mean I can't say "do what you think is best without bothering me", or anything else that involves planning my mind state in advance? Does it mean volition has to be re-verified at every moment?
Do I have volitional "authority" over anything other than whether my mind state changes, is destroyed, etc?
If yes, over what?
Can a Post-Singularity "information sea" (I like this term that Stephen Baxter uses better than "brain soup" ) always be unambiguously resolved into different minds (or persons, or decision mechanisms, and so on)? If minds can overlap, what happens if they overlap for 99.9%? Are you counting them as having double normal validity?
What about changes to reality that don't change one's mind state, but would otherwise be considered important - such as unnoticed uploading? If unnoticed uploading is something that volition applies to, what other aspects of physical reality that do not affect one's mind state does it apply to?
If unnoticed uploading can be a violation of volition, can other changes to the matter one is implemented on that don't change the mind that is implemented on it - such as small translations in space - also be violations of volition?
Is it a violation of volition to suspend someone's mind-program for one attosecond (against vis will)? If yes, is it an equally large violation as killing? (If not equivalent to killing, does the recreating in fact cancel out the killing? If yes, can recreating other people cancel out a killing?)
If this is not a violation, what time does it have to be (instead of an attosecond) to become a violation?
Do people always have volitions? How are they defined?
If you intend to find out volitions by asking, doesn't the volition often depend sensitively on how the asking is done? And can't the asking itself be a violation of volition, to those who don't want to know things, or don't want to be disturbed?
If you intend to find out volitions by scanning brains, how do you deduce volitions from this information? If by simulating, isn't that unethical? And what about privacy?
If this is solved by substituting "informed preferences", how informed? Superintelligently informed?
Volitional Friendliness has counterintuitive consequences, such as people killing themselves on a whim (easy with a Sysop that satisfies requests), insane people torturing themselves on a whim, and so on.
If all these questions are not for us to answer and will have to be answered by the AI, what is evidence that these questions have correct answers at all? By what standard would different definitions of volition be judged?
And what of all this will be needed when first programming or explaining volition to the AI?
I also think there should be another term for volitional Friendliness. The way it is now is confusing and allows for conflating different concepts. Random suggestion: v-Friendliness for volitional Friendliness, s-Friendliness ("seed") for Friendliness meaning developing from the human moral frame of reference.
(*) I don't believe an obvious definition for diachronic identity exists (at most an approximate one) (I think it is just a matter of definition rather than fact), because:
- physical continuity is not necessary or sufficient, as any uploadist will agree;
- continuity of consciousness isn't necessary, as is proven by going to sleep;
- memories aren't necessary or sufficient, otherwise one could become a different person by having different memories implanted, or by forgetting things; also, this is a matter of degree and not kind
- the only thing I see left is similarity (of personality, etc); that is also a matter of degree, and I think it wouldn't work for other reasons
- there is no reason it should be possible to extend a notion of personal identity formed in a world where there is conveniently one brain per meatbag to stranger posthuman situations; if any ethical philosophy requires it, that's that philosophy's problem.
---
Discuss, or else