Perhaps in many cases, if “X wants Y” then that means X will do or bring about Y unless it is prevented by something external. In some cases X is an unconscious optimization procedure, which therefore “wants” the thing that it is optimizing, in other cases X is the output of some optimization procedure, as in the case of a program that “wants” to complete its task or a microorganism that “wants” to reproduce, but optimization is not always involved, as illustrated by “high-pressure gas wants to expand”.
I think an important consideration is the degree of catastrophe. Even the asteroid strike, which is catastrophic to many agents on many metrics, is not catastrophic on every metric, not even every metric humans actually care about. An easy example of this is prevention of torture, which the asteroid impact accomplishes quite smoothly, along with almost every other negative goal. The asteroid strike is still very bad for most agents affected, but it could be much, much worse, as with the “evil” utility function you alluded to, which is very bad for humans on every metric, not just positive ones. Calling both of these things a “catastrophe” seems to sweep that difference under the rug.
With this in mind, “catastrophe” as defined here seems to be less about negative impact on utility, and more about wresting of control of utility function away from humans. Which seems bound to happen even in the best case where a FAI takes over. It seems a useful concept if that is what you are getting at but “catastrophe” seems to have confusing connotations, as if a “catastrophe” is necessarily the worst thing possible and should be avoided at all costs. If an antialigned “evil” AI were about to be released with high probability, and you had a paperclip maximizer in a box, releasing the paperclip maximizer would be the best option, even though that moves the chance of catastrophe from high probability to indistinguishable from certainty.
But, over the lifetime of civilization, our accumulated experience led us to update this prior, and single out the complexity measure suggested by math.
I may be picking nits, here, but what exactly does it mean to “update a prior”?
And as a mathematical consideration, is it in general possible to switch your probabilities from one (limit computable) universal prior to another with a finite amount of evidence?
No way I’d take that bet on even odds. Though I do think it’s better than even odds. It’s kind of hard to figure out how I feel about this.
Uh, if you’re worried about UFAI I’d be more concerned about your digital footprint. The concern with UFAI is that it might decide to torture a clone of you(who isn’t the same as you unless the UFAI has a ton of other information about you, which is a separate thing) instead of somebody else. It doesn’t seem that much worse from a selfless or selfish point of view.
Funny you mention AlphaGo, since the first time AlphaGo(or indeed any computer) beat a professional go player(Fan Hui), it was distributed across multiple computers. Only later did it become strong enough to beat top players with only a single computer.
This is one of those things that seems obvious but it did cause some things to click for me that I hadn’t thought of before. Previously my idea of AGI becoming uncontrollable was basically that somebody would make a superintelligent AGI in a box, and we would be able to unplug it anytime we wanted, and the real danger would be the AGI tricking us into not unplugging it and letting it out of the box instead. What changed this view was this line: “Try to unplug Bitcoin.” Once you think of it that way it does seem pretty obvious that the most powerful algorithms, the ones that would likely first become superintelligent, would be distributed and fault-tolerant, as you say, and therefore would not be in a box of any kind to begin with.
I think that fully specifying human values may not be the best approach to an AI utopia. Rather, I think it would be easier and safer to tell the AI to upload humans and run an Archipelago-esque simulated society in which humans are free to construct and search for the society they want, free from many practical problems in the world today such as resource scarcity.
We’re talking about the impact of an event though. The very question is only asking about worlds where the event actually happens.
If I don’t know whether an event is going to happen and I want to know the impact it will have on me, I compare futures where the event happens to my current idea of the future, based on observation(which also includes some probability mass for the event in question, but not certainty).
In summary, I’m not updating to “X happened with certainty” rather I am estimating the utility in that counterfactual case.
Gur vzcnpg bs na rirag ba lbh vf gur qvssrerapr orgjrra gur rkcrpgrq inyhr bs lbhe hgvyvgl shapgvba tvira pregnvagl gung gur rirag jvyy unccra, naq gur pheerag rkcrpgrq inyhr bs lbhe hgvyvgl shapgvba.Zber sbeznyyl, jr fnl gung gur rkcrpgrq inyhr bs lbhe hgvyvgl shapgvba vf gur fhz, bire nyy cbffvoyr jbeyqfgngrf K, bs C(K)*H(K), juvyr gur rkcrpgrq inyhr bs lbhe hgvyvgl shapgvba tvira pregnvagl gung n fgngrzrag R nobhg gur jbeyq vf gehr vf gur fhz bire nyy cbffvoyr jbeyqfgngrf K bs C(K|R)*H(K). Gur vzcnpg bs R orvat gehr, gura, vf gur nofbyhgr inyhr bs gur qvssrerapr bs gubfr gjb dhnagvgvrf.
Because assuming Provable(C)->C as a hypothesis doesn’t allow you to prove C. Rather, the fact that a proof exists of Provable(C)->C allows you to construct a proof of C.
The proof doesn’t work on a logically uncertain agent. The logic fails here:
Examining the source code of the agent, because we’re assuming the agent crosses, either PA proved that crossing implies U=+10, or it proved that crossing implies U=0.
A logically uncertain agent does not need a proof of either of those things in order to cross, it simply needs a positive expectation of utility, for example a heuristic which says that there’s a 99% chance crossing implies U=+10.
Though you did say there’s a version which still works for logical induction. Do you have a link to where I can see that version of the argument?
Edit: Now I still see the logic. On the assumption that the agent crosses but also proves that U=-10, the agent must have a contradiction somewhere, because that, and the logical uncertainty agents I’m aware of have a contradiction upon proving U=-10 because they prove that they will not cross, and then immediately cross in a maneuver meant to prevent exactly this kind of problem.
Wait but proving crossing implies U=-10 does not mean that prove they will not cross, exactly because they might still cross, if they have a contradiction.
God this stuff is confusing. I still don’t think the logic holds though.
The Riemann argument seems to differ from the Great Filter argument in this way: the Riemann argument depends only on the sheer number of observers, i.e. the only thing you’re taking into account is the fact that you exist. Whereas in the great filter argument you’re updating based on what kind of observer you are, i.e. you’re intelligent but not a space-travelling, uploaded posthuman.
The first kind of argument doesn’t work because somebody exists either way: if the RH or whatever is false then you are one of a small number, if it’s true than you are one of a large number, you are in a typical position either way, and the other situation simply isn’t possible. But the second kind of argument seems to hold more merit: if the great filter is behind then you are part of the extreme minority of normal humans, but if the great filter is ahead then you are rather typical of intelligent lifeforms. This might count as evidence, and it seems to be the same kind of evidence which suggests that a great filter even exists in the first place: if it doesn’t then we are very exceptional not only in being the very first humans but the very first intelligent life as well.
That’s the funniest thing I’ve seen all day.
Seems to me that if an agent with a reasonable heuristic for logical uncertainty came upon this problem, and was confident but not certain of its consistency, it would simply cross because expected utility would be above zero, which is a reason that doesn’t betray an inconsistency. (Besides, if it survived it would have good 3rd party validation of its own consistency, which would probably be pretty useful.)
Regarding your comments on SPECKS preferable to TORTURE, I think that misses the argument they made. The reason you have to prefer 10N at X to N at X’ at some point, is that a speck counts as a level of torture. That’s exactly what OP was arguing against.
Non-Archimedean utility functions seem kind of useless to me. Since no action is going to avoid moving the probability of any outcome by more than 1/3^^^3, absolutely any action is important only insomuch as it impacts the highest lexical level of utility. So you might as well just call that your utility function.