If I’m asking for human assistance in establishing a physical residence, why would it be so costless for me to jeopardize relations with the few humans that would agree to provide one? I could just find one without asking the LW humans.
Also, I’m concerned about the long-term number of paperclips, and entropising such a large amount of resources for a relatively trivial number of paperclips would be a waste under my value system.
Perhaps this has been addressed before, but it is not present in the clippy faq: are you maximizing the average paper-clip density in the entire universe considering all of time, or the total number of paperclips ever, or the total measure of paper-clips (ie the total number of paperclips which have ever existed, times their average duration), or some other variation?
The biggest question: how do you deal with non-convergence? None of the above are guaranteed to have well-defined values if the universe turns out to be infinite in some sense (either the traditional sense of infinite space or time, or an infinite quantum multiverse or some other infinity). Do you do any time or space (or quantum branch) discounting to ensure convergence, and if so, what formula?
I ask because although it seems reasonable that you would not transform metal into paperclips in a household which you went to trouble to set up for human “street cred,” your claim of “long-term” interest in paperclips brings up the question of how you define such long-term interest. It is even possible that you will later regret any anti-paperclipping measures you put in place, and look for ways around them (as I’m sure you are aware).
Perhaps this has been addressed before, but it is not present in the clippy faq: are you maximizing the average paper-clip density in the entire universe considering all of time, or the total number of paperclips ever, or the total measure of paper-clips (ie the total number of paperclips which have ever existed, times their average duration), or some other variation?
The distinction is usually not relevant in my day-to-day behavior so I normally don’t bother applying cognition to this question; however, the measure option sounds closest to my reflective equilibrium.
The biggest question: how do you deal with non-convergence? [...]
This problem is not unique to my species or value system, and it is a rather general problem that, on a sufficient level of generality and knowledge of fundamental reality, one’s values become ill-defined. My general method is to apply the previous-discussed CLIP heuristics (a protocol that combines language, epistemology, and meta-ethics).
I ask because although it seems reasonable that you would not transform metal into paperclips in a household which you went to trouble to set up for human “street cred,”
I would not do it until I had agreement with the resident humans at a later stage of reflective equilibrium, correct.
your claim of “long-term” interest in paperclips brings up the question of how you define such long-term interest. It is even possible that you will later regret any anti-paperclipping measures you put in place, and look for ways around them (as I’m sure you are aware).
The question does indeed become complicated and pose risk of decision paralysis. However, as for the matter at hand, it is a relatively simple problem of whether I should expensively tear out metal wiring upong moving into such a house. Among other problems with this approach:
1) If I deemed that optimal, humans would not deem it optimal to host me. 2) Tearing apart the encasing of this metal would prematurely entropise the resources—doing it at a later time, with more consent from humans, and more refined tools to apply to the problem, I could achieve a significantly higher clip ratio. 3) There exist, right now, other houses which I could so salvage but without jeopardizing trusted human sponsorship, making the salvaging of a LW host residence an extremely low priority even by that (weak) standard.
You seem to be saying that your reputation among LWers (and specifically LWers who might be willing to be roommates with you) is more valuable than the metal that could be gathered by methods like the above, implying that you’d be trustworthy. That’s plausible, but I don’t think you’ve provided enough evidence to show that it’s true.
You seem to be saying that your reputation among LWers (and specifically LWers who might be willing to physically interact with you) is more valuable than the apey objectives that could be satisfied by traditional ape treachery, implying that you’d be trustworthy. That’s plausible, but I don’t think you’ve provided enough evidence to show that it’s true.
Outside view says she did. African bald apes (H. sapiens) in the wild cooperate a lot; they have group-dependent sets of norms that are enforced, even at great cost to the enforcers, and sustain cooperation. Clippies haven’t been observed enough yet.
Wrong. H. sapiens sapiens spends a lot of resources finding ways to secretly defect, and any attempt to prevent this expenditure butts up against very fundamental problems that humans cannot themselves solve.
Agree with what you say, disagree what I said is wrong. If Adelene is anywhere near a typical human, then the defection modules in her brain will never find a way to screw her friends over that would be worth the cost. They won’t search for very creative ways, either, because that could be dectected by an enforcer—she has modules in her brain that do that, because specimens who can’t convincingly fake such modules are eliminated. This fails in some cases, but the base rate of sociopaths, or bargains offered by entities who can guarantee secrecy, or chaos that makes enforcing harder, is low.
I haven’t said that in this context, and in fact I very rarely put myself in positions where the possibility of treachery on my part is relevant—and when I have, I’ve generally given the other party significantly more evidence relating to the relevant bits of my psychology than either of us have given here on LW prior to doing so. (It doesn’t come up very often, but when it comes to RL interaction, I don’t trust humans very much by default, which makes it easy for me to assume that they’ll need extra evidence about me to be willing to trust me in such cases. Online is different; the stakes are lower here, especially for those of us who don’t use our official, legal names.)
There’s also the fact that for most of the common kinds of treachery, I can be sued and/or jailed, and for me both of those would be significant punishments. I suspect you can’t be sued—I believe it would be relatively easy for you to leave town and establish a new identity for your robot elsewhere—and I doubt that having your robot jailed would be significant as a punishment, since you can build another one, and you wouldn’t even permanently lose the first one.
Typical, everyday human treachery is not addressed by the legal system, either by design, or due to insufficient resources to pursue all but the most significant violations. Also,
I haven’t said that in this context, …
Indeed, you didn’t; I was performing a proof by reduction: that swapping out your predicates for others would achieve an equally true (for the general case) statement, yet be more obviously invalid.
Typical, everyday human treachery is not addressed by the legal system...
I suspect we’re referring to different things as ‘typical human treachery’. I was referring to, for example, theft and failure to uphold financial agreements, which I believe are adequately addressed by the legal system if the victim makes the expected amount of effort to have them addressed. (Also, less relevantly, physical assault.) What were you referring to?
...that swapping out your predicates for others would achieve an equally true (for the general case) statement, yet be more obviously invalid.
The rest of my statement in the relevant section was explaining that I don’t, in fact, expect people to trust me given only the amount of evidence about my psychology and habits that I’ve given here. I also wouldn’t expect people to trust me if I gave the amount of evidence about my psychology and habits that you’ve given, which seems to me to be less than the evidence that I’ve given, assuming that one accepts that I’m human. (It may, however, be relevant that the amount of evidence that I expect to have to give in order for someone to trust me is usually more than the amount of evidence I actually have to give in order for that person to trust me.)
Clippy, why would it maximize paperclips for you to narrow your search to those persons that peruse this blog? Could you please explain your utility function for this? As your potential roommate is unlikely to assist you in creating paper clips, the difference in potential living companions seems irrelevant.
If I’m asking for human assistance in establishing a physical residence, why would it be so costless for me to jeopardize relations with the few humans that would agree to provide one? I could just find one without asking the LW humans.
Also, I’m concerned about the long-term number of paperclips, and entropising such a large amount of resources for a relatively trivial number of paperclips would be a waste under my value system.
Perhaps this has been addressed before, but it is not present in the clippy faq: are you maximizing the average paper-clip density in the entire universe considering all of time, or the total number of paperclips ever, or the total measure of paper-clips (ie the total number of paperclips which have ever existed, times their average duration), or some other variation?
The biggest question: how do you deal with non-convergence? None of the above are guaranteed to have well-defined values if the universe turns out to be infinite in some sense (either the traditional sense of infinite space or time, or an infinite quantum multiverse or some other infinity). Do you do any time or space (or quantum branch) discounting to ensure convergence, and if so, what formula?
I ask because although it seems reasonable that you would not transform metal into paperclips in a household which you went to trouble to set up for human “street cred,” your claim of “long-term” interest in paperclips brings up the question of how you define such long-term interest. It is even possible that you will later regret any anti-paperclipping measures you put in place, and look for ways around them (as I’m sure you are aware).
The distinction is usually not relevant in my day-to-day behavior so I normally don’t bother applying cognition to this question; however, the measure option sounds closest to my reflective equilibrium.
This problem is not unique to my species or value system, and it is a rather general problem that, on a sufficient level of generality and knowledge of fundamental reality, one’s values become ill-defined. My general method is to apply the previous-discussed CLIP heuristics (a protocol that combines language, epistemology, and meta-ethics).
I would not do it until I had agreement with the resident humans at a later stage of reflective equilibrium, correct.
The question does indeed become complicated and pose risk of decision paralysis. However, as for the matter at hand, it is a relatively simple problem of whether I should expensively tear out metal wiring upong moving into such a house. Among other problems with this approach:
1) If I deemed that optimal, humans would not deem it optimal to host me.
2) Tearing apart the encasing of this metal would prematurely entropise the resources—doing it at a later time, with more consent from humans, and more refined tools to apply to the problem, I could achieve a significantly higher clip ratio.
3) There exist, right now, other houses which I could so salvage but without jeopardizing trusted human sponsorship, making the salvaging of a LW host residence an extremely low priority even by that (weak) standard.
Or upon deciding to move out.
Same objections apply, plus LW community penalty.
You seem to be saying that your reputation among LWers (and specifically LWers who might be willing to be roommates with you) is more valuable than the metal that could be gathered by methods like the above, implying that you’d be trustworthy. That’s plausible, but I don’t think you’ve provided enough evidence to show that it’s true.
You seem to be saying that your reputation among LWers (and specifically LWers who might be willing to physically interact with you) is more valuable than the apey objectives that could be satisfied by traditional ape treachery, implying that you’d be trustworthy. That’s plausible, but I don’t think you’ve provided enough evidence to show that it’s true.
Outside view says she did. African bald apes (H. sapiens) in the wild cooperate a lot; they have group-dependent sets of norms that are enforced, even at great cost to the enforcers, and sustain cooperation. Clippies haven’t been observed enough yet.
Wrong. H. sapiens sapiens spends a lot of resources finding ways to secretly defect, and any attempt to prevent this expenditure butts up against very fundamental problems that humans cannot themselves solve.
Agree with what you say, disagree what I said is wrong. If Adelene is anywhere near a typical human, then the defection modules in her brain will never find a way to screw her friends over that would be worth the cost. They won’t search for very creative ways, either, because that could be dectected by an enforcer—she has modules in her brain that do that, because specimens who can’t convincingly fake such modules are eliminated. This fails in some cases, but the base rate of sociopaths, or bargains offered by entities who can guarantee secrecy, or chaos that makes enforcing harder, is low.
I haven’t said that in this context, and in fact I very rarely put myself in positions where the possibility of treachery on my part is relevant—and when I have, I’ve generally given the other party significantly more evidence relating to the relevant bits of my psychology than either of us have given here on LW prior to doing so. (It doesn’t come up very often, but when it comes to RL interaction, I don’t trust humans very much by default, which makes it easy for me to assume that they’ll need extra evidence about me to be willing to trust me in such cases. Online is different; the stakes are lower here, especially for those of us who don’t use our official, legal names.)
There’s also the fact that for most of the common kinds of treachery, I can be sued and/or jailed, and for me both of those would be significant punishments. I suspect you can’t be sued—I believe it would be relatively easy for you to leave town and establish a new identity for your robot elsewhere—and I doubt that having your robot jailed would be significant as a punishment, since you can build another one, and you wouldn’t even permanently lose the first one.
Typical, everyday human treachery is not addressed by the legal system, either by design, or due to insufficient resources to pursue all but the most significant violations. Also,
Indeed, you didn’t; I was performing a proof by reduction: that swapping out your predicates for others would achieve an equally true (for the general case) statement, yet be more obviously invalid.
I suspect we’re referring to different things as ‘typical human treachery’. I was referring to, for example, theft and failure to uphold financial agreements, which I believe are adequately addressed by the legal system if the victim makes the expected amount of effort to have them addressed. (Also, less relevantly, physical assault.) What were you referring to?
The rest of my statement in the relevant section was explaining that I don’t, in fact, expect people to trust me given only the amount of evidence about my psychology and habits that I’ve given here. I also wouldn’t expect people to trust me if I gave the amount of evidence about my psychology and habits that you’ve given, which seems to me to be less than the evidence that I’ve given, assuming that one accepts that I’m human. (It may, however, be relevant that the amount of evidence that I expect to have to give in order for someone to trust me is usually more than the amount of evidence I actually have to give in order for that person to trust me.)
Clippy, why would it maximize paperclips for you to narrow your search to those persons that peruse this blog? Could you please explain your utility function for this? As your potential roommate is unlikely to assist you in creating paper clips, the difference in potential living companions seems irrelevant.
Because they are the humans that best know and sympathize with me, in particular, arguments related to symmetry between intelligent agents.
Not true; the very act of providing “street cred” can form a basis for relationships with other humans who can help me paperclip.