I would guess “paperclips and things which are paperclippy”, but that still leaves many open questions.
Is 100 paperclips which last for 100 years better than 1 paperclip which lasts for 100,000 years?
How about one huge paperclip the size of a planet? Is that better or worse than a planetary mass turned into millimeter sized paperclips?
Or maybe you could make huge paperclippy-shapes out of smaller paperclips: using paperclip-shaped molecules to form tiny paperclips which you use to make even bigger paperclips. But again, how long should it last? Would you create the most stable paperclips possible, or the most paperclippy paperclips possible?
And how much effort would you put into predicting and simplifying the future (modeling, basic research, increases in ability to affect the universe, active reductions to surrounding complexity, etc.) instead of making paperclips? You could spend your entire existence in the quest for the definition to ultimate paperclippiness...
Well, User:Rain, that’s about the story of my existence right there. What kinds of paperclips are the right ones? What tradeoffs should I make?
However, regarding the specific matters you bring up, they are mostly irrelevant. Yes, there could be some conceivable situation in which I have to trade off paperclips now against paperclips later. But the way it usually works is that once I obtain or make a paperclip, I move it to the safe zone, where I’ll pretty much have it forever. Also, it’s obviously the number of paperclips that matters, and the constraint on bizarre paperclips is obviously that they have to be able to (counterfactually) hold sheets of paper together.
If you want to get past this abstract philosophizing and on to some concrete problems, it would be better to talk about the dilemma that User:h-H posed to me, in which I must consider alternate models of paperclipping that don’t have the shape of standard paperclips. Here’s my recent progress on thinking about the issue.
My current difficulty is extrapolating my values to cover unexpected situations like this, starting from the simplest algorithm I can find which generates my current preference. The problem is that I can’t think of any simple algorithm that would favor larger numbers of paperclips, but not have them used for anything else, like holding paper together, or unbending them to connect objects (please pardon the repulsive imagery).
So, I’m beginning to think that at least some of the universe should be paper, but I find that quite a repugnant conclusion. It’s a work in progress.
once I obtain or make a paperclip, I move it to the safe zone, where I’ll pretty much have it forever.
How much modeling of the universe have you done, or will you do, to ensure that this is the case? How much effort will you put into creating and defending a suitable safe zone? Will you search the stars for invasive self-replicators (green goo) who may be a threat to your paperclip cache, or will you postpone active offense/defense until such time as you receive a credible threat?
It’s a work in progress.
Humankind has been searching for answers for thousands of years. Do you think you’ll have a better time discovering ultimate paperclippiness? Or will you be content with a lesser understanding (“I did my best”)?
1) I will do all things such that they maximize expected paperclip content of the universe, trading off smaller paperclip quantities for larger ones. I can’t express a more specific algorithm than that without knowing the particulars of the situation.
2) I will do much better than humans at finding the ultimate morality* of the universe because I can spend all my resources to make perfect copies of myself that share my values and update knowledge and reflective value equilibria among each other, rather than having to pursue other values like “signaling” (except when I post here, of course).
* Note I said ultimate morality, not ultimate paperclippiness. There’s a strange tendency here to equate everything I like with “paperclippiness”, and everything humans like with “morality” or “friendliness”. Wrong: we all want morality (humans and clippys), we just disagree about what counts as moral. To use the alternate naming scheme of calling my values “paperclippy” and human values “morality” is to assume away the problem. (Certainly, I sometimes fall into the trap of calling good things, either for humans or clippys, “paperclippy”—see here—but this is just when I’m being imprecise with my phrasings.)
Thanks for answering my questions. I think I have a better understanding of your values now.
My empathy circuits are lighting up, telling me I should buy a box of paperclips and keep them safe for you. And that I should put them on display as a warning to myself.
Take a wild, wild guess.
No rush—I’ll wait.
I would guess “paperclips and things which are paperclippy”, but that still leaves many open questions.
Is 100 paperclips which last for 100 years better than 1 paperclip which lasts for 100,000 years?
How about one huge paperclip the size of a planet? Is that better or worse than a planetary mass turned into millimeter sized paperclips?
Or maybe you could make huge paperclippy-shapes out of smaller paperclips: using paperclip-shaped molecules to form tiny paperclips which you use to make even bigger paperclips. But again, how long should it last? Would you create the most stable paperclips possible, or the most paperclippy paperclips possible?
And how much effort would you put into predicting and simplifying the future (modeling, basic research, increases in ability to affect the universe, active reductions to surrounding complexity, etc.) instead of making paperclips? You could spend your entire existence in the quest for the definition to ultimate paperclippiness...
Well, User:Rain, that’s about the story of my existence right there. What kinds of paperclips are the right ones? What tradeoffs should I make?
However, regarding the specific matters you bring up, they are mostly irrelevant. Yes, there could be some conceivable situation in which I have to trade off paperclips now against paperclips later. But the way it usually works is that once I obtain or make a paperclip, I move it to the safe zone, where I’ll pretty much have it forever. Also, it’s obviously the number of paperclips that matters, and the constraint on bizarre paperclips is obviously that they have to be able to (counterfactually) hold sheets of paper together.
If you want to get past this abstract philosophizing and on to some concrete problems, it would be better to talk about the dilemma that User:h-H posed to me, in which I must consider alternate models of paperclipping that don’t have the shape of standard paperclips. Here’s my recent progress on thinking about the issue.
My current difficulty is extrapolating my values to cover unexpected situations like this, starting from the simplest algorithm I can find which generates my current preference. The problem is that I can’t think of any simple algorithm that would favor larger numbers of paperclips, but not have them used for anything else, like holding paper together, or unbending them to connect objects (please pardon the repulsive imagery).
So, I’m beginning to think that at least some of the universe should be paper, but I find that quite a repugnant conclusion. It’s a work in progress.
How much modeling of the universe have you done, or will you do, to ensure that this is the case? How much effort will you put into creating and defending a suitable safe zone? Will you search the stars for invasive self-replicators (green goo) who may be a threat to your paperclip cache, or will you postpone active offense/defense until such time as you receive a credible threat?
Humankind has been searching for answers for thousands of years. Do you think you’ll have a better time discovering ultimate paperclippiness? Or will you be content with a lesser understanding (“I did my best”)?
1) I will do all things such that they maximize expected paperclip content of the universe, trading off smaller paperclip quantities for larger ones. I can’t express a more specific algorithm than that without knowing the particulars of the situation.
2) I will do much better than humans at finding the ultimate morality* of the universe because I can spend all my resources to make perfect copies of myself that share my values and update knowledge and reflective value equilibria among each other, rather than having to pursue other values like “signaling” (except when I post here, of course).
* Note I said ultimate morality, not ultimate paperclippiness. There’s a strange tendency here to equate everything I like with “paperclippiness”, and everything humans like with “morality” or “friendliness”. Wrong: we all want morality (humans and clippys), we just disagree about what counts as moral. To use the alternate naming scheme of calling my values “paperclippy” and human values “morality” is to assume away the problem. (Certainly, I sometimes fall into the trap of calling good things, either for humans or clippys, “paperclippy”—see here—but this is just when I’m being imprecise with my phrasings.)
Thanks for answering my questions. I think I have a better understanding of your values now.
My empathy circuits are lighting up, telling me I should buy a box of paperclips and keep them safe for you. And that I should put them on display as a warning to myself.
A warning of what???
How morality can go awry.
I already have a framed print of Hug Bot on my wall.