Suppose that it’s some point in the future, and we’re able to run conscious simulations of our ancestors. We’re considering whether or not to run such a simulations.
We are also curious about whether we are in a simulation ourselves, and we know that knowledge that civilizations like ours run ancestor simulations would be evidence for the proposition that we ourselves are in a simulation.
Could the choice at this point whether or not to run a simulation be used as a form of acausal control over the probability that we ourselves are living in a simulation?
English tip: the possessive ending ” ’s ” carries an implicit “the”. Thus “Greg Egan’s story” means “the story of Greg Egan”, not just “story of Greg Egan”. (This is unlike the corresponding construction in, for example, German.) Instead of the above, you wanted to write:
Wasn’t there a Greg Egan story about it?
(This particular mistake occurs often among non-native-speakers, and indeed is a dead giveaway of one’s status as such, so it’s worth saying something about.)
[Note to self: I should re-read the relevant chapter in my English grammar when I get back home. Meanwhile, I’ll look at the overview here.]
(Semantically, “ten minutes’ walk” still means ‘a ten-minute walk’ rather than ‘the ten-minute walk’, but your point in reply to shminux was about syntax not semantics anyway.)
(Semantically, “ten minutes’ walk” still means ‘a ten-minute walk’ rather than ‘the ten-minute walk’, but your point in reply to shminux was about syntax not semantics anyway.)
The “proof of synonymy” looks like this:
ten minutes’ walk = (the walk) of (ten minutes) = a (walk of ten minutes) = a ten-minute walk
...the second “equality” being where semantics is invoked.
The determiner position in an NP [noun phrase] is usually filled
by one of two kinds of expression.
In all the examples so far it has been a determinative [a word
like the, a, this, some, or three], and some of these can
be accompanied by their own modifiers, making a determinative
phrase, abbreviated DP.
In addition, the determiner may have the form of a genitive NP.
Examples, with the determiners underlined [bolded], are given
[below]:
DETERMINATIVE *the city **some* rotten eggs
DP *almost all politicians **very few* new books
GENITIVE NP *her income **the senator’s* young son
p. 109:
As a determiner, the genitive is always definite. Note, for example,
that [one patient’s father] corresponds to *the father of one
patient, not **a* father of one patient.
Note, for example, that [one patient’s father] corresponds to the father of one patient, not a father of one patient.
Hm. So how do you express the concept of an undetermined relative of some patient? The text you quoted would say that [one patient’s relative] means the relative of one patient—how do I express a relative of one patient?
Well, of course there are ways to rephrase most anything. I am, however, interested in whether there’s a way to express the “a relative of one patient” notion through the possessive ’s.
A related question is whether a native speaker would be sure that one patient’s relative necessarily means the relative, or he would be ambiguous whether it means the relative or a relative.
In a specialized context (such as among people who work at a hospital), “patient’s relative” could conceivably become a set phrase, in which case sentences such as “there are some patient’s relatives waiting outside” would become possible (contrast * “there are some Greg Egan’s stories on the shelf”).
This is presumably what happened with “girls’ school”. Very rarely, it can even happen with proper nouns, as in the mathematical term Green’s function. But this is not part of the syntax of the possessive ; it is the result of the whole possessive phrase being treated as a unit. (When you hear “the Green’s function for this operator” for the first time, you immediately know that “Green’s function” is a jargon phrase, because of the irregular syntax.)
(My comment was generated by the spontaneous reaction and reflection of a native speaker rather than memory of any deliberately learned rule.) Wikipedia has this to say:
In English and some other languages, the use of such a word implies the definite article. For example, my car implies the car that belongs to me/is used by me; it is not correct to precede possessives with an article (* the my car) or other definite determiner such as a demonstrative (* this my car)
One should indeed think of ” ’s ” in this context as the equivalent for nouns of what “my” is for the pronoun “I”.
Hmm, okay, to put it another way—if we avoid running ancestor simulations for the purpose of maximizing the probability that we are not in a simulation, is it valid to, based on this fact, increase our credence in not being in a simulation?
I think so. If we decided not to run a simulation, any would-be-simulators analogous to us would also choose not to run a simulation, so you’ve eliminated a bunch of worlds where simulations are possible.
That is true, but irrelevant. Making the decision eliminates possible worlds in which we are simulations. Therefore we end up with fewer simulation-worlds out of our total list of potential future worlds, and thus our probability estimate must increase.
Or, to put it in Bayesian terms: P(we’re in a simulation|we chose not to be in a simulation)/P(we choose not to be in a simulation) is greater than 1.
if we avoid running ancestor simulations for the purpose of maximizing the probability that we are not in a simulation, is it valid to, based on this fact, increase our credence in not being in a simulation?
Unless you round sufficiently small increases down to zero, which is what people generally do. If somebody asked me that, and I estimated that the difference in probability was .00000000001, then I would answer “no”.
That is granted. However, I’m also fairly sure (p=.75) that the probability isn’t that small, because by deciding not to simulate a civilization yourself, you have greatly decreased the probability of being in an infinite descending chain. There remains singleton chance simulations and dynamic equilibria of nested simulations, but those are both intuitively less dense in clones of your universe—so you’ve ruled out a significant fraction of possible simulation-worlds by deciding not to simulate yourself yourself.
you have greatly decreased the probability of being in an infinite descending chain.
No matter what there aren’t going to be any infinitely descending chains unless our understanding of the laws of physics is drastically wrong. You can’t simulate n+1 bits with n qubits. So, even if you assume a quantum simulation for a purely classical setting, you still have strict limits.
There remains singleton chance simulations and dynamic equilibria of nested simulations, but those are both intuitively less dense in clones of your universe
Imagine that some Clarktech version of ourselves dedicates an entire galaxy to simulating the Milky Way. Would we have noticed by now?
Neither does the simulation need to be perfect: it only needs to be perfect wherever we actually look. This makes for a much more complex program, but might save on computing costs.
Anyway, yeah, you probably won’t get an infinite chain, but you’ll get a very long one, which leads to my second point:
A “singleton chance simulation” just means that someone randomly decided to simulate our universe in particular. This is rather unlikely.
A “dynamic equilibria of nested simulation” just means that Universe A simulates Universe B simulates Universe C which simulates Universe A, creating a descending chain that is not as dense as an immediate recursion, A->A->A.
Both these cases will contribute less possible universes than a (near-)infinite descending chain, so by eliminating the descending chain you’ve greatly decreased the probability of being in a simulation.
No. It is unreasonable to think that all simulations are ancestral anyway. Even if no one runs ancestral simulations people will still run simulations of other possible words for a variety of reasons and we will be likely in one of those. And anyway, as soon as you can make a complete ancestral simulation (without knowing of any way to do so without giving consciousnesses/qualia/whatever to the simulated) you can be >99% that you live in a simulation no matter if you run anything yourself or not.
That is less insulting, and therefore an improvement. A version that’s not even a little insulting might look something like “Not all simulations are ancestral.” That approach expresses disagreement with the original claim, but doesn’t connote anything about the person who made it.
A version that’s not even a little insulting might look something like “Not all simulations are ancestral.”
There’s a difference between “it is unreasonable to think X” and “not X”. (Let X equal “the sixteenth decimal digit of the fine structure constant is 3”, for example.)
(I’d use “There’s no obvious good reason to think that all simulations are ancestral.”)
It is unreasonable to think that all simulations are ancestral anyway.
Point taken regarding ancestor simulations, but I don’t think that resolves the question. What we choose to do is still evidence about what others will choose to do whether or not the choice is about simulating ancestors or just other possible worlds.
as soon as you can make a complete ancestral simulation … you can be >99% that you live in a simulation
In Bostrom’s formulation there is also the possibility that civilizations capable of ancestor simulations will overwhelmingly choose not to. It’s not obvious to me that this is one of the horns of the trilemma to reject.
I can think of at least two reasons why it might be a convergent behavior not to run ancestor simulations:
1) Civilizations capable of running ancestor simulations might overwhelmingly have morals that dissuade them from subjecting sentient beings to such low standards of living as their ancestors had.
2) Such civilizations may wish to exert acausal control over whether they are in a simulation. This is the motivation for my question.
In Bostrom’s formulation there is also the possibility that civilizations capable of ancestor simulations will overwhelmingly choose not to. It’s not obvious to me that this is one of the horns of the trilemma to reject.
Again, you are making Bostrom’s mistake of focusing on ancestral simulations. This is likely why this option seems plausible to you like it did to him—it looks much more plausible that people will decide not to run any ancestral simulations because of their morals than it is that people will decide not to run any simulations whatsoever.
1) Civilizations capable of running ancestor simulations might overwhelmingly have morals that dissuade them from subjecting sentient beings to such low standards of living as their ancestors had.
This is theoretically possible but realistically there is little reason to expect all posthuman civilizations to have such morals in regards to arbitrary creatures. We certainly don’t seem to be the type of civilization which would sacrifice the utility gained by running simulations for some questionable moral reasons—or at least not with a probability that is close to 1. Additionally, The mindspace for all posthuman agents is huge—you need a large amount of evidence to conclude that it is likely for all posthuman civilizations to be so moral.
Such civilizations may wish to exert acausal control over whether they are in a simulation. This is the motivation for my question.
Similarly, mind space is huge and it seems really unlikely by default that most posthuman societies will never run a simulation just on that basis. Furthermore, it is enough if only 1 for every billion posthuman civilizations runs simulations for it to be more likely that we are in a simulation than not, provided that the average simulator civilization runs more than a billion simulation in it’s history.
Furthermore, in order for most posthuman civilizations to not run any simulations there needs to be some sort of a 100% efficent way to prevent rogue agents to develop simulations. This also could be possible but still mostly unlikely. Even if somehow all posthuman societies always decide to never run a single simulation (for which there is no evidence) it is unlikely that all those civilizations also have a world-wide simulation-prevention mechanism in place from the very moment when simulations are technologically possible in that world.
you are making Bostrom’s mistake of focusing on ancestral simulations
Again, this seems irrelevant. I talked about ancestor simulations because that’s how it’s worded in the Simulation Argument, but as I said in the post above, as far as I can tell the logic doesn’t depend on it. Just replace ‘simulations of ancestors’ with ‘simulations of worlds containing sentient beings’.
As for the rest of your post, those are fine arguments for why the second horn of the trilemma should be rejected. I don’t find them absolutely convincing, so I still assign non-negligible credence to option 2 (and thus still find the acausal control question interesting), but I don’t have strong counterarguments either, so if you do assign negligible credence to option 2, perhaps we’ll have to agree to disagree on this point.
And anyway, as soon as you can make a complete ancestral simulation (without knowing of any way to do so without giving consciousnesses/qualia/whatever to the simulated) you can be >99% that you live in a simulation no matter if you run anything yourself or not.
Do inaccurate ancestral simulations count for anything in this argument? Admittedly, I’m extrapolating from humans as I know them, but the combination of incomplete research, simulations modified for convenience and/or tolerability and/or to improve the story, and interest in what-if scenarios implies that even if you’re a ancestor of an ancestor simulation creating civilization, you won’t be that much like the actual ancestor.
It completely doesn’t matter whether you are a simulation of an accurate ancestor, inaccurate ancestor or HJPEV. As I am trying to point out there is nothing special to ancestral simulations and no real reason to focus only on them.
I have a question about the Simulation Argument.
Suppose that it’s some point in the future, and we’re able to run conscious simulations of our ancestors. We’re considering whether or not to run such a simulations.
We are also curious about whether we are in a simulation ourselves, and we know that knowledge that civilizations like ours run ancestor simulations would be evidence for the proposition that we ourselves are in a simulation.
Could the choice at this point whether or not to run a simulation be used as a form of acausal control over the probability that we ourselves are living in a simulation?
The most you can say is that all reflectively consistent ancestors would behave the same way you do. Wasn’t there a Greg Egan’s story about it?
English tip: the possessive ending ” ’s ” carries an implicit “the”. Thus “Greg Egan’s story” means “the story of Greg Egan”, not just “story of Greg Egan”. (This is unlike the corresponding construction in, for example, German.) Instead of the above, you wanted to write:
(This particular mistake occurs often among non-native-speakers, and indeed is a dead giveaway of one’s status as such, so it’s worth saying something about.)
(Except in constructs like “girls’ school” or “a ten minutes’ walk”.)
You’re right about “girls’ school”, but “a ten minutes’ walk” is wrong (should be “a ten-minute walk” or “ten minutes’ walk”).
Thanks. I myself am a non-native speaker.
[Note to self: I should re-read the relevant chapter in my English grammar when I get back home. Meanwhile, I’ll look at the overview here.]
(Semantically, “ten minutes’ walk” still means ‘a ten-minute walk’ rather than ‘the ten-minute walk’, but your point in reply to shminux was about syntax not semantics anyway.)
The “proof of synonymy” looks like this:
ten minutes’ walk = (the walk) of (ten minutes) = a (walk of ten minutes) = a ten-minute walk
...the second “equality” being where semantics is invoked.
Thanks. This sounds plausible (if irrelevant), but I could not find an authoritative reference confirming it. Any links?
A Student’s Introduction to English Grammar, p. 90:
p. 109:
See also the Wikipedia determiner and genitive case articles.
Thanks! Now, if only someone linked that Egan story :)
“Crystal Nights”
This story is not by Egan, but it might be what you’re looking for.
Ah, yes, thanks. I wondered why I couldn’t find it :) Hmm, I thought it was longer...
Hm. So how do you express the concept of an undetermined relative of some patient? The text you quoted would say that [one patient’s relative] means the relative of one patient—how do I express a relative of one patient?
Didn’t you just?
Well, of course there are ways to rephrase most anything. I am, however, interested in whether there’s a way to express the “a relative of one patient” notion through the possessive ’s.
A related question is whether a native speaker would be sure that one patient’s relative necessarily means the relative, or he would be ambiguous whether it means the relative or a relative.
In a specialized context (such as among people who work at a hospital), “patient’s relative” could conceivably become a set phrase, in which case sentences such as “there are some patient’s relatives waiting outside” would become possible (contrast * “there are some Greg Egan’s stories on the shelf”).
This is presumably what happened with “girls’ school”. Very rarely, it can even happen with proper nouns, as in the mathematical term Green’s function. But this is not part of the syntax of the possessive ; it is the result of the whole possessive phrase being treated as a unit. (When you hear “the Green’s function for this operator” for the first time, you immediately know that “Green’s function” is a jargon phrase, because of the irregular syntax.)
(My comment was generated by the spontaneous reaction and reflection of a native speaker rather than memory of any deliberately learned rule.) Wikipedia has this to say:
One should indeed think of ” ’s ” in this context as the equivalent for nouns of what “my” is for the pronoun “I”.
Haven’t read it, but perhaps you mean this one? It sounds very interesting!
Taboo “acausal control.”
Hmm, okay, to put it another way—if we avoid running ancestor simulations for the purpose of maximizing the probability that we are not in a simulation, is it valid to, based on this fact, increase our credence in not being in a simulation?
I think so. If we decided not to run a simulation, any would-be-simulators analogous to us would also choose not to run a simulation, so you’ve eliminated a bunch of worlds where simulations are possible.
Only if those simulators are extremely similar to us. It may only take a very minor difference to decide to run simulations.
That is true, but irrelevant. Making the decision eliminates possible worlds in which we are simulations. Therefore we end up with fewer simulation-worlds out of our total list of potential future worlds, and thus our probability estimate must increase.
Or, to put it in Bayesian terms: P(we’re in a simulation|we chose not to be in a simulation)/P(we choose not to be in a simulation) is greater than 1.
Sure, but by how much? If the ratio is something like 2 or even 5 or 10 this isn’t going to matter much.
That’s not the question.
That’s the question, and the answer is “yes.”
Unless you round sufficiently small increases down to zero, which is what people generally do. If somebody asked me that, and I estimated that the difference in probability was .00000000001, then I would answer “no”.
That is granted. However, I’m also fairly sure (p=.75) that the probability isn’t that small, because by deciding not to simulate a civilization yourself, you have greatly decreased the probability of being in an infinite descending chain. There remains singleton chance simulations and dynamic equilibria of nested simulations, but those are both intuitively less dense in clones of your universe—so you’ve ruled out a significant fraction of possible simulation-worlds by deciding not to simulate yourself yourself.
No matter what there aren’t going to be any infinitely descending chains unless our understanding of the laws of physics is drastically wrong. You can’t simulate n+1 bits with n qubits. So, even if you assume a quantum simulation for a purely classical setting, you still have strict limits.
I’m not sure what you mean here. Can you expand?
Imagine that some Clarktech version of ourselves dedicates an entire galaxy to simulating the Milky Way. Would we have noticed by now?
Neither does the simulation need to be perfect: it only needs to be perfect wherever we actually look. This makes for a much more complex program, but might save on computing costs.
Anyway, yeah, you probably won’t get an infinite chain, but you’ll get a very long one, which leads to my second point:
A “singleton chance simulation” just means that someone randomly decided to simulate our universe in particular. This is rather unlikely.
A “dynamic equilibria of nested simulation” just means that Universe A simulates Universe B simulates Universe C which simulates Universe A, creating a descending chain that is not as dense as an immediate recursion, A->A->A.
Both these cases will contribute less possible universes than a (near-)infinite descending chain, so by eliminating the descending chain you’ve greatly decreased the probability of being in a simulation.
No. It is unreasonable to think that all simulations are ancestral anyway. Even if no one runs ancestral simulations people will still run simulations of other possible words for a variety of reasons and we will be likely in one of those. And anyway, as soon as you can make a complete ancestral simulation (without knowing of any way to do so without giving consciousnesses/qualia/whatever to the simulated) you can be >99% that you live in a simulation no matter if you run anything yourself or not.
I strongly recommend not using stupid. It’s less distracting to just point out mistakes without using insults.
changed to unreasonable if that helps
That is less insulting, and therefore an improvement. A version that’s not even a little insulting might look something like “Not all simulations are ancestral.” That approach expresses disagreement with the original claim, but doesn’t connote anything about the person who made it.
However, your version completely skips what I am actually saying—that I think that whole line of thinking is bad.
There’s a difference between “it is unreasonable to think X” and “not X”. (Let X equal “the sixteenth decimal digit of the fine structure constant is 3”, for example.)
(I’d use “There’s no obvious good reason to think that all simulations are ancestral.”)
“Unreasonable” is an improvement, but I’d take it further to “mistaken” or “highly implausible”.
Actually, I agree with you about the likelihood of numerous sorts of simulations that highly outnumber ancestor simulations.
Point taken regarding ancestor simulations, but I don’t think that resolves the question. What we choose to do is still evidence about what others will choose to do whether or not the choice is about simulating ancestors or just other possible worlds.
In Bostrom’s formulation there is also the possibility that civilizations capable of ancestor simulations will overwhelmingly choose not to. It’s not obvious to me that this is one of the horns of the trilemma to reject.
I can think of at least two reasons why it might be a convergent behavior not to run ancestor simulations:
1) Civilizations capable of running ancestor simulations might overwhelmingly have morals that dissuade them from subjecting sentient beings to such low standards of living as their ancestors had.
2) Such civilizations may wish to exert acausal control over whether they are in a simulation. This is the motivation for my question.
Again, you are making Bostrom’s mistake of focusing on ancestral simulations. This is likely why this option seems plausible to you like it did to him—it looks much more plausible that people will decide not to run any ancestral simulations because of their morals than it is that people will decide not to run any simulations whatsoever.
This is theoretically possible but realistically there is little reason to expect all posthuman civilizations to have such morals in regards to arbitrary creatures. We certainly don’t seem to be the type of civilization which would sacrifice the utility gained by running simulations for some questionable moral reasons—or at least not with a probability that is close to 1. Additionally, The mindspace for all posthuman agents is huge—you need a large amount of evidence to conclude that it is likely for all posthuman civilizations to be so moral.
Similarly, mind space is huge and it seems really unlikely by default that most posthuman societies will never run a simulation just on that basis. Furthermore, it is enough if only 1 for every billion posthuman civilizations runs simulations for it to be more likely that we are in a simulation than not, provided that the average simulator civilization runs more than a billion simulation in it’s history.
Furthermore, in order for most posthuman civilizations to not run any simulations there needs to be some sort of a 100% efficent way to prevent rogue agents to develop simulations. This also could be possible but still mostly unlikely. Even if somehow all posthuman societies always decide to never run a single simulation (for which there is no evidence) it is unlikely that all those civilizations also have a world-wide simulation-prevention mechanism in place from the very moment when simulations are technologically possible in that world.
Again, this seems irrelevant. I talked about ancestor simulations because that’s how it’s worded in the Simulation Argument, but as I said in the post above, as far as I can tell the logic doesn’t depend on it. Just replace ‘simulations of ancestors’ with ‘simulations of worlds containing sentient beings’.
As for the rest of your post, those are fine arguments for why the second horn of the trilemma should be rejected. I don’t find them absolutely convincing, so I still assign non-negligible credence to option 2 (and thus still find the acausal control question interesting), but I don’t have strong counterarguments either, so if you do assign negligible credence to option 2, perhaps we’ll have to agree to disagree on this point.
I do and based on the wording of your comment you have no real reason not to either.
Did you miss this part?
Nope. They weren’t meant to be absolutely convincing—option 2) is possible just not probable.
Perhaps. I will have to think about it some more.
Do inaccurate ancestral simulations count for anything in this argument? Admittedly, I’m extrapolating from humans as I know them, but the combination of incomplete research, simulations modified for convenience and/or tolerability and/or to improve the story, and interest in what-if scenarios implies that even if you’re a ancestor of an ancestor simulation creating civilization, you won’t be that much like the actual ancestor.
Just for the fun of it, the Borgias on tv.
It completely doesn’t matter whether you are a simulation of an accurate ancestor, inaccurate ancestor or HJPEV. As I am trying to point out there is nothing special to ancestral simulations and no real reason to focus only on them.