The simulation argument was famously elaborated by Nick Bostrom. The idea is that our descendants may be able to create simulated creatures like you, and put them in simulated environments that look like the one you now find yourself in. If so, you can’t be sure that you are not now one of these future simulated people. The chance that you should assign to this possibility depends on the number of such future creatures, relative to the number of real creatures like you today.
Interesting argument. More generally I believe that this kind of reasoning on probability distributions of simulations can put strong constraints on the simulation hyphotesis.
My problem with it is that it's ignoring other factors which make simulations more likely to be run (a.k.a Simplicity Assumption), in particular the computational complexity of the simulations. If a simulation is 1000x easier to simulate, just few simulators are enough to quickly make easier simulations dominant. Imagining that nobody in the future will take a shot at simulating very easy sims is unlikely, similarly to how Fermi's paradox cannot be solved with "ALL possible aliens will do X".
I might have missed something, but it seems to me that the above merely looks at the fraction of past sims at a singular point in time in the future. Shouldn't we rather integrate over the entire future to find out whether most sims of, say, humans in the year 2021 are found in 2021 or in later simulations (in the rest of the history of civilization)?
In other words, I assume, given our evidence, that the question is: conditional on seeming to be a 2021_human, what is the probability of you being a sim or an actual 2021_human? Or equivalently: what is the fraction of actual vs. simulated such humans *over all time*?Not: conditional on being an em in some future point in time, what is the probability of being a 2021_human?
I'd expect "recency bias" to fade as well, which might make this argument substantially weaker. The fact that more people were interested in the science of, for example, gravitational waves recently (due to current events) than GR itself, is probably a social artefact that will fade.
Recency bias seems like an attractor ("talk about what everyone's talking about") that will be coordinated away from. Better individual targeting for media/education could make that happen. So could better surfacing and aggregation mechanisms that put a calibrated emphasis on newness.
Of course it isn't clear that our time will survive interest, in whatever competing pressures remain.
Extrapolating the current technology level out billions of years without taking account of the historical rise of simulation technology does not seem very reasonable to me. Most world simulations are performed inside minds. Minds have got better and better at considering more possibilities that never physically exist, but are instead future projections, or forms of counterfactual historical analysis. The capabilities of minds seem likely to grow in the future, so that most position evaluations are of virtual situations that never happened physically. Go programs already consider millions of virtual positions for every move they make. In the future, for the dominant creatures, the virtual will likely outnumber the real by billions to one.
I mentioned predicting the nature of aliens as an application of simulism in my "Intelligent design" essay - from 2016. It is a subcategory of ancestor simulations - which is a pretty mainstream idea. I wrote: "Simulating major evolutionary transitions might be of interest to future intelligent agents because it might throw light on what aliens they might encounter in the future. That could be an event of critical importance to them." I haven't received much critical feedback about the idea. However, I don't regard the idea that our observations are "too detailed" as a particularly good criticism. We don't know how detailed the parent universe is - what we see could be a very crude simulation of it. Nor do we know that any more than one person's experience is the subject of the simulation. All I really know is that the simulation includes Tim Tyler. The rest of you folk could all be NPCs. That would cut out a lot of apparent detail. A lot of the remaining detail could be provided by illusions and fake memories. The actual evidence for large quantities of unnecessary detail seems pretty weak overall. We don't know very much about the level of detail of the simulation, or how it compares to that of the parent universe.
I have a different version of Shulman's argument: we are in a category of simulation designed to explore the rise of artifactual intelligence, and that this would be one of the most frequently-run kind of simulation in a universe dominated (as ours will likely be) by artifactual intelligence.
But for people who insist the simulation arguments has to be about simulating humans, there is another argument why this era is special: https://www.overcomingbias....
First, your argument is quite different from Bostrom's, and so deserves a different name. Have you written it up somewhere?
Second, even if the % of effort on history is non-trivial, little of that is spread across historical doubles evenly, and most of that effort doesn't go into making detailed sims. We need that % spent on detailed sims of even doubles, and then spread that over all the doubles expected ever, which is quite a lot.
Third, I'm pretty skeptical that many detailed sims of entire histories of random possible civs is a sensible way to prepare to meet aliens.
We might be, but the simulation argument doesn't apply to those. Without that argument, we'd just look at the % of observer moments that are in sims, which is small, and conclude we are probably not in one.
In our own society it is far above 1/billion, well above 1/million (consider just history teacher, students, books that do broad surveys.
One example field for future civs to do that would be predicting the nature of aliens they might meet. which has major call for modeling biological evolution, early biosapient cultural developments, and later post-human machine developments, each contributing important variance (even moreso if old civillizations stabilize because it is technologically possible and necessary to avoid self-destruction/successfully colonize). For an intergalactic civilization that project alone is worth spending millions of stars, enough to simulate trillions of histories like ours from abiogenesis through interstellar colonization, with extra simulation of times and things where variance has more effect on the long-run future.
Simpler physics is an example of a 'computational shortcut' which means either the world would either be simpler (so we are not them) or its basically Boltzmann brain
What would be your estimate of the % of observer moments devoted to sims that are allocated equally in proportion to historical doubling periods? This is surely vastly less than allocated for other reasons, and there's some threshold it needs to be above to give us a significant chance of living in a sim now.
So a mature society that puts at least some tiny but not-vanishing share of attention to economic history of each of those doublings has to put more than 1% on these past few centuries. Increasing galactic populations a trillion trillion times will not lead to economic growth histories giving less than a trillion trillionth of their coverage to more than 1% of all the growth (and much more if there's a growth explosion in the next few centuries with influence from preexisting humanity, like you predict in Age of Em).
For science the share has to be larger given how far things like relativity and the Standard Model have already reached (and how little new has been revealed by vast increases in discovery efforts since). A future physics textbook in a billion years will still need to describe protons and neutrons, and if that gets significant coverage, then history of science has to give coverage to their discoverers. Increasing intergalactic population by a trillion trillion times will not change the page count of such coverage in a survey textbook from 5% to vastly less than 1 character, as your parameters (and assumption of fixed decay rate with growth, extrapolated to extremes) predict.
Because if you take the extremely exciting things that humans engage in and extrapolate future performance gains. War isn't going to be any fun to play at when it's mass execution at the push of a button. Flying a fighter plane probably peaked in WW2 as far as level of excitement as a factor of vehicle performance. I agree that I am falling into some logical traps with my idea. The idea that one is living at the most significant time is hard to escape. As we become longer lived, we will become more risk averse also. The fun is now.
What about "base reality has different physics"? We simulate a lot of "worlds" now in video games and most of them don't have exactly the same physics as ours.
Interesting argument. More generally I believe that this kind of reasoning on probability distributions of simulations can put strong constraints on the simulation hyphotesis.
My problem with it is that it's ignoring other factors which make simulations more likely to be run (a.k.a Simplicity Assumption), in particular the computational complexity of the simulations. If a simulation is 1000x easier to simulate, just few simulators are enough to quickly make easier simulations dominant. Imagining that nobody in the future will take a shot at simulating very easy sims is unlikely, similarly to how Fermi's paradox cannot be solved with "ALL possible aliens will do X".
For more info on this line of reasoning check https://philpapers.org/rec/...
My math above does integrate over the entire future.
I might have missed something, but it seems to me that the above merely looks at the fraction of past sims at a singular point in time in the future. Shouldn't we rather integrate over the entire future to find out whether most sims of, say, humans in the year 2021 are found in 2021 or in later simulations (in the rest of the history of civilization)?
In other words, I assume, given our evidence, that the question is: conditional on seeming to be a 2021_human, what is the probability of you being a sim or an actual 2021_human? Or equivalently: what is the fraction of actual vs. simulated such humans *over all time*?Not: conditional on being an em in some future point in time, what is the probability of being a 2021_human?
I'd expect "recency bias" to fade as well, which might make this argument substantially weaker. The fact that more people were interested in the science of, for example, gravitational waves recently (due to current events) than GR itself, is probably a social artefact that will fade.
Recency bias seems like an attractor ("talk about what everyone's talking about") that will be coordinated away from. Better individual targeting for media/education could make that happen. So could better surfacing and aggregation mechanisms that put a calibrated emphasis on newness.
Of course it isn't clear that our time will survive interest, in whatever competing pressures remain.
Extrapolating the current technology level out billions of years without taking account of the historical rise of simulation technology does not seem very reasonable to me. Most world simulations are performed inside minds. Minds have got better and better at considering more possibilities that never physically exist, but are instead future projections, or forms of counterfactual historical analysis. The capabilities of minds seem likely to grow in the future, so that most position evaluations are of virtual situations that never happened physically. Go programs already consider millions of virtual positions for every move they make. In the future, for the dominant creatures, the virtual will likely outnumber the real by billions to one.
I mentioned predicting the nature of aliens as an application of simulism in my "Intelligent design" essay - from 2016. It is a subcategory of ancestor simulations - which is a pretty mainstream idea. I wrote: "Simulating major evolutionary transitions might be of interest to future intelligent agents because it might throw light on what aliens they might encounter in the future. That could be an event of critical importance to them." I haven't received much critical feedback about the idea. However, I don't regard the idea that our observations are "too detailed" as a particularly good criticism. We don't know how detailed the parent universe is - what we see could be a very crude simulation of it. Nor do we know that any more than one person's experience is the subject of the simulation. All I really know is that the simulation includes Tim Tyler. The rest of you folk could all be NPCs. That would cut out a lot of apparent detail. A lot of the remaining detail could be provided by illusions and fake memories. The actual evidence for large quantities of unnecessary detail seems pretty weak overall. We don't know very much about the level of detail of the simulation, or how it compares to that of the parent universe.
I have a different version of Shulman's argument: we are in a category of simulation designed to explore the rise of artifactual intelligence, and that this would be one of the most frequently-run kind of simulation in a universe dominated (as ours will likely be) by artifactual intelligence.
https://blog.knowinghumans....
But for people who insist the simulation arguments has to be about simulating humans, there is another argument why this era is special: https://www.overcomingbias....
First, your argument is quite different from Bostrom's, and so deserves a different name. Have you written it up somewhere?
Second, even if the % of effort on history is non-trivial, little of that is spread across historical doubles evenly, and most of that effort doesn't go into making detailed sims. We need that % spent on detailed sims of even doubles, and then spread that over all the doubles expected ever, which is quite a lot.
Third, I'm pretty skeptical that many detailed sims of entire histories of random possible civs is a sensible way to prepare to meet aliens.
We might be, but the simulation argument doesn't apply to those. Without that argument, we'd just look at the % of observer moments that are in sims, which is small, and conclude we are probably not in one.
This assumes all simulations are historical. What if sims are the equivalent of video games in which we are the NPCs?
In our own society it is far above 1/billion, well above 1/million (consider just history teacher, students, books that do broad surveys.
One example field for future civs to do that would be predicting the nature of aliens they might meet. which has major call for modeling biological evolution, early biosapient cultural developments, and later post-human machine developments, each contributing important variance (even moreso if old civillizations stabilize because it is technologically possible and necessary to avoid self-destruction/successfully colonize). For an intergalactic civilization that project alone is worth spending millions of stars, enough to simulate trillions of histories like ours from abiogenesis through interstellar colonization, with extra simulation of times and things where variance has more effect on the long-run future.
Simpler physics is an example of a 'computational shortcut' which means either the world would either be simpler (so we are not them) or its basically Boltzmann brain
What would be your estimate of the % of observer moments devoted to sims that are allocated equally in proportion to historical doubling periods? This is surely vastly less than allocated for other reasons, and there's some threshold it needs to be above to give us a significant chance of living in a sim now.
Your own argument for modest number of total cumulative economic doublings (on pain of absurd economic value per atom):
https://www.overcomingbias....
So a mature society that puts at least some tiny but not-vanishing share of attention to economic history of each of those doublings has to put more than 1% on these past few centuries. Increasing galactic populations a trillion trillion times will not lead to economic growth histories giving less than a trillion trillionth of their coverage to more than 1% of all the growth (and much more if there's a growth explosion in the next few centuries with influence from preexisting humanity, like you predict in Age of Em).
For science the share has to be larger given how far things like relativity and the Standard Model have already reached (and how little new has been revealed by vast increases in discovery efforts since). A future physics textbook in a billion years will still need to describe protons and neutrons, and if that gets significant coverage, then history of science has to give coverage to their discoverers. Increasing intergalactic population by a trillion trillion times will not change the page count of such coverage in a survey textbook from 5% to vastly less than 1 character, as your parameters (and assumption of fixed decay rate with growth, extrapolated to extremes) predict.
Because if you take the extremely exciting things that humans engage in and extrapolate future performance gains. War isn't going to be any fun to play at when it's mass execution at the push of a button. Flying a fighter plane probably peaked in WW2 as far as level of excitement as a factor of vehicle performance. I agree that I am falling into some logical traps with my idea. The idea that one is living at the most significant time is hard to escape. As we become longer lived, we will become more risk averse also. The fun is now.
What about "base reality has different physics"? We simulate a lot of "worlds" now in video games and most of them don't have exactly the same physics as ours.