Nick Bostrom - The Simulation Argument (Full)

well there is this article that I published back in 2003 presenting the simulation argument this is an argument that tries to show that at least one of three propositions is true although it doesn't tell us which of these three the three propositions in question is first that almost all civilizations at our stage of technological development go extinct before they reach technological maturity so that's the first possibility uh a second possibility is that there is a very strong convergence among all technologically mature civilizations in that they all lose interest in creating ancestor as I call them

these would be very detailed computer simulations of people like their historical forbears detailed enough that the simulated people in the simulations would be conscious so the second possibility is that they just lose interest in doing this and the third possibility is that we are almost certainly living in a simulation so there's this argument that shows that one of these three is true and the full argument involves some probability Theory but the basic idea can be grasped quite simply which is that suppose it were the case that the first possibility did not obtain so then some

non-trivial fraction of civilizations at our stage eventually reach technological maturity then suppose the second possibility also does not obtain so some non- negligible fraction of those mature civilizations are still interested in using the resources to running ancestry simulations you can then saw that because each mature civilization that devoted some resources to this purpose could run astronomical numbers of ancestor simulations you can show that if the first two possibilities do not obtain then there will be many many more simulated People Like Us than there will be non-simulated people like us in other words almost all people

with our kinds of experiences would be living inside simulations rather than outside them if the first two possibilities are false and conditional on that we should therefore think we are probably one of the typical simulated people rather than one of the exceptional non-simulated people so the structure of the argument then is that if you reject the first two hypotheses then the third one follows which then means you can coherently redact all three that that's the structure of the simulation argument there's a further possibility that if you have a simulated civilization that that simulated civilization might

inside the simulation develop the technology to run its own simulations and you might then have a kind of nested simulation with simulation inside the simulation you might have many levels of simulation but that's an optional extra doesn't the basic simulation argument doesn't prop presuppose that it's just that if you extrapolate the kind of computing power that an advanced civilization will eventually have and you figured if they would use some of that immense computing power to running simulations they could run an astronomical number of them so therefore in that scenario most people with our kinds of

experiences would be among the simulated people because for each original history there would then be maybe millions or or or billions of simulated histories so if that were the case we should believe the simulation hypothesis but the simulation argument doesn't imply the simulation hypothesis it just says that there is these three possibilities one of which is true leaving it open which one of them it's also possible that we will all fail to reach this level of technological maturity that all civilizations that our current stage of development go extinct for some other reason maybe there is

some advanced technology such that when you discover it you invariably use it to destroy yourself or there might be some other reason why why civilizations will not get through to this level of technological maturity but if they do and if they're still interested in that stage that using some of their resources to producing these ancestor simulations then most people like us would be living in the simulation so the simulation argument imposes a constraint on what you can coherently believe about the future and about our place in the world so it's different in disrespect from these

traditional arguments in philosophy that have challenged somebody to try to prove that prove that the external World exists or prove to me that uh I'm not dreaming or something like that prove to me that I'm not in a that I'm not a brain in a vat just being fed sensor input for electrons this is like a traditional discussion in philosophy which is a kind of an intellectual game to to to explore what we mean by knowledge and so forth but the simulation argument is different from that in that it doesn't start from a position of

doubt it doesn't start by challenging somebody to prove Beyond reasonable doubt that that that the external world really exists that it's not just a dream rather the simil argument starts by assuming everything is as it seems it starts by assuming that science tells us about the world that we have computers in the external world that those computers are getting faster and better with passing time and then using that scientific picture to think about what kind of capabilities will eventually be available to a mature civilization and then drawing out the implications of that and then ending

up in this position where we got to recognize that one of these three propositions is true and in particular the simulation hypothesis seems worth taking seriously as I said so the simulation argument shows that one of these three possibilities is true now I don't think we have very strong evidence for or against either one of them and therefore we should maybe distribute our Credence more or less evenly between them we cannot rule out that there is this strong filter that prevents any civilization at our stage from reaching technical maturity that seems consistent with what we

know we also can't rule out that there would be this strong convergence like maybe all sufficiently advanced civilization just loose interest in creating ancestor simulations now this this would be a departure from the current situation where there are certainly are many people in the world today who if they could would like to run these simulations of conscious being I mean we look at computer gaming like the more realistic that the simulated world is the more feeling it is to the people playing the games we have literature which tries to conjure up Virtual Worlds we have

maybe if the simulations could be made more accurate maybe historians who would be interested in in creating these to to sort of study the past there are many possible reasons why somebody might want to create an anest simulation if they could nevertheless we can't rule out the possibility that once some civilization reaches a sufficient level of maturity maybe they will lose interest in doing this maybe they'll realize that it would be horribly unethical for example to create simulated people who suffer if those if those simulated people are conscious maybe there is some other reason so

so there the reason why I don't assign an overwhelming probability to the simulation hypothesis is that there are these two Alternatives so I believe that the simulation argument is sound um but that's consistent with thinking that there is less than 50% chance that the simulation hypothesis is true well I had um sort of two strands of uh research interest that converged in the simulation arguments on the one hand I had had a long-standing interest in the future implications of Technology just thinking about what once we push closer to the physical limits what kind of technological

capabilities might Humanity one dat Vel on the other hand I had also done my PhD on developing a mathematical theory of observation selection effects this is a piece of methodology that you need to reason about questions uh that involve anthropic bias so that reason about questions that involve indexical information that's information about who you are what time it is where you are that kind of locates an observer within a particular model so I done a lot of research on that now once you have these two pieces of background then the simulation argument is really just

one inferential step away like once you combine thinking about where technology might lead with this kind of observation selection Theory then you just like one insight to kind of put these pieces together and you get a simulation argument popping out of that and U it's significant because it it imposes this surprising constraint on what you can coherently believe about uh the future and our place in the world so it might seem at first Like Anything Could Happen how we we have no evidence to kind of limit the range of possibilities we don't know anything about

what other civilizations if there are any out there in the universe How likely they are to survive what they would want to do if they became technologically mature uh we have no way of knowing whether in a simulation or not there might be many other possibilities but the simulation argument kind of constrains the space of possibilities in the surprising way so aside the possibility that we're ination it also helps inform us to some extent about the kinds of existential risk that we might confront in the future like threats to the survival of Earth originating civilization

um so it's one of those Clues we have that maybe together with other insights can can help um inform us about where we are in the world and what might what might lie in our future well the simulation argument does not presuppose that future people would be interested in running ancestor simulations it's in fact one of the three possibilities that are consistent with this argument is that there is a complete loss of interest in creating ancestor simulations but not that this loss does not only have to happen in our own lineage among our own descendants

but among all Advanced technological civilizations throughout the universe or almost all advanced civilization there has to be this convergence that they all lose this interest in order to account for why it is that um we are not in a simulation if we are not yeah so the idea this concept of an ancestor simulation would be a very particular kind of computer simulation unlike any we have today a computer simulation in which the simulated people are conscious so that might require simulating these virtual people to a level where individual neurons in their brains would be included

in the simulation so if all you're simulating is a kind of two-dimensional little little diagram then obviously there is no conscious experience arising from that but if you were simulating say a brain down to the level of individual neurons then on many theories that simulation itself could create Consciousness that what makes something conscious is not that it's built out of carbon atoms like we are inside the brain but that it implements a certain kind of computation and and so it is even when you had the ability to create simulations with that level of granularity where

you could simulate individual brains down to the level of individual neurons and and then have sentient people in those simul that's when you would create this possibility of running ancest simulations and that then is where this simulation argument becomes relevant that's been a a great deal of interest in it um which is understandable because it kind of does place this interes in constraint uh on where we might be in the world and in particular it seems to force us to take very serious the possibility that we are in the simulation which is like a kind

of very radical claim and it's important to understand it's not that we are in in a simulation in a metaphorical sense it's not that we could sort of think of the universe as if it were a digital computation or a cell autom that that kind of it's in that it would be a simulation in the literal sense and that there would be some some very intelligent individuals in some Advanced civilization that would build a computer and then run some prog program on that computer and we would be patterns inside that program um so it's a

striking it has kind of implications for these striking hypotheses so it's it's it's not surprising that that it has attracted a fair amount of attention I think um it seems to be coming in in waves that um like every every year or every other year or so there is a kind of I guess a read like a new set of people come across this for the first time and there's like this wave of uh of media attention um and uh at the same time a kind of gradual accumulation of scholarly research on this so there

have been a number of followup papers and so forth that that are coming out yeah the most common misunderstanding is to conflate the simulation hypothesis with the simulation argument the simulation hypothesis is the statement that we are living in a computer simulation I think that is less than 50% likely to be the case however I do believe in the simulation argument which shows or purports to show that one of three propositions is true one of which is the simulation hypothesis but then there are these other two propositions as well that also are alternatives the possib

that almost all civilizations go extinct before reaching technological maturity and the possibility that there's this strong convergence among all technologically mature civilizations such that they all lose interest in creating ancestor simulations um and it's it's hard it's hard to for people to keep those two ideas apart I think there is a kind of complexity limit in in in a lot of media commity communication that it's easy enough to convey kind of one idea and in in more highbrow media you might convey an idea that has two parts like on the one hand this and on

the other hand that but the simulation argument kind of has three components and it just seems to exceed the communication bandwidth it just gets a little bit too complex for popular media sometimes but it's it's a little bit complex but it's really not more complex than most like educated people could probably get their heads around if they thought about it for half an hour or an hour kind of if they actually tried it it doesn't require a PhD to get your head I mean there is a little bit of complexity there but it's just a

little bit too much to fit in normally in a sort of 5 minute interview setting well I get every once in a while uh emails from various people who claim to have seen some anomaly that they claim prove to me that we are in a simulation however I don't credit these reports because if we are not in the simulation you would expect a certain number of people to believe themselves to have spotted weird things like there's one guy I don't know he claimed to see pixels in the bathroom mirror or something like that but we

know from psychology and from Paranormal phenomena that there are just these some people will have hallucinations some people will misremember what they've seen some people will report erroneously and I think that's true whether we are in a simulation or not so even if we are in a simulation I think that these kinds of reports are much more likely to be due to these normal psychological factors than they are to anybody actually observing a glitch in the simulation to run a simulation at all would require some extremely advanced technology presumably the people doing this would be

some kind of super intell postum life forms not thinkers like us and I think that if they had the ability to create a simulation like this at all they would also have the ability to paper over any cracks any glitches that that we might be able to detect or if we did detect something that they didn't want us to detect then they could erase the memory or rewind the simulation and and redo it yeah it's possible that the simulators might make a mistake but that would also be in an extremely strong position to cover that

mistake up from the people in the simulation they would presumably be super intelligent and easily able to outsmart us they have complete control over the whole structure of reality and if you're running the simulation you could even in principle go in and edit the brain states of people inside the simulation you could erase the memories you could rewind the simulation there would be any number of tools to avoid a kind of glitch from ruining the simulation if if they wanted to do that there would conversely be many ways in which they could let us know

that we were in a simulation if they wanted to do that could be a big window popping up in front of you like informing you you are in a simulation like so it's a hypothesis that certainly there are possible observations which if we made them would count say heavily in favor of being in a simulation and therefore there are observations that would count against it as well um but one way to get some at least weak probabilistic evidence for or against the simulation hypothesis is via the simulation argument itself so if you remember the argument

tries to show one of these three propositions is true so that means that any evidence that we have that makes the other two Alternatives less likely would have to make the simulation hypothesis more likely so for instance if we discovered strong evidence that there is some very great filter that would tend to make it impossible for any civilization to reach technological maturity that would tend to reduce the probability of the simulation hypothesis um now now one situation in which we would get very strong evidence that we are in a simulation is if we ourselves or

our descendants one day reach technological maturity and actually develop the technology to create their own answers to simulations imagine that they buil this massive computer sometime in the future set it all up so that it will Implement an ancestor simulation and they are about to flick the on switch well at that stage they could pretty much rule out the other two alternative hypothesis right because they knew that if they had made it through to technological maturity that would be strong evidence against the first possibility it would show that not almost all civilizations go extinct before

becoming technologically mature and if they are still interested in creating anest dis simulations that would rule out the second possibility it's not the case then that almost all technologically immature civilizations have lost interest if you are a technological mature civilization and you are interested that's strong evidence against the second possibility and that would then only leave them with a third possibility that they are in a simulation so even when we got to that stage we would have very strong reason to infer that we ourselves are simulated well so another major research um focus here uh

in my research group is uh studying existential risks so an existential risk is one that would either imperil the survival of uh Earth originating intelligent life like a risk that could cause our Extinction or that could permanently and drastically destroy our future our future potential for desirable development so an existential catastrophe is one in a sense that would destroy our entire future so there's never been an existential catastrophe I mean we are still here and they POS a unique set of challenges to precisely because we have very limited opportunity to learn from experience we got

to be proactive with regard to these risks um we're all in the same boat when it comes to existential risk and the values uh that are on the line are extremely large so if one asks what kind of risks are there that raise to this level of existential risk I would say that all the really big existential risk are ones that arise in one way or another out of human activity and there's an easy way to see that which is that our human species has lived on this planet for more than 100,000 years surviving all

kinds of threats from nature we have survived earthquakes and volcano eruptions and asteroid impacts and fire storms and everything so it's unlikely that any of those would do us in within the next hundred years if they have haven't done in within the previous 100,000 years by contrast we are introducing completely new types of Hazards into the world through our own activities and more specifically I think that all the really big existential risks are related to anticipated future technological breakthroughs for example in synthetic biology or in advanced forms of molecular nanotchnology or in machine intelligence artificial

intelligence I think those are some of the Aras where in the future uh there might be major existential risks