The Simulation Argument FAQ

Nick Bostrom
(2008) Updated 2011; Version 1.10
 
 

1. What is the simulation argument?

The simulation argument was set forth in a paper published in 2003. A draft of that paper had previously been circulated for a couple of years.

The argument shows that at least one of the following propositions is true: (1) the human species is very likely to go extinct before reaching a “posthuman” stage; (2) any posthuman civilization is extremely unlikely to run a significant number of simulations of their evolutionary history (or variations thereof); (3) we are almost certainly living in a computer simulation. It follows that the belief that there is a significant chance that we will one day become posthumans who run ancestor-simulations is false, unless we are currently living in a simulation. A number of other consequences of this result are also discussed.

The argument has attracted a considerable amount of attention, among scientists and philosophers as well as in the media.

References:

N. Bostrom, “Are You Living in a Computer Simulation?” Philosophical Quarterly, 2003, Vol. 53, No. 211, pp. 243-255. URL:  http://www.simulation-argument.com/simulation.html.

N. Bostrom & M. Kulczycki, “A Patch for the Simulation Argument” Analysis, 2011, Vol. 71, No.1, pp. 54-61. URL:  http://www.simulation-argument.com/patch.pdf.

 

2. Do you really believe that we are in a computer simulation?

No. I believe that the simulation argument is basically sound. The argument shows only that at least one of three possibilities obtains, but it does not tell us which one(s). One can thus accept the simulation argument and reject the simulation hypothesis (i.e. that we are in a simulation).

Personally, I assign less than 50% probability to the simulation hypothesis – rather something like in 20%-region, perhaps, maybe. However, this estimate is a subjective personal opinion and is not part of the simulation argument. My reason is that I believe that we lack strong evidence for or against any of the three disjuncts (1)-(3), so it makes sense to assign each of them a significant probability.

I note that people who hear about the simulation argument often react by saying, “Yes, I accept the argument, and it is obvious that it is possibility #n that obtains.” But different people pick a different n. Some think it obvious that (1) is true, others that (2) is true, yet others that (3) is true. The truth seems to be that we just don’t know which of the disjuncts is true.

 

3. Is the simulation argument a variant of Descartes' daemon or the brain-in-a-vat argument?

No, the simulation argument is fundamentally different from these traditional philosophical arguments (as explained in my reply to Brian Weatherson). The purpose of the simulation argument is different: not to set up a skeptical problem as a challenge to epistemological theories and common sense, but rather to argue that we have interesting empirical reasons to believe that a certain disjunctive claim about the world is true (that is, (1)v(2)v(3)). The simulation argument relies crucially on non-obvious empirical premises about future technological abilities. And the conclusion of the simulation argument is not simply that we cannot be certain that we are not living in a simulation. If we knew that fSIM (the faction of all human-like beings who are simulated) was very small but non-zero, we might not be able to be completely certain that we are not in a simulation, but this would not be a very interesting contention. (If we think that somewhere in our infinite universe there are a few “envatted brains”, then maybe we shouldn’t assign a strictly zero credence to us being envatted brains either, but so long as we thought that the proportion of brains in vats to brains in crania was small enough, we would have no ground for seriously doubting that we are not brains in vats, at least if we lacked specific evidence to the contrary). The simulation argument is also different from ordinary brain-in-a-vat arguments in that it doesn’t begin from a starting point of doubt and ask for some compelling reason for canceling that doubt. Rather, it begins from the starting point that things are the way we believe they are, and then, while granting us that we might be justified in assigning a high initial credence to these beliefs, nevertheless tries to show that we have specific empirically-grounded reasons for revising these initial beliefs in a certain way – not so as to make us generally agnostic about the existence of an external world, but to accept the disjunctive conclusion. Thus, the simulation argument is not best thought of as a skeptical argument that would have us be more agnostic, but rather as an argument that would have us increase our credence in one particular disjunction (and decrease our credence in its negation). It aims to tell us something about the world rather than to advise us that we know less about the world than we thought we did.

References:

N. Bostrom, “The Simulation Argument: Reply to Brian Weatherson,” Philosophical Quarterly, 2005, Vol. 55, No. 218. URL: http://www.simulation-argument.com/weathersonreply.pdf.

 

4. If we are in a simulation, doesn’t that undermine the reasoning in the simulation argument?

The simulation argument relies on an assumption about the technological capabilities of a mature civilization. Our evidence for this assumption is empirical: it is based on our best theories about the physical limits of computation and the kinds of constructions that could be built with advanced molecular manufacturing techniques etc., and our confidence in these theories rests ultimately on observations of the world we see around us. But if we are in a simulation, then how could we trust these observations? Might they not inform us only about the simulated reality and not about any hypothesized underlying of reality in which the simulation is being run? And if so, does this not undermine the simulation argument by casting doubt on its empirical assumptions?

If we reflect more carefully on this objection, however, we see that it fails. The claim that we cannot have any information about the underlying reality if we are in a simulation is false. In a simulation, we can certainly know the following two conditional claims:

A. If we are in a simulation, then the underlying reality is such as to permit simulations, it contains at least one such simulation, and (3) is true.

B. If we are not in a simulation, then the empirical evidence noted in the simulation argument is veridical taken at face value, suggesting that a technologically mature civilization would have the ability to create vast number of simulations; and consequently, by the simulation argument, there is a very high probability at least one of the disjuncts in (1)-(3) is true.

Since we either are, or are not, in a simulation, we can conclude, even if we are in a simulation, that at least one of the disjuncts (1)-(3) is true.

References:

N. Bostrom, “Self-Locating Belief in Big Worlds: Cosmology’s Missing Link to Observation,” Journal of Philosophy, 2002, Vol. 99, No. 12, pp. 607-623. URL: http://www.anthropic-principle.com/preprints/cos/big.html.

F. Besnard, "Refutations of the Simulation Argument," 2004, Unpublished manuscript. URL: http://perso.wanadoo.fr/fabien.besnard/pdfrefut.pdf [presents a variation of this objection].

 

5. “I can see glitches in the Matrix!”

I don’t buy that. We should expect to hear such reports occasionally even if we are not in a simulation. Even if we are in a simulation, the most plausible explanation for such reports is not that they result from any real “glitch” but rather that they originate in the ordinary frailties of the human mind (hallucinations, psychiatric problems, visual illusions, self-deception, fraud, and so forth).

It seems likely that the hypothetical simulators, who would evidently have to be technologically extremely advanced to create simulations with conscious participants, would also have the ability to prevent these simulated creatures from noticing anomalies in the simulation. This could be done by avoiding anomalies altogether, or preventing them from having noticeable macroscopic ramification, or by retrospectively editing the brain states of observers who had happened to witness something suspicious. If the simulators don’t want us to know that we are simulated, they could easily prevent us from finding out. Consider that even our own humble brains – unaided by technology – usually manage to prevent us from realizing when we are dreaming at night, even though the typical dream is teeming with the most fantastic anomalies.

 

6. What happens to the simulation argument if the universe is infinite?

This is an interesting issue, which is deliberately set aside in the original paper. Certainly some modifications are necessary once we admit the possibility of an infinite universe that may contain infinitely many simulated and non-simulated people. In this case, the ratio of simulated people to the total number of people is not defined.

To deal with these infinite cases, we need to do something like thinking in terms of densities rather than total populations. A suitable density-measure can be finite even if the total population is infinite. It is important to note that we to use some kind of density-measure of observation types quite independently of the simulation argument. In a “Big World” cosmology, all possible human observations are in fact made by somebody somewhere. (Our world is may well be a big world, so this is not a farfetched possibility.). To be able to derive any observational consequences from our scientific theories in a Big World, we need to be able to say that certain types of observations are more typical than others. (See my paper “Self-Locating Belief in Big Worlds” for more details on this.)

The most straightforward way of making this notion precise in an infinite universe is via the idea of limit density. Start by picking an arbitrary spacetime point. Then consider a hypersphere centered on that point with radius R. Let f(A) be the fraction of all observations that are of kind A that takes place within this hypersphere. Then expand the sphere. Let the typicality of type-A observations be the limit of f(A) as R--->infinity.

To apply this idea to the case where we might be living in a simulation, we can use a similar rule, except that we define the seed point to be the location at the bottom level of reality where the computer is located that is ultimately running our simulation, and we take the hypersphere expansion at that level of reality. (One could modify this rule so that it would work also in the case where it is possible that there is no lowest level of reality but rather an infinite regress of simulations within simulations “all the way down”.)

References:

N. Bostrom, “Self-Locating Belief in Big Worlds: Cosmology’s Missing Link to Observation,” Journal of Philosophy, 2002, Vol. 99, No. 12, pp. 607-623. URL: http://www.anthropic-principle.com/preprints/cos/big.html.

F. Besnard, "Refutations of the Simulation Argument," 2004, Unpublished manuscript. URL: http://perso.wanadoo.fr/fabien.besnard/pdfrefut.pdf [presents a variation of this objection].

 

7. Couldn't we simply be in the first generation, so that all the simulated creatures that will one day be created don't yet exist?

Indeed we could, but the question is what probability we should assign to this possibility. If we assume that the total number of simulated people with experiences like ours that will eventually have lived is vastly greater than the number of non-simulated people with such experiences that will ever have lived, then we should believe with very high probability that we are among the simulated majority rather than the non-simulated minority. None of all these simulated and non-simulated people have any way of telling which generation they are in. If they all guess that they are in the first generation, almost everybody will guess wrong. If they all guess that they are in some later generation, then almost everybody will guess right.

The part of the argument most directly pertinent to this point is the "Bland Indifference Principle", defended briefly in section V of the original paper. For a defense of a much stronger principle from which the Bland Indifference Principle can be derived as a trivial special case, see my book Anthropic Bias.

References:

N. Bostrom, “The Simulation Argument: Reply to Brian Weatherson,” Philosophical Quarterly, 2005, Vol. 55, No. 218. URL: http://www.simulation-argument.com/weathersonreply.pdf.

N. Bostrom, Anthropic Bias: Observation Selection Effects in Science and Philosophy (Routledge, New York, 2002).

 

8. How has the Simulation argument affected how you live?

Barely. (Except, of course, it has affected my work in one particular research area.)

The original paper has some discussion of various potential practical implications of the simulation-hypothesis, and Robin Hanson has also published a paper on this. My view on Hanson’s paper is that the considerations he point out may well be sound as far as they go, but they are quite weak and in some parts counterbalanced or outweighed by other considerations (such as the ones explored in my paper). The upshot, I think, is that the simulation-hypothesis currently does not seem to have any radical implications for how one should live, although there might be a multitude of weak or subtle implications. The practical impact of these is of course further modulated by the possibility that the simulation-hypothesis is false.

The simulation argument does, however, narrow down the set of tenable possibilities and by doing so it helps to shed light on, among other things, the prospects of our species. It is also conceivable that, when combined with other considerations or arguments, some of which may yet to be discovered, it could have significant implications for our individual or collective planning.

 

9. Isn’t the simulation-hypothesis untestable?

There are clearly possible observations that would show that we are in a simulation. For example, the simulators could make a “window” pop up in front of you with the text “YOU ARE LIVING IN A COMPUTER SIMULATION. CLICK HERE FOR MORE INFORMATION.” Or they could uplift you into their level of reality.

We could also obtain strong indirect evidence, such as one day observing that we ourselves have created the appropriate kind of computer simulations. If we were to learn more about the probability of survival for human-like species, that could also be relevant information. For instance, if we learnt that the existential risks we will be confronting are so large that we should expect practically every advanced civilization to succumb to them, that would reduce the probability of the simulation-hypothesis. One can think of a large number of other possible pieces of observational evidence, as well as more indirect theoretical results, that would either increase or decrease the probability of the hypothesis. So in this sense, the simulation-hypothesis is clearly testable.

Now one might wonder whether there is any kind of test that we could conduct in the near term that would cast light on which of the three disjuncts is true. I doubt that there is any simple experiment that will do this – like mixing some chemicals together in a bottle and checking whether they turn red or blue. (Most theoretical science is of course untestable in this sense, so it is not a useful criterion for whether something is worth taking seriously.) But there might well be some clever analyses that we could do that would show that some piece of observational evidence is relevant in some surprising way to probabilities of the simulation-hypothesis and the other disjuncts. It might be too much to hope for that we could think of any such analysis that would give conclusive or extremely strong evidence about which of the three disjuncts is true, but I think it might be feasible to get evidence that would at least give weak clues and conditional hints.

 

10. What is the relation between the simulation argument and religion?

It has no direct connection with religious conceptions of a literally omniscient and omnipotent deity. The simulation-hypothesis does not imply the existence of such a deity, nor does it imply its non-existence.

The last section of the original paper speculated about certain parallels that could be drawn between traditional religious conceptions and our relations to our hypothetical simulators. These simulators would have created our world, they would be able to monitor everything that happens here, and they would be able to intervene in ways that conflict with the simulated default laws of nature. Moreover, they would presumably be superintelligent (in order to be able to create such a simulation in the first place). An afterlife in a different simulation or at a different level of reality after death-in-the-simulation would be a real possibility. It is even conceivable that the simulators might reward or punish their simulated creatures based to how they behave, perhaps according to familiar moral or religious norms (a possibility that gains a little bit of credibility from the possibility that the simulators might be the descendants of earlier humans who recognized these norms).

One person who had been a hardcore atheist his whole life, told me, when I explained the simulation argument to him, that it was the best argument for God’s existence that he had ever heard. He became an agnostic.

However, it is important to stress that the simulators implied by the simulation-hypothesis would be naturalistic entities, subject to the laws of nature at their own level of reality; they would not be strictly omniscient or omnipotent, and they might well be finite. Moreover, we need to keep in mind that the simulation argument does not imply the simulation-hypothesis.

 

11. What if we are simulated by aliens?

"Isn't the argument too anthropocentric? What if some kind of alien race constructed the super computer that simulates our universe, and the simulation was made to simulate mainly their kind? In this double-what-if scenario (we are simulated, but by an alien race that has nothing to do with humanity), what are the possible implications to your original theory?"

Formally, the simulation-hypothesis includes the possibility that we are simulated by an extraterrestrial civilization. However, the inclusion is redundant. If the simulation-hypothesis is true, then we are living inside a computer, and whichever civilization built that computer is our "home" civilization by definition.

Of course, it is possible that the simulators and their ancestors are more similar to some extraterrestrial civilization in our universe (if there are any) than they are to us, so in that sense it is possible that we are simulated by the descendants of an alien-like civilization.

More generally, simulators might create many simulated people who are very different from their own ancestors, or who live in worlds that are very different from the one that the simulators live in. It is possible that we are living in such a simulation. I don't know of any way of estimating the probability that our hypothetical simulators (or their ancestors) are similar to us, or that their world is similar to the world we experience. (The original paper focuses on ancestor-simulations because the methodology is more solid for that case. It is less clear whether some kind of principle of indifference could also be applied to a reference class of "observer-moments" that are very different from one another. For more on the reference class problem, see Anthropic Bias.)

References:

N. Bostrom, Anthropic Bias: Observation Selection Effects in Science and Philosophy (Routledge, New York, 2002).

 

12. Why do you think there’s been so much interest in this?

If the simulation argument is sound, it tells us something surprising and profound about the world. The argument is powerful because from some rather simple empirical assumptions it derives a remarkable conclusion about the world (i.e. that it satisfies the abovementioned tri-partite disjunction). It is rare to get so much leverage out of a short philosophical argument.

 

13. If we are merely simulated, does that mean that the world isn't "really real"?

No clarity is gained by asserting that world isn't "really real" if we are in a simulation. The simulated world that we experience would, however, be only a part of reality. Reality would also contain the computer that runs the simulation, the civilization that built the computer, and perhaps many other simulations and much else beside.

 

14. How did you come up with this?

In my doctoral work, I had studied so-called self-locating beliefs and developed the first mathematical theory of observation selection effects, which affects such beliefs. I had also for many years been thinking a lot about future technological capabilities and their possible impacts on humanity. If one combines these two areas – observation selection theory and the study of future technological capacities – then the simulation argument is only one small inferential step away.

Before the idea was developed in its final form, I had for a couple of years been running a rudimentary version of it past colleagues at coffee breaks during conferences. Typically, the response would be “yeah, that is kind of interesting” and then the conversation would drift to other topics without anything having been resolved.

I was on my way to the gym one evening and was again pondering the argument when it dawned on me that it was more than just coffee-break material and that it could be developed in a more rigorous form. By the time I had finished the physical workout, I had also worked out the essential structure of the argument (which is actually very simple). I went to my office and wrote it up.

(Are there any lessons in this? That new ideas often spring from the combining of two different areas or cognitive structures, which one has previously mastered at sufficiently a deep level, is a commonplace. But an additional possible moral, which may not be as widely appreciated, is that even when we do vaguely realize something, the breakthrough often eludes us because we fail to take the idea seriously enough.)

 

15. "If the Universe is not simulated to a quantum degree of accuracy, the simulation can be immediately exposed via Bell's inequality or some similar test."

"Thus," the objection continues, "in order to create a virtual universe sufficient to withstand quantum physics experiments, you would need either a quantum computer or a truly super-astronomical amount of non-quantum computing power."

The assumption that a universe not simulated to a quantum degree of accuracy can be immediately exposed is unfounded.The original paper notes that even an atomic-scale simulation (much less a fully accurate quantum simulation) of the inside of your desk is not needed, and that extremely crude resolution would suffice for distant astronomical objects.  If you train an STM on a piece of your desk, then you could distinguish atomic-scale feature, and these would then have to be added to the simulation.  But this could be done on an ad hoc basis: wherever somebody is paying attention, the requisite detail could be created.  Moreover, if through some bug in the simulation, somebody were to notice an anomaly, their brains could be edited afterwards to remove the memory (or the whole simulation could be rerun from some cached earlier state in a way that avoids the problem).

So basically, the idea is that some of our experimental findings could be "faked" by the simulators, if they wanted to conceal the fact that we are in a simulation and if such faking was the most efficient way to conceal it. Consider the fact that while you are dreaming, your own brain often succeeds in making you unaware of the fact that you are dreaming.  If your own humble brain can do this, then presumably it would be quite feasible for some technologically super-advanced builders of ancestor-simulations to achieve the same delusion.

Note also that we do not know how much computing power might be available in the simulating basement universe: the simulators might have access to quantum computers, or "super-astronomical" classical computers. However, it is not necessary to assume that they have any more computing power than we already know to be feasible in our own universe if it is not a simulation.

 

16. Can the simulation argument be generalized?

"For example, would a terraforming argument, that holds either we'll almost certainly never terraform and seed many worlds like our own or we're almost certainly living in a terraformed and seeded world, remain valid?"

Yes, the form of the simulation argument could in principle be applied more generally, but one would have to check in each case whether the necessary empirical presuppositions obtain and whether the conclusions are interesting and significant.

For example, with the terraforming argument, we would have to assume that the seeded civilizations could not detect directly whether or not they had been seeded; also, that each advanced civilization that decided to terraform would create very large numbers of terraformed planets on which creatures similar to ourselves would evolve.  Both of these assumptions might be doubted.  The argument would also have to consider our temporal position.  Civilizations resulting from the seeding or terraforming of a planet would be able, like we are, to determine the cosmic epoch in which they lived, and this might place some constraints on how many planets civilizations arising earlier than ours would have had time to seed by now (taking into account the lag required for the seeds to mature).

Likewise, one can consider a "cosmoforming argument", focusing on the possibility that an advanced civilization with some as-yet unknown technology might be able to induce the creation of baby universes (perhaps expanding into hidden dimensions).  The transposition of the simulation argument form to this hypothesis is straightforward.  However, the conclusion of this cosmoforming argument seems less significant than that of the simulation argument.  It would not be particularly surprising to learn, for example, that it is impossible for any ever-so-advanced civilization to spawn new universes; nor, perhaps, would be as shocking to learn that our universe had been spawned from some parent universe as a consequence of the actions of intelligent agents there as it would be to learn that we are living in a computer simulation.

The bottom line is that terraforming, cosmoforming, and "Trueman-show" arguments are compatible with the simulation argument, and are parallel in terms of their logical structure.  Some of these alternative applications might be interesting.  They each have to be evaluated on their merits.

References:

B. Dainton, "Innocence Lost: Simulation Scenarios: Prospects and Consequences" (2002, October), Draft. URL: http://www.simulation-argument.com/dainton.pdf