This is sort of a follow up to “Brian Leiter’s post”:http://webapp.utexas.edu/blogs/bleiter/archives/001665.html on philosophy blogs in Newsweek. And equally belated.
Last week the Sydney Morning Herald ran an article by Paul Davies about “the possibility that our universe is a simulation”:http://www.smh.com.au/articles/2004/07/21/1090089219062.html. What was interesting, to me at least, was that the article cited Nick Bostrom’s argument to this effect in _The Philosophical Quarterly_. (An online version of Bostrom’s paper is “here”:http://www.simulation-argument.com/.) It isn’t every day you see a philosophy journal cited in the morning newspaper. Sadly Davies didn’t cite, or even talk about, my “refutation of Bostrom’s argument”:http://brian.weatherson.net/sims.pdf also in _The Philosophical Quarterly_. So I thought I may as well take the chance to revisit that debate and say what I thought was most important about it. Anyone who wants to write to the SMH making either of the points below is much more than welcome!
(Blog history note: I first found out about Bostrom’s paper through a chain of links starting with “Instapundit”:http://www.instapundit.com/archives/003465.php, which probably makes my paper the first philosophy paper to be the result of a blog entry.)
The core of Bostrom’s argument is fairly simple. Given some plausible assumptions it is likely that there will be many more simulated conscious beings than material conscious beings over the course of the history of the world. (Bostrom makes all his conclusions conditional on these assumptions, but it makes life easier to just assume they are true, and I won’t question any of the assumptions in any case.) Now if we accept that we’re just as likely to be any one of the simulees as one of the material entities, it follows that we’re more likely to be a simulation than a material entity.
There are two problems with this argument, both due in different ways to the fact that we actually have quite a bit of evidence about just what our place in the universe is.
First, I know right now that I’m not feeling cold. So it’s not just as likely that I’m one of the people as any other, because I’m not one of the ones feeling cold. A much more plausible principle is that for any of the conscious beings _whose conscious experiences are indiscriminable from mine_ I’m just as likely to be any one as any other. (I think even that’s not true, but let’s grant it for now.) The problem is that the assumptions don’t give us the conclusion that there are likely to more simulees than materials with conscious experiences indiscriminable from mine. Indeed, when we think about the vast variety of conscious experiences it starts to seem unlikely that there is any other being, simulation or material, with the same experiences as I have. So for all the argument has told us, the proportion of beings with conscious experiences indiscriminable from mine who are material is 1. And then applying the more plausible principle, we get that the probability that I’m material is 1.
To get the original argument to work, Bostrom has to come up with stronger principles than the plausible one I listed. But these principles all seem either ad hoc or inconsistent for reasons related to Goodman’s ‘grue’ paradox. I won’t go into the detail, because what really seems important is that the makeup of classes of beings that I know I’m not in just doesn’t seem that relevant to what probability I should assign to my being material or simulated.
The other point is that part of Bostrom’s argument relies on a principle of indiscriminability that is importantly false. I said above that Bostrom-style arguments can’t affect the probability I assign to being cold, because I can just tell that I’m not cold. Most modern theorists of perception would say the same thing about the probability I should assign to being a material being. I can just look at _see_ that I’m material, whatever other philosophers would tell us.
Why wouldn’t we believe this? One intuitive reason turns out to be quite bad, for reasons Tim Williamson has recently stressed. It’s intuitive to think that if we were simulees rather than material beings we couldn’t tell the difference, which I guess is true. And it’s intuitive to think that this implies if we are material rather than simulees that we couldn’t tell the difference. But this doesn’t follow, because ‘is indiscriminable from’ is not symmetric.
We can make this intuitive by considering real-life cases. After 2 martinis there’s all sorts of things I can safely do, e.g. walk down steep steps, that I can’t safely do after 4 martinis. But when I’ve had 4 martinis I can’t tell the difference between by abilities and what they were when I’d only had 2 martinis. My ability to tell what’s happening is deteriorating as what is happening changes. That in no way means that after the first 2 martinis I can’t tell that I can safely walk down the steps.
The point generalises. Just because I can’t tell A from B when in situation A doesn’t mean I can’t tell A from B when in situation B _if my capacities are greater in B than in A_. And that’s just what happens in the simulation/material case. True, I can’t tell if I’m a simulee that I’m the product of a simulation rather than material. But if I’m material I have epistemic capacities that I don’t have if I’m a simulation. E.g. I can easily form reliable beliefs by perception. So nothing follows about what I can and can’t detect as a material being by thinking about what happens if I’m the product of simulation. The catchphrase version of this is that what happens in bad scenarios is not that relevant to what I can and can’t know by perception if I’m actually in a good scenario.
The upshot of all this is that we don’t need to worry about philosophical arguments talking us out of our naive belief that we are material beings any more than we could possibly worry about philosophical arguments talking us out of our naive beliefs that we are feeling warm (if we are). Compare: even if we knew that 99.999% of conscious beings feel cold, if we feel warm we just won’t be moved to change our beliefs by any philosophical arguments based on that statistic. I’m inclined to think that questions about whether we are material or a simulation are similar.
{ 23 comments }
John Roberts 07.27.04 at 3:01 am
Do “Java script web-applets” run on a virtual machine?
Lukas 07.27.04 at 4:31 am
the martini example should be more like “if i’m drunk, i’m not good at telling whether or not i’m drunk, and am likely to think i’m not. can i tell if i’m drunk?”
Martey 07.27.04 at 4:34 am
“Javascript” is lines of code embedded in HTML or in separate files (with the *.js extension). It is processed by the web browser. I am pretty sure that it does not run in a virtual machine (VM) normally. There do seem to be javascript virtual machines floating around the Internet, if you have a need for one, though.
Java applets on the other hand, do run in a virtual machine. If you are running Windows, you probably have either the Microsoft VM or the Sun VM installed on your computer. In my opinion, Sun’s virtual machine is better; there are also the company that makes Java.
Thomas Dent 07.27.04 at 10:52 am
Why do you make the assumption that the simulation is a ‘bad scenario’ in which
you have lesser capacities than you think you have now? Surely this prejudices
the argument in an unfair way. It depends how good the simulation is. Moreover,
your apparent sensorial capacities aren’t so superexcellent anyway, at least if
they’re anything like mine. I assume you also have had the experience of seeing
a man or a black-and-white cat and then a moment later just a pattern of light
and shadow. Might this not have been a moment when the limitations of the
simulation were showing? Might there not be a world of beings whose senses, not
being simulated, don’t belie them nearly so often?
Of course, this isn’t Bostrom’s argument if he wants to say that we can’t tell
by any feature of our conscious experience whether or not we are in a simulation.
But in this case, I don’t see how we can cite a naive belief as evidence. We would
have equally good simulated perceptions of the simulated world as our material
perceptions of the material world.
Surely arguments of the type Bostrom uses are part of our everyday repertoire of
thinking? For example, if we happen to be thinking about a particular song and
it comes on the radio. We can perfectly well imagine the class of people who are
having similar or different conscious experiences and perform statistical
operations on this class, to come up with the conclusion that such a thing is
or is not the product of some sort of supernatural synchronicity or divine
providence.
I don’t understand the analogy with feeling cold. Whether or not you are
feeling cold is not an objective externally measurable property of your
existence, it is a subjective property of your internal consciousness, about
which it is impossible to be deceived. Whether or not you are a simulation
is potentially an externally measurable property (at least if you are one!)
about which your naive intuition may be wrong.
If you take the analogy seriously, you end up thinking that naive intuitions
about an objective property of existence are as reliable as our knowledge of
our own conscious experience (which is tautologously reliable).
Thomas Dent 07.27.04 at 10:55 am
er, I mean the *classes* of people having similar experiences or different experiences.
Matt 07.27.04 at 11:24 am
I’m with Thomas — aren’t we supposed to accept that the simulation is so good we can’t tell if we are in it or not, as part of the thought-experiment?
Your asymmetry argument makes sense in that, from the viewpoint of this “real” world, we can tell the difference between this “real” world and a simulation subworld inside it — but how do we know there’s not another world one level above us, to which our world is itself only a subworld?
Here’s an example: I’m shortsighted, but until I was about 11 years old I didn’t realise it. Never having seen out of anyone else’s eyes, how was I to know that it wasn’t normal for things at a distance to be blurry? (Being a nerd, I sat at the front of the class right near the blackboard, and was already bad at ball games, so the adults around me didn’t notice either)
I was in a kind of mental subworld where vision is bad as a matter of course. I didn’t know it until people one world up, who had good vision and knew the difference, let me in on the secret.
The REAL issue with this simulation hypothesis is surely that if we grant “you can’t tell the difference between the simulation and the ‘real’ world” as a precondition, then it becomes meaningless, and we should no more concern ourselves with it than with the by-definition unanswerable question of whether or not we have an invisible and intangible demon ™ on our shoulder.
Matt 07.27.04 at 11:26 am
I’m with Thomas — aren’t we supposed to accept that the simulation is so good we can’t tell if we are in it or not, as part of the thought-experiment?
Your asymmetry argument makes sense in that, from the viewpoint of this “real” world, we can tell the difference between this “real” world and a simulation subworld inside it — but how do we know there’s not another world one level above us, to which our world is itself only a subworld?
Here’s an example: I’m shortsighted, but until I was about 11 years old I didn’t realise it. Never having seen out of anyone else’s eyes, how was I to know that it wasn’t normal for things at a distance to be blurry? (Being a nerd, I sat at the front of the class right near the blackboard, and was already bad at ball games, so the adults around me didn’t notice either)
I was in a kind of mental subworld where vision is bad as a matter of course. I didn’t know it until people one world up, who had good vision and knew the difference, let me in on the secret.
The REAL issue with this simulation hypothesis is surely that if we grant “you can’t tell the difference between the simulation and the ‘real’ world” as a precondition, then it becomes meaningless, and we should no more concern ourselves with it than with the by-definition unanswerable question of whether or not we have an invisible and intangible demon ™ on our shoulder.
Matt 07.27.04 at 11:30 am
Sorry. It appears a perfect simulation of me was making the same comment at the same time. Great minds…
lambda 07.27.04 at 12:13 pm
Re: Java, Javascript, and VM’s…
Not to get too pedantic, but the technical way of saying this is:
“Java” (a different language from Javascript, sharing only the name and some syntactic similarities in common) is compiled into an intermediate language (“bytecode,”) which is then interpreted by the V(irtual) M(achine). The term ‘machine’ is applied here because bytecode looks a lot like the low-level instructions that are interpreted by the hardware (chip) in your computer… and so you can think of the VM as an idealized form of a ‘register machine’ (actually, in this case, a ‘stack machine,’ but ok).
Javascript, is a language which is ‘interpreted’ by your browser… but there’s nothing unique about the browser being able to do that. I could just as easily write a stanadlone app that interprets JS, or build a browser with an integrated VM.
The only real difference, at a high level, is the intermediate compilation step that Java (typically) goes through… but it would be just as easy to write a javascript interpreter that used an intermediate compilation stage, or a Java “VM” that interpreted the language without compilation.
So theoretically, the Java VM and the ‘interpreter’ embedded in your browser are the same thing… they’re both ‘evaluating’ an artificial language, and ‘showing’ you the result…
All the rest of the terminology (in addition to the decision to call it a “VM” in the first place) is basically marketing-speak.
sorry to interrupt; please resume the original discussion….
Monica 07.27.04 at 1:00 pm
Yeah, I’m with Thomas and Matt. In addition, there’s another problem with your refutation: it hinges on assuming that if sets of experiences are unique, such that no two beings share them, then all beings are material. How in the world do you get that?
John Quiggin 07.27.04 at 1:16 pm
What would Descartes say about this? Having first concluded that his existence is robust to the simulee hypothesis, he eventually uses the goodness of God (proved in intermediate steps) to reject it.
But if we suppose a world in which everyone can be an omnipotent deceiver, and many people are, it’s hard to see how Descartes would make his case.
John Quiggin 07.27.04 at 1:24 pm
Here’s a version of Bostrom that seems robust to your refutation. Suppose we create simulees whose experiences are very similar to ours, but coarser in some way we can detect and they cannot.
By hypothesis, they also create coarser simulees and so on ad infinitum or to levels we can’t observe.
Now there’s no way for us to tell that we aren’t at some intermediate point in a sequence of this kind.
We can represent the problem by a set of integers, arbitrarily large but bounded above, in which, if n and n-1 are in the set, n-1 is a simulee generated by n. Then our chance drawing the first number is 1/N which can be arbitrarily small.
John Quiggin 07.27.04 at 1:26 pm
Sorry, I missed Matt’s comment making the same point.
lindley 07.27.04 at 1:34 pm
This might be a naive question… but, in the paper, what is the difference between the “creedence” Cr(.|.) formulation, and something like “subjective probability?”
That is, when the paper says that proposition (##) is “for all Phi, Cr(Phi | frequency(Phi)=x) = x”, how is this different from a subjective belief, with some basic exchangeability argument?
And doesn’t that make the First, Second, and Third “Interpretations” just refutations of a mis-understanding of conditioned probabilities? (i.e., Pr(x | y) means you’re conditioning on y, and that Pr(x) is conditionally independent, given y, of _all_ other ‘knowledge’ you may possess?)
Scott Martens 07.27.04 at 3:33 pm
There are a number of problems with the question of whether or not the universe is a simulation, but they’re not so easily dismissed. First, I originally heard the question a little differently: Imagine that computer technology progresses and we become able to simulate the universe. We then do so, for entertainment and scientific value. Inevitably, we do so repeatedly. What are the odds, given these stipulations, that we are living in one of the simulations?
It’s a goofy question, but I started thinking about it.
First, Daniel Wolpert proved back in the 80’s that an accurate simulation of any space which contains the computer doing the simulating must run slower than the real universe. This follows as a correlary of the Halting Problem, but it applies to any type of physically realisable machine, not just Turning machines.
Thus, simulating an entire universe within a universe of comparable size possessed of the same laws of physics – and thus the same ability to build computers within subspaces of the same size – means that the simulated universe must run more slowly than the real universe. In the limiting case, the simulated universe runs on a computer which takes up the entire universe, and may therefore approach running just as fast. But if so, no one can live outside the simulation.
Alternately, if the simulated universe has different laws of physics, which either forbid the construction of computers with the same level of problem solving ability as the simulators, or which allow computers of identical power, but requires that they take greater volume to build, then the real universe is measurably different from the simulated one. It has different physics. In that case, either we live in the real universe or we live in a simulation built by aliens – we might as well call them gods – but there is no chance that it is either the one or the other, because we can know that there is a real difference between the two, and that if we knew what that difference was, we could know which we were in. The simulation is not indistinguishable from the real thing, it’s just that we can’t be sure we’re not in the simulation and have never seen the real world.
The same applies to the case where the simulated universe is smaller than the real universe. Once again, there would be a real difference between the two, even if we don’t know which one we’re in.
Lastly, I considered one possiblity for how one might distinguish a simulated universe from the real one: physics at different scales. A simulated universe can not have scale-invariant laws of physics. A real universe might. Scale-invariant laws of physics would require an infinite amount of computing power to simulate.
The existence of scale-invariant laws of physics is a scientific theory in the Popperian sense. It is not a metaphysical supposition. It would be easy to falsify: identify a scale at which the laws of physics change, so that no new theory could account for physical activities at both the macro and micro scale. In a simulation that uses finite computing power and runs in finite time such a level must exist.
The contrary, however, is harder to show. Scale variant physics is a necessary property of simulated universes. It is not necessarily true that real universes must have scale invariant physics.
Daniel Elstein 07.27.04 at 5:43 pm
I think that you’re wrong about indiscriminability being asymmetric. Your example is not convincing, because I’m inclined to say that if you really can’t tell that you’re drunk when you’re drunk then you can’t tell that you’re sober when you’re sober either. I take it that the moral of the example is just that we can tell when we’re drunk (though it may be a little tricky) just as pace Descartes we can tell when we’re dreaming.
It seems that the substantive point in your argument is meant to depend on some form of epistemic externalism – when you say that as material beings we have the ability to form reliable perceptual beliefs. I thought that no such strategy had a hope of defeating sceptical arguments precisely because of the indisciminability point – that was (I thought) what Edward Craig showed in his article ‘Nozick and the Sceptic’ (Analysis 1989).
Brian Weatherson 07.27.04 at 9:17 pm
On John’s point, the idea here is that Descartes was engaged in the wrong project. We simply don’t have to be able to justify our beliefs to a determined sceptic in order for it to be true that they are justified. Descartes presupposes that we do, but there’s no reason for this, and nothing but trouble seems to come from accepting it.
I take it Bostrom’s argument is meant to show something stronger. It is meant to show that with just making ordinary presuppositions about the actual world we can call into question our belief that we are material beings. And that’s what I’m arguing against. And that’s why I’m fully prepared to use the fact that we are actually material in my defence. If I was trying to argue with the sceptic this would be question-begging. (Arguments with the sceptic seem to always end up question-begging.) But since I’m just objecting to a particular argument, I’m allowed to use assumptions I bring to the table unless and until the arguer has shown me they are false. That’s why, pace Thomas, it’s not question-begging _in this context_ to say that I have mostly reliable beliefs. Sure they aren’t perfect, but they are way better than the guy in the simulation.
I wasn’t making the assumption Monica thinks I was making. Here was the argument there less compressed. Assume, as I think is plausible, that nothing in the universe has experiences from which mine are indiscriminable (by me). And assume, as I’m going to do until someone gives me a proof that I shouldn’t, that I’m material. Then the proportion of beings in the universe with experiences from which mine are indiscriminable is 1. So there’s no obvious statistical argument that I should be uncertain whether I’m material unless we start filling up the reference classes with beings that I know I am not. And there’s no reason to fill up the reference classes that way. (And if we do there are so many ways to fill them up that we get inconsistent results – but that’s another story.)
There’s a simpler case to show that indiscriminability is asymmetric. I can tell, just by looking and feeling, that I’m not a rock. I can discriminate my experiences from those a rock has. (I’m assuming here that rocks don’t have conscious states. If not we have to *really* rethink our ethics.) But a rock can’t discriminate anything, hence it can’t discriminate its state from mine. Now some might say this is a special case because the rock has no discriminating power at all, but I think it’s just an extreme case of a fairly standard kind of situation.
On John’s point, the worry here isn’t that the simulees experience is ‘coarser’ than ours. The worry is that thinking about what we have in common with simulees underestimates how much information we have about the world. There’s a picture Bostrom (and Descartes, and many other philosophers) are working with that our evidence about the external world starts and ends with our phenomenal experiences. That’s just what is being denied here. (Daniel is right that it’s a kind of externalism, though more like Austin’s and his followers than Nozick’s.) There’s a difference of type between us and the simulee – we can see tables, chairs and beer mugs and they can’t. And since what one can see is part of ones basic evidence about what the world is like, we have a different kind of evidence to what they do. (If, as always, we are actually material. If we are simulations then all bets are off.) Since we aren’t even in the same kind of epistemic situation as the simulees, making more and more of them, or making hierachies of them, doesn’t particularly change *our* epistemic situation.
Matt 07.28.04 at 1:10 am
To counter the second part of your argument, wouldn’t one only have to change the description of a state as simply “indiscriminable from” another state to, “is indiscriminable from, and lacks the ability to recognize this indiscriminability?” Take your analogy as an example: at 2 martinis, I can recall having 2 martinis, but at 4, I may recall having had 2 martinis, or 3, or 5. I have no idea that my memory abilities are (temporarily) diminished. Now imagine that we are, in fact, simulations. It might be the case (and, I suspect it would be easy to argue that it is the case) that we a.) would not be able to tell whether we are, in fact, simulations or materials, and b.) would not be able to recognize that we cannot actually determine the facts of the situation. This would have the effect of making our belief, and our belief that we can distinguish between the two states, irrelevant to the question of whether we are in fact material.
Brian Weatherson 07.28.04 at 2:42 am
Matt, your (a) and (b) are true, but why do they matter? We in fact can tell material things from simulations, so why does it matter that people with _very different evidence_ to ours cannot tell this. If you thought evidence was all phenomenal and they are in the same phenomenal state we are I could understand the worry. But why think that?
John Quiggin 07.28.04 at 8:58 am
Brian, I’m not clear about the claim here. You are asserting that we know we are material beings. If so, it seems obvious that we are not simulees and no probabilistic argument is going to change this.
But what about the simulees? It seems to me there are two possibilities
(i) they know they are not material
(ii) they falsely believe they are material [unlike the simulees they themselves have created]
Are you asserting one of these, or something else completely?
matt 07.28.04 at 1:45 pm
Evidence may not be phenomenal, but belief is, is it not? In which case, our belief both about the evidence and about the implications of that evidence could be identical to those of simulations. In which case, it’s still relevant.
Monica 07.28.04 at 2:00 pm
“assume, as I’m going to do until someone gives me a proof that I shouldn’t, that I’m material…So there’s no obvious statistical argument that I should be uncertain whether I’m material unless we start filling up the reference classes with beings that I know I am not.”
If we assume you are material, then you can show that you are material? What an odd argument. Or perhaps I need to go read your paper.
Jamie 07.28.04 at 5:18 pm
If we assume you are material, then you can show that you are material? What an odd argument.
Brian is not trying to demonstrate that he is a material being. (If anyone doubts that he is one, I am willing to swear out an affidavit.) He is responding to Bostrom’s argument. Bostrom’s conclusion was not that Brian is immaterial.
An argument to the conclusion that Brian is a material being would be questionbegging if it took as a premise that he is one. That’s obvious. It is not obvious that it is question-begging to use that premise in the present argument. It may be, but it’s not obvious. That would take some showing.
Comments on this entry are closed.