Why the Simulation Argument is Invalid

In this philosophical piece, I will try to summarize the main reasons why the so-called simulation argument a la Bostrom is most certainly invalid. I will show, from the perspective of science and empiricist philosophy, that it hinges on some rather unreasonable assumptions, and unscientific thinking, and therefore it must be placed in the same category as mythology.

Rarely do I speak with such certainty. I have spoken against only dualism, non-reductionism/creationism with such certainty in the past. However, as I will show, this argument clearly is a form of creationism and therefore must be dispensed with in the same forceful and rigorous manner as we treat other forms of creationism. Unfortunately, these “arguments”, purported to be “good philosophy” by many gullible thinkers, are not respectable arguments for a serious philosopher. Rather, they are superstitious drivel and metaphysical nonsense redressed in scientific-sounding jargon and mathematical-looking notation, making a mockery of both science and philosophy, and they must be regarded as a source of embarrassment by any intellectual worth his material.

On the surface, the Simulation Argument (SA for short) looks like a disjunctive argument that talks of three possibilities: first is, civilizations don’t reach post-human stage, second is, post-human civilizations aren’t interested in ancestor simulations, third is, we are most likely living in a simulation. Of course, this is all there to make the argument look sensible. In fact, I believe Bostrom wants to disregard the first two possibilities, convincing us that they have low probability and make us believe that the third possibility is all there really is. Or more precisely, he wishes to sneak in an extremely improbable scenario along with probable scenarios, insidiously elevating its ontological status, which is a classical tactic of deception. The latter two possibilities constitute a false dilemma, let us analyze why.

Hence, the correct reading of the argument shall follow. SA claims that in the future post-humans will be so powerful, that they will be able to create realistic physical simulations that can contain Earth and its inhabitants. And they will create ancestor simulations because somehow ancestor simulations are common/crucial/unavoidable in a technological civilization. Therefore, the argument goes, it is low probability that we will be the original post-humans. We are most likely, according to Bostrom, in an ancestor simulation within an ancestor simulation… and so forth. Of course he plays it down, he actually gives a 1/3 chance that we are almost certainly in a simulation. Though regardless, that’s a ridiculously high probability given to a mythological scenario as we will see.

Let us review the premises before we get to deeper mistakes and methodological errors common with this sort of ontological argument. There is, first of all, absolutely no reason why the first premise should not be true. Like many bad arguments in philosophy, SA takes advantage of a gray area: let us start with what sounds like a plausible premise, and then let us have the reader swallow ever more unreasonable premises and inferences. Indeed, it is possible to construct physical simulations. We have been simulating many physical phenomena for a long time. Simulation is a valid and respectable method of science. Known current theories of physics are computable, hence with enough computing speed it is theoretically possible to make such simulations. Bostrom’s assumption in the paper that it is possible to create a virtual reality prison somewhat like in the movie “The Matrix”, of the entire Earth is acceptable. Such a computer would likely require enormous resources, but it is possible given access to stellar energy sources in the future, given energy-efficient computers, and energy drawn from blackholes, etc.

That is where the plausibility of SA ends, unfortunately. To begin, his claims about the ease of such computations is entirely unfounded scientifically:

We can conclude that the computing power available to a posthuman civilization is sufficient to run a huge number of ancestor-simulations even if it allocates only a minute fraction of its resources to that purpose. We can draw this conclusion even while leaving a substantial margin of error in all our estimates.

The suggestions in his analysis are all far off. They disregard how much information the human brain processes and the resolution of our perceptual and scientific apparatus, and what sort of complexity would be ultimately distinguishable to human observers, leaving out obvious artifacts that would show the world to be a simulation. In present day, some uninitiated thinkers seem to think that if digital physics turned out to be true, this would be evidence for SA, which I argued against separately. However, that is not all there is to say. One cannot argue on the basis of estimates made by people who are not experts in theoretical computer science and theoretical physics. The obvious thing is that quantum events work everywhere, we have our atomic microscopes and various quantum devices, and for those to work you need a complete and consistent physical simulation. And no, you could not simply try to monitor all brains and prevent them from making science, which would be just as complex. Perhaps, Bostrom suspects that theologians are here to prevent us from ever advancing science so that we cannot find the bitter truth about our world, like the agents in “The Matrix”. There is no end to delusions after that leap of faith.

Bostrom attempts at a probability calculation of whether a human observer is in a simulation. This calculation is quite simplistic, ignoring almost everything we know about cosmology and physics, and involves these quantities: fp: fraction of technological post-human civilizations, navg: average number of ancestor simulations run by a post-human civilization, havg: average number of “humans” that lived in a civilization before post-human stage. The probability that we’re in a simulation is then (fp.navg.havg)/(fp.navg.havg)+havg. Now, this rather looks like something that you could encounter in a secondary grade probability class. It seems that we know nothing about the concept of “probability distribution”, and we think we can forecast such an unprecedented probability by estimating a few deterministic variables directly in a simple probability calculation. Obviously, this was inspired by Drake’s equation, but Drake tried to base his estimate on scientifically plausible forecasts, and involve as many relevant variables as possible. On the other hand, here Bostrom tries to make us ignore relevant other variables and conditions, and make us believe that because there are (according to him) many more human individuals that should live in ancestor simulations made by posthumans, than original humans, that we should be almost certainly living in one. Why should there be many such simulations? Because Bostrom thinks there will be plenty of computational resources. Therefore, some of these computational deities will be, out of curiosity or for recreation, perhaps, be making such simulations. That is, Bostrom thinks such simulations will be relatively cheap. Well, this is the second part that is troublesome. Such simulations are never going to be too easy to program or run. Simulating quantum systems requires a lot of computational speed and memory, that cannot be easily shrunk down as Bostrom thinks (his first major mistake). A universal quantum computer of that size would cost a lot of energy and space-time no matter how you design it. One problem here is that we do not even know if constructing an enormous universal quantum computer is possible. If not, that is a big obstacle, because simulating n qubits would take asymptotically more space and time (say it might require n squared units given current quantum computing theory). Even when you approach physical limits of computation, simulating a previous Earth, and a fake universe backdrop would take up a lot of physical resources, so much that we need not quantify. However, even on a galactic scale, that is not a negligible cost. Bostrom obviously wants us to think that this is going to be the equivalent of a video game, when it most certainly isn’t. That would be the video game to fool even extremely competent philosophers and scientists, and that is where the cost premise breaks down. Simulating a previous planet down to quantum level is simply a waste of resources that could have been put to better use. Post-human civilization will be beyond earth scarcity, but given the intergalactic distances, the local physical resources within a galaxy will be extremely valuable. Given that it is exponentially difficult to solve some scientific problems, it seems rather unreasonable to think that such computational deities would be so reckless and hedonist as to waste their resources on unnecessary computer games.

This is already an irrecoverable blow to SA, it gives the first hint that in no plausible technological civilization, there are infinite resources, and infinite means. There will still be an economy that quantifies energy and space-time resources. And at the galactic scale, such a project would still be significant, and it would require a rather good reason to start, even if we do not question its feasibility. Hence, the fans of SA provided us with a “solution” to this obvious weakness in the premises. They proposed that since Solomonoff induction requires (to a careless reader) reproducing past data with the smallest program, that future civilizations would need to recreate their past, so that they could predict and control their future! Of course Bostrom has done well to not make this part of his argument, because that is obviously false. A future post-human civilization would hardly depend on our musings here on Earth. Our entire culture and products would be like a drop of water in an ocean. And more so, Solomonoff induction would be perfectly adequate starting from a well-known future state description not involving Earth, because it doesn’t require exact simulation, it only requires fitting stochastic models. The upshot is that a successful post-human civilization would not depend on an ancestor simulation to survive, therefore such simulations are not essential. However, even more significantly, such data is certainly not available, and one cannot recover the intricate history that has been lost via “simulation” due to algorithmic incompleteness; that data is truly random.

This brings us to the plausibility of the probability calculation. The probability calculation fails precisely because it is a variation of intelligent design. In his paper, Bostrom spends quite a bit of time to persuade us that:

if we knew that a fraction x of all observers with human-type experiences live in simulations, and we don’t have any information that indicate that our own particular experiences are any more or less likely than other human-type experiences to have been implemented in vivo rather than in machina, [then the conclusion should be correct]

Bostrom is even careful to set his argument apart from the typical eschatology nonsense. However, his reasoning should still be treated with the same well-deserved suspicion, because it is the theory of evolution itself that suggests that we predict evolution to be much preferable to any variation of intelligent design. Therefore, from a scientific point of view, it is definitely not the case that the so-called bland indifference principle applies here. The principle is not merely bland, it is also mentally handicapped and missing both legs.

Inductively, this reasoning simply does not work. You cannot prove intelligent design by claiming that there might be a God that designed our world, when there are much better scientific theories as to how our universe, solar system, Earth and its inhabitants evolved. But, would not intelligent design (or its variation SA) explain how we are all here? Yes, it would, but it would do no better job at that than anything that the Bible says. That is to say, given what we know, that is, the entirety of human scientific knowledge which is gleaned through much rigorous thinking and experimentation, the hypothesis of evolution is much more simpler compared to the hypothesis that some God designed this world, and therefore has much higher a priori probability.

In fact, this is exactly what the theory of induction tells us; Occam’s razor tells us to pick the simplest theory among those that explain the same facts. Sure, that our Earth is a video game would explain our observations. But it would not be a good explanation because it is a much more complex explanation than that of evolution which simply requires laws of physics to work. What is the numerical sense here? The laws of physics can likely be encoded in under 1000 bits. Possibly less than that but we are exaggerating for the sake of argument. In reality, we know much shorter universal codes, digital physicists claim very compact universal machines to encode the entirety of physical law, which would amount to under 100 bits. However, an intelligent design scenario, that is a world designed by an intelligent agent smarter than a human requires an initial state of at least the complexity of human brain and civilization required to get it to post-human level, which is many orders of magnitude larger than 1000 bits. Looking at the genome, an estimate of 100.000 bits would be even very conservative, again just for the sake of argument. The evolution theory is then 2^{100.000}/2^1000 times more likely than any intelligent design scenario, which is of course a much more scientifically grounded back-of-the-envelope calculation than Bostrom’s theological fantasies. The careful reader should have noticed that the bland indifference principle does not apply here. Inductive reasoning is the valid method of thinking no matter what circumstance you start from. Such deductive reasoning simply cannot be assumed, especially if they are based on superstitious assumptions.

If we were in a simulation, this would be discovered sooner or later. However, for that one requires evidence, for extraordinary claims require extraordinary evidence, and Bostrom’s SA fails miserably in providing such an extraordinary evidence. Metaphysical arguments, in general, should not be seen as providing evidence. Here, we must point to the serious methodological error: we will not believe in the possibility of something only because Bostrom thinks so. Especially if he is referring to entities beyond the observable, which casts doubt on whether he is talking of nomological possibilities. It is not even clear if we can treat these possibilites as strictly constrained to physical law. They refer to realms that seem quite supernatural with respect to our own. By definition, such claims are not very different from claims about Valhalla, so they must be even regarded schizophrenic a priori. When a madman tells us that he is communicating with aliens from a parallel universe (with metaphysical arguments to back him up), we place him in a mental institution, when an Oxford philosopher tells us that “The Matrix” movie is real, we congratulate him and give him a medal because it resembles old religious superstitions. On the contrary to Bostrom’s expectations, we empiricists are obliged to scrutinize such possibility much like we scrutinize any other. In this case, it is not even squarely nomological possibility that refers rigorously to our world. It refers to another world, a fantasy realm. Such possibility must be empirically justified and scientifically plausible to be thus accepted. There are many other parallel universe cosmogenies that must be treated by empiricists before we take DMT and venture into “The Matrix” trips as it is popular in Oxford. Otherwise, we could always say that because the Bible said that the Earth was created in seven days, this was pretty good evidence, because that might just be the case. Or that in a cosmos with many Gods, every God would have a creation realm of his own, and therefore any person is likely to be in a realm created by a God. When put that way, it sounds much less credible, of course. Or that we should believe in intelligent design, because our world is too complex to exist on its own: it requires a maker, an intelligent watchmaker. The scientific consensus is contrary to that uneducated point of view, evolution requires no intelligent designer whatsoever, therefore regardless of anything else, the theory of evolution is a very good refutation of any intelligent design claim that doesn’t deserve the title of a scientific hypothesis. That is the only principle of indifference that works here: any sort of intelligent design is already inductively known to be invalid. Which is precisely why Bostrom tries so desparately. He tries to take intelligent design, and make it look like a scientific hypothesis, and by that action I believe that we must question his philosophical integrity, for in his words there is a certain intention to convince the readers that old superstitions were in fact true. I call this “pulling a Chalmers”, take an old superstition, metaphysical fantasy, or mythological tale and rephrase it in new words, and then appeal to the superstitious majority. Yes, we know that most people are not intelligent enough to understand why the theory of evolution is extremely well supported by evidence. We know that most people are not knowledgeable enough to understand how very precise our physical theory is, and what sort of consistency it has entailed, and why countless physical experiments affirm each other, increasing their likelihood several orders beyond theological claims, and even more importantly offering us a complete world view that starts from Big Bang to present day, only requiring a very small set of basic physical laws.

Here, we’ve seen another very serious flaw in SA. It ignores the completeness of the scientific world-view. There is no difference between positing super-natural entities or intelligent agents that designed our world. This is by definition intelligent design. SA is wrong also because it is just a theory of turtles all the way down. It offers absolutely no valid theory of cosmology. It doesn’t answer what the age of the cosmos is, what its extent is, how much energy there is, what the energy distribution of the cosmos is, how much entropy there is and so forth. It doesn’t explain whether there was a big bang or not. No sense of the vastness of the “real” cosmos. It only suggests that we are living in a fake world, just like in “The Matrix”, and none of this is real, that what we think to be real is just an illusion. In philosophy, this is known as solipsism. There, we have combined two ancient superstitions: intelligent design and solipsism at once. Unfortunately, SA paper neglects to mention these two well-known, and almost universally discredited metaphysical claims. On the other hand, it fails to account for the evolution of the original humans that it claims that are not us, and how they evolved. That is to say, it still disregards the most interesting question, how did the original humans ever form?

If these original humans existed, surely they should have believed in SA themselves. Why should they reject creationism? SA’s answer is that they should not reject creationism, that creationism is always a very big probability, and every reasonable person must embrace creationism. That is, these original humans were so privileged and we are so unprivileged, although they had more or less the same observations to base their conclusions on. Creationists at heart, what reason would they have to conduct any scientific work? What is this difference that is undetectable to empiricists, then? They were just so lucky! But we are very unlucky because we have terrible philosophers. Maybe, they didn’t have any philosophers that wasted their time, trying to stall singularity. Or maybe they knew they were the real Gods!

Then, we must question if we really know the total number of simulated human-like minds in the entire space-time of the cosmos. To show the absurdity of such thinking, let us grant that there will be many ancestor simulations. Let us neglect that, enormous resources would have to be invested into the design and implementation of such a great simulation that would fool every empiricist on Earth, while it channels privileged information into the hearts and minds of believers. The problem is that there is simply no plausible estimate given. How many such simulations are supposed to exist in a galaxy? In the entire universe? How large is this “real” universe? Are its resources infinite? None of these questions are really answered. What is done here is to simply suggest that the real universe is much larger than this, and we are being fooled by Gods, and as if we are all retarded, pretend that this is a new or philosophical claim. In all Abrahamic religions, exactly the same claim is made. That there are realms beyond this one, and that there is a heaven that is eternal, and so forth. How is SA any different from those fairy stories? It isn’t in principle, and any deep reflection should reveal that.

Of course, all of that simulation talk is merely a substitute for the theological nonsense that Bostrom is driving at. You could have simply replaced all the scientific sounding jargon with theological jargon and his paper would not change a single bit. SA is sadly, deeply religious and unscientific. He even proudly makes the connection to theology, explaining that simulators will be Gods compared to the simulated. One cannot but imagine that he must have experienced the same divine ecstasy as the scholastic philosophers did when they proposed their ontological arguments, reveling in their blind faith.

The only indifference principle that applies here is thus that of induction. Induction disproves intelligent design no matter what. Inductively, the probability that we are the first humans is much, much, higher, than the probability that we are of the future simulated humans. This distribution would change depending on the precision of the simulation for simulated humans in the future. The truly undetectable simulation itself would have to be as big as a universe, which is impossible by Bostrom’s own admission (strangely). Anything else is detectable. In fact, we would think that it would be rapidly detected with a cheap simulation that only simulates earth, astrophysics and space travel would quickly find the artifacts, which we have not. More sophisticated readers will note that one cannot really fake the astronomic observations at every wavelength and scale imaginable. One would quick infer inconsistencies and incompleteness. There is no simulation that would not reveal its information content bounds, or space-time bounds. That is to say, for us the probability distribution for the present isn’t uniformly decomposed, it favors the original humans, it has the shape of a power-law that gives very high probability to us being the original humans (well, no extra huge posits!) and very low probability to us being simulated entities. The probability of us being simulated doesn’t even go very far in the long-tail of this distribution, because the cost of simulation is inordinate, even in the future there would be few exact world simulations. Since post-humans would be smart, they would conserve resources to increase their lifetimes. Most likely, future scientists would be content with simulating just specimens, and or vistas. Otherwise, sadly, this caricature of an argument is not any different from the eschatology argument that since it is highly implausible that we should be “so privileged” as to be the first humans that have ever lived, we should be close to the end times! All in all, such arguments are made by people with an inadequate understanding of probability theory. You cannot make such claims without carefully reviewing dependencies to everything else that we know. Doing such is artless prophecy and not worthy of philosophical investigation. Unfortunately, the inductive probability of such things would not come out as simple as that.

In every era, creationists have pretended to understand and use scientific knowledge to prove God. This is just another riff on the same theme. The proponents of intelligent design tried to insert their ridiculous views as if they were scientific theories into textbooks everywhere in the world. How is this any different? If we really lived in a virtual “Truman’s World” or “The Matrix”, this would be detectable. Something would be sorely missing from our world-view, something fundamental would be inexplicable, forever shrouded in mystery, quite possibly there would be much disconnect in our knowledge. Perhaps, that is how non-scientists like Bostrom experience the world. They only have a superficial understanding of the vastness and unification of scientific knowledge, they feel that much about our existence has not been explained, and the bits of explanations they see, they are quite suspicious of, and hence they think it is possible that all of this might be a game. They obviously have not liberated themselves from ancient superstitions. It is easy to understand the psychological factors involved in creationism. Science is too much for the faint hearted. It would be much more convenient if we knew this world to be just a game, not the real deal. Then, we would not have any real responsibility, or any real purpose, or any real values, or any real death, even. We would be just characters in a supreme God’s sadistic video game. Our whole existence would just be an illusion. This is the basic psychosis of the solipsist: his mind is not strong enough to face reality. The Cosmos is simply too big and too scary for him. Therefore, he believes that he or someone else is making this all up.

For those that might have intuited correctly, the only way such an argument could be made probable is by decreasing the complexity of the “designer”. So much that its complexity is more or less the same as the laws of physics. It could even be simpler. I recall Schmidhuber call this scenario a “bacteria god” (with a small “g”), a mindless procedure that simply generates and tests many child processes. However, note the huge difference here, this is not theological nonsense, it is clearly a hypothesis of digital cosmology. Such a process would not be an intelligent agent, it would not be simulating anything that existed before, or deliberately programming anything. It is just a kind of evolution in action. It is quite reasonable that the cosmos is a computer given universal quantum computer theory. If so, then it is conceivable that very small complexity generative processes can evolve in such a system. A very important distinction, such processes are not simulation, they are just sub-processes generated in succession, it is just an operating system like feature that has evolved. Thus, each such process could be like a universe with its own physical law. One could have a 2D space-time, one could have 4D, maybe another 11D. This is of course much more reasonable than SA. Thus, by induction, if we would like to venture beyond the observable universe, this would be the hypothesis to take seriously: Juergen Schmidhuber’s Theory of Everything. Contrast freely to SA.