In 2003, Swedish philosopher Nick Bostrom proposed that it is overwhelmingly probable that we are in fact living in a computer simulation. This is a hypothesis that, tapping into the current age of AI mania we seem to be living in, has gained in popularity and garnered robust support from a number of intellectual heavyweights. The interesting thing about the argument is it makes absolutely no appeal to evidence of any kind. Instead, it derives its conclusion based purely on probability. If this immediately strikes you as flimsy grounding for a hypothesis, let alone a ‘scientific’ hypothesis, you aren’t alone. In this article, after outlining exactly what the simulation hypothesis claims, I will proceed to argue that it tells us absolutely nothing about whether we are living in a simulation or not.
The simulation argument revolves around three propositions. Either:
(1) Almost no civilisations survive to a post-human, highly technologically advanced stage, or
(2) Almost no post-human, highly technologically advanced civilisation has any interest in creating realistic computer simulations, or
(3) It is almost certain that we are living in such a simulation.
For (1) to be true, there would have to be some kind of limiting factor on almost all civilisations. Perhaps it is almost always true that technology develops faster than a culture’s ability to control it, so some significant percentage of all civilisations bomb themselves back to an agrarian way of life before developing the technological capacities to create truly realistic computer simulations. Maybe in order to develop the technology to build such a simulation, a civilisation has to experiment with AI, and 95% of the time something like an AI-controlled paperclip-producing factory ends up taking over the planet (In this scenario, the AI doesn’t have any interest in building conscious simulations because it may not be conscious or perhaps just doesn’t value it). (1) definitely isn’t impossible, but I agree with Bostrom that it’s hard to imagine any feasible scenario in which it is more often true than not, let alone almost always true.
On the other hand, (2), I think, isn’t quite as certain as Bostrom assumes. For the moment though, just to see how the argument plays out, I won’t question it here.
If Bostrom is right that (1) and (2) aren’t likely, we can now formulate the argument proper:
Premise 1: Some not insignificant percentage of civilisations become technologically advanced
Premise 2: Some not insignificant percentage of technologically advanced civilisations desire to create realistic computer simulations inhabited by conscious, digital beings
Premise 3: A single advanced civilisation is capable of creating as many simulations as they desire
Premise 4: Based on premises 1-3, there will be many more simulations of worlds (Bostrom talks about numbers in the millions) than, what we can call for convenience, real worlds
Conclusion: Therefore, based on the laws of probability, proposition (3) in the original formulation is true; i.e. we are almost certainly living in a simulation
After having read that, the chances are pretty good that you found yourself agreeing all the way through the argument until you came to the conclusion, which jarred you back to reality (or should that be back to the simulation?). In this, the simulation hypothesis reminds me of the ontological argument for the existence of God, the premises of which, each seem plausible and innocuous on their own, but which lead to a less than appealing conclusion, forcing you to go back through the argument and see what could have gone wrong. That is exactly what we will do now.
(1) Simulating Consciousness
The first, and probably the biggest, weakness of the simulation hypothesis is a hidden assumption that the whole edifice rests on; namely, that it is possible to simulate consciousness in a computer. It is far from certain that this is possible (there is certainly no evidence (scientific or otherwise) in favour of this), in fact, there are legitimate reasons for thinking that it isn’t, and yet, it is absolutely crucial to the success of the argument. If consciousness turns out not to be realisable in a computer program, the simulation hypothesis fails at the title.
I have written about this in other places, so I’ll restrict myself to a brief discussion here. The intuition guiding the AI prophets on this idea of conscious programs seems to rest on two main planks; (1) the idea that brains are just fancy computers, and (2) consciousness is purely a result of information processing. It still surprises me at the way the legitimate brain-computer analogy has been illegitimately misinterpreted as a relation of identity. ‘Brains are like computers’ has somehow, seemingly without people noticing it, become ‘brains are computers.’ The first statement seems as obviously true to me as the second one is false. The implication of the confused notion that brains are computers is that whatever brains can do, computers can also do, as long as we program them in the right way, never mind that no one has any idea what that ‘way’ might be.
The other plank, that consciousness is pure information processing, seems just as mistaken to me. Don’t get me wrong, it might be true, but at this stage, assuming we are still in the realm of science fact here, as opposed to science fiction, this has yet to receive even the slightest hint of a confirmation. The fastest supercomputer on the planet, despite being able to process information millions (billions? trillions?) of times faster than a mere human brain, is no more conscious than a rock. Computers can learn to play chess and go (the game, not the verb) on their own, devise strategies that no human has ever thought of, beat our best representatives, and yet they don’t care one way or the other because they have no consciousness of what they’re doing. Ah, that’s because the information in these computer programs isn’t integrated in the right way, they tell us. There’s that phrase again; the one that doesn’t seem to mean anything – in the right way. What way is that? The way that human brains work? Hmm, nothing circular there.
(2) Desires of a Post-human Civilisation
Bostrom rejects the second proposition, that a post-human, technologically-advanced civilisation wouldn’t be interested in creating life-like simulations, without a single word of discussion (at least in the Sam Harris podcast episode I listened to), as if it were obviously false. I disagree.
Let me just put things in perspective a little. The post-human, technologically-advanced civilisation Bostrom is talking about here isn’t a couple of hundred years down the track. He’s talking about a time when we have the resources to turn an entire planet into a computer. Considering we are still working on getting a single human to another planet, your guess is as good as mine as to when this will be possible, but one thing we should be able to agree on is that it won’t be anytime soon. Now, given this post-human civilisation will be so far beyond us (the hint is in the name – post-human; i.e. not even human anymore), I find it pretty presumptuous on Bostrom’s part to think that he can articulate what such a race of beings would want to do with their staggering computing power. It’s like a caveman thinking about his future descendants (us) and concluding that with all our massive intellects and technology, we must live in the biggest caves and have the biggest fires to keep ourselves warm at night. Now, there’s no harm in speculating; indeed that’s precisely what makes science fiction so interesting, but Bostrom isn’t writing a novel here, he’s proposing a supposedly scientific hypothesis. I find it a bit of a stretch that a hypothesis that relies on predicting the desires of a race of beings so far beyond us that we don’t even think of them as human anymore can be scientific in any way.
One possible objection to my caveman analogy is to suggest that actually we do live in the biggest caves and have the biggest fires; we just call them houses and central heating. While the caveman couldn’t have predicted apartment buildings and electrical heating devices, he could have predicted that we would still want to protect ourselves from the elements and, with greater intellectual resources at our disposal, would be able to devise improved ways of accomplishing this. There are three quick rebuttals I would offer here. First, this has nothing to do with the simulation argument. Bostrom isn’t arguing that a technologically advanced civilisation will build on current technology in some way we can’t predict; he’s arguing for a specific form of technology, i.e. computer simulations. This is the caveman predicting massive, blazing fires that reach as far as the eye can see. Second, it strikes me as a little disingenuous to argue that the caveman’s ‘prediction’ of a massive fire has actually been realised in central heating. Yes, our central heating serves the same purpose as the caveman’s fire, but central heating is not a fire. The caveman would be amazed beyond words to see how our ‘fires’ work. Why? Because he didn’t (and couldn’t possibly) have predicted anything even remotely similar. Third, this objection suggests that we are forever condemned to merely regurgitate the same fundamental behaviours again and again forever, merely in different garb. FaceBook is just a way to communicate with others, the Internet is just a means of gathering information, apartments are just to protect us from the elements, etc; all activities we have been engaged in for hundreds of thousands, if not millions, of years. ‘Taxonomising’ human behaviour like this is, of course, possible, and does seem appealing from a reductionist, objectivist viewpoint, but, as with all taxonomies, the ‘higher’ categories obscure as much as they clarify. To say a snake is just a part of the animal kingdom isn’t to have said very much about a snake. In the same way, dismissing FaceBook as a tool for communication, as if it’s merely an extension of two caveman grunting a greeting to each other as they head down to the waterhole for a morning wash, is clearly a dramatic and unhelpful simplification.
(3) Real Worlds
Bostrom’s argument essentially divides ‘worlds’ into two categories; simulated and real. Since, there can be many of the former (Bostrom talks about numbers in the millions here, which seems more than a little intellectually reckless to me) and only one of the latter, the odds that we would find ourselves in the lone real world are astronomically low. Reasoning like this has failed to ask just what we mean by ‘world’. To say, “There is only one real world,” is to implicitly suggest that we are talking about what we usually mean when we use the word ‘universe.’ This can’t be right though, because the argument is centred on civilisations, of which there may not be only one in our universe. How many civilisations are there in the universe? Well, no one knows. Given the size of the universe, we certainly can’t rule out the fact that there could very well be many; one might even say this is probably true. What’s more, of all of these civilisations, the vast majority are likely less technologically advanced than Bostrom’s post-human super-race, meaning they won’t have contributed to the simulations. But it gets worse.
Just before writing this article, I happened to be reading Max Tegmark’s Our Mathematical Universe, in which he spends the first few chapters talking about the current state of cosmology. One genuine scientific idea seems to refute the simulation argument quite nicely; what Tegmark calls, the level I multiverse. A couple of quick definitions before we go on:
- Inflation is the theory that in the early stages of things (before the Big Bang, in fact) space underwent an exponential growth, doubling in size at regular intervals. Because of this kick start, space is still expanding (we have measured galaxies receding from us as confirmation of this).
- The universe is the region of space we have access to, i.e. the spherical region of space (about 14 billion light years in radius) from which light has had time to reach us.
Since the volume of space created by inflation is continually expanding and the speed of light is finite, there are vast regions of space beyond our ‘universe’ we have never seen. These spaces constitute Tegmark’s level I multiverse. Because they are beyond our perception, or outside our ‘universe,’ Tegmark calls them parallel universes. They are in the same volume of ‘space’ as our universe, have the same laws of physics, and are not unreachable by any means, we just have to wait for the light from these ‘universes’ to reach us.
What does this mean for the simulation argument? Well, it changes the calculus (which is a big deal because the simulation argument is nothing but a calculation of probability), specifically, by increasing the number of ‘real’ civilisations. It’s true that this will also increase the number of simulated worlds, but, as I argued above, the vast majority of these worlds won’t be capable of creating simulations. Basically, for every civilisation that crosses this technological barrier, there will be (how many? Thousands? Tens of thousands? Let’s follow Bostrom’s lead and just call it millions) millions of civilisations at various stages beneath the simulation-creating level. Suddenly, it doesn’t seem so unlikely that we are in the real world.
One might try to resist this by claiming that I cannot include any and all random civilisations in my calculus. This is only concerning our own civilisation. The argument is specifically that a post-human civilisation would create multiple simulations of their own history (i.e. us). In this more circumscribed context, the odds of us being ‘real’ are slim.
The problem with this is that there is no logical reason to believe any of it. Why would an advanced civilisation restrict itself to creating simulations of its own history? If we are a simulation, there is no reason why we couldn’t be a simulation of a race of beings that never looked anything like us. In fact, wouldn’t it be more interesting to investigate different ways life could have evolved? There isn’t even any reason to suppose the laws of physics for the creators of our simulation are the same as ours. Again, wouldn’t future scientists be more interested in these types of simulations? Attempting to ‘localise’ the argument like this just doesn’t work.
Another argument a Bostrom defender might employ here could be to say that all of our cosmological theories are bankrupt if we actually are in a simulation. Rather than describing reality, we are simply describing how the simulation is set up. Obviously, this is circular reasoning. If you are arguing that we are probably in a simulation, you can’t reject evidence that appears to contradict this thesis on the grounds that we might be in a simulation.
(4) Self-Referential Probability
The simulation argument depends entirely on probability. There is absolutely no concrete evidence in its favour. This is why it is so hard to refute; the parameters are so broad (an infinite future, unlimited technological capabilities, the unlimited perfectibility of the simulation such that is asserted to be identical to our lived reality), and there are no concrete constraints. Of course, being the purely speculative, intellectual argument that it is, also means that it lacks substance.
One weakness of an argument like this is that one is then free to leverage those same parameters to come up with similar arguments that also lack substance. I could, for example, ask, “What is the probability that, out of all the simulations a race of highly-advanced beings could create, we would happen to find ourselves in one that has precisely the history ours has?” The odds of this are obviously astronomically low; therefore, it is unlikely we are in a simulation. I bet on reading this you think something is wrong. But what exactly?
The problem, it seems to me, arises when we attempt to think probabilistically about our own situation. One sometimes hears people make a similar mistake when they reflect on the unfathomable number of random chances that led to their personal existence. “What are the chances,” they say, “that everything happened just the way it did so that I was born. I mean, if my great-great-great-grandfather had crossed the street two seconds earlier that day, he wouldn’t have met my great-great-great-grandmother, and I would have been done for. And there were millions of these moments that could have been different but weren’t. Amazing! It must be fate!” Of course, the outrageously unlikely sequence of events that led to you being born only appears outrageously unlikely after you have been born, as you look backwards, in effect, presupposing the existence of the very thing that precipitated the wonder; i.e. you. The wonder only arises from having already assumed a particular perspective.
Returning to my argument against the simulation hypothesis then; to even begin to inquire into the probability of my particular situation (i.e. a world with this particular history) is already to have reasoned illegitimately. If one doesn’t stop the argument here, nipping it in the bud, so to speak, whatever conclusion you draw will be irrelevant because the whole thing is abstract nonsense, conceptual chicanery. It’s a show which reveals more about our concepts and language than about the real world. Talk about the probability of finding oneself in any kind of world always seems to lead to gibberish. The only reason I can think that this would be the case is because the whole scenario is so abstract and contrived, that it just doesn’t relate to anything meaningful.
Either way, I’m arguing here that my counter-argument against the simulation hypothesis is invalid. The point though, is that Bostrom’s argument is of exactly the same type. If my counter-argument is flawed, not because of the specific details, but because of the broader framework it assumes, then so is Bostrom’s. We can rephrase his argument to better highlight the correspondence between them; “What is the probability that, out of the sum total of all possible worlds, including both simulated and real (of which there may only be one), we would happen to find ourselves in one that is real?” If you rejected my counter-argument above (as I suggested you probably did), you must also reject Bostrom’s.
(5) The Religion Analogy
Several times in this article, I have pointed out how scandalously speculative the simulation argument is. From assuming that a simulation full of conscious beings is possible, to claiming to know how much computing power is required to produce it, to asserting we can have any insight at all into the motivations of a post-human civilisation almost unrecognisably superior to us; the unrestrained speculation involved here makes the whole thing an exercise in science fiction masquerading as science. While thinking about this one day, I was struck by the idea that this argument actually runs parallel to another set of arguments I have spent a lot of time rejecting; namely, religious arguments that try to prove the existence of God and/or an afterlife.
Think about this for a moment. Religious believers assert the existence of an all-powerful, all-knowing, omnipresent being who created our universe. He (or she, but usually ‘he’) exists in some fanciful realm beyond time and space in some way no one can understand. There is no concrete evidence that this being exists, but there are a number of highly contrived, speculative, or circumstantial arguments that people like to claim as ‘evidence.’ Finally, death isn’t the end. Instead, it marks the beginning of a new life, which one qualifies for, by pleasing God.
Compare this to the simulation argument. A being (or group of beings) have created our universe as a simulation in a computer. As such, at the press of a button, they can gain access to any part of our universe and manipulate anything any way they choose (even suspending our ‘laws’ of physics thereby performing what we would call miracles). They exist beyond our time and space in a universe that may be governed by laws similar to ours or that may be completely different. There is no concrete evidence for any of this, but a handful of highly contrived, speculative, or circumstantial arguments that people claim as ‘evidence.’ Finally, death in this simulated universe where we are just lines of computer code, is presumably as easy to overcome as pressing Ctrl + C, Ctrl + V and starting up another simulation. This next life isn’t guaranteed, but presumably, if one can become a ‘favourite’ character of the creator(s), your odds of being re-booted will go up.
How does one argue against a religious believer? One points out certain facts that seem incompatible with the existence of an all-good God; the problem of evil, human vestigial structures (which evolution explains but seem inexplicable if we were designed), etc. One highlights the many problems/barbarisms in the Bible. One appeals to common sense. And so on. Yet, the religious believer is seldom convinced. Why? Because they aren’t restrained by things as inconvenient as common sense and reality. Theirs is a comforting (in an existential sense, at least) world of magical beings where the ultimate being has unlimited power. Take the problem of evil. The believer can assert that actually there is no problem. What we perceive as evil only appears to be so from our limited perspective. Or they can take the opposite route and claim that evil is wrought by humankind. God has left us to our own devices as per our wishes, but He’s waiting in the wings ready to re-enter and when He does… look out! Even if you should happen on a rock solid argument that leaves them stumped, they have one last unbeatable ace up their collective sleeve; “We petty humans can’t possibly understand the mind of God.”
Likewise, you could marshal your best arguments against the simulation hypothesis, but you will never succeed with common sense and perceptive observations about reality because the argument is impervious to all such assaults. The underlying premise – that a simulation would be indistinguishable from reality – guarantees that there is nothing you could possibly observe which would render the hypothesis false. This makes the theory non-falsifiable, and a non-falsifiable theory isn’t a theory at all. If there is no conceivable evidence that could disprove a theory, there is no evidence that can prove it either (it goes without saying, all religious doctrine is non-falsifiable).
Theoretical physicist Sean Carrol sceptically asks why these technologically-advanced beings would have created such a detailed and complex simulation. That’s easy. Why wouldn’t they? This universe seems detailed and complex to us, but to them it’s nothing. Or equally effectively, the simulation adherent can argue that our universe isn’t detailed. Look at what happens when you get to the deepest levels of reality. There seems to be nothing there! Certainly nothing resembling particles, at least. Max Tegmark argues that fundamental reality is ultimately mathematical; not just describable with mathematics, but nothing more than mathematics; i.e. reality is fundamentally, and literally, insubstantial. Do you see the problem? Whether the premise of the question is true or false, the simulation theory adherent can answer satisfactorily either way. Any theory that can do this is worse than useless.
(6) Occam’s Razor
My last argument is that the simulation hypothesis is a flagrant violation of Occam’s razor, which basically says we ought not to multiply entities beyond necessity. If you can explain something without invoking entities X, Y, and Z, then, all things considered, you would be better off removing them from your theory. It will come as no surprise that this guideline is often used against religion. Once a more sophisticated understanding of cosmology gave us the means to explain the formation and motion of stars and planets, and evolution by natural selection uncovered how humans came to be wandering upright about the plains of Africa brandishing those opposable thumbs, God was no longer a hypothesis we had need of. Get rid of it.
The irony is that almost every single simulation argument defender, while fully endorsing the usage of Occam’s razor against religion, suddenly and inexplicably seems to forget the wisdom of William of Occam right when they need it most. This universe can be explained more efficiently without appeal to a realm of highly advanced beings living in a base reality we, by definition, can never experience (unless they should take pity on one of us and upload us into some type of body (artificial or otherwise) in their reality). Given this state of affairs; rather than elevating the supposition to highly probable, we ought to do the exact opposite and jettison it.
Okay, let’s quickly recap. I disagree with the simulation hypothesis for six reasons:
- The assumption that we can create genuine consciousness in a computer simulation remains completely unproven.
- Bostrom’s rejection of his second proposition, that almost no post-human, highly technologically advanced civilisation would have any interest in creating realistic computer simulations, is difficult to justify and is really no more certain than the assumption about consciousness.
- Current cosmology predicts that there is a vast region of ever-increasing space beyond our ‘universe’ which could quite conceivability harbour many civilisations in ‘base level’ reality, changing the calculus in such a way that it is no longer clear that simulations outnumber ‘real’ worlds.
- These types of purely probabilistic, highly speculative, evidence-free chains of reasoning that reference our real situation are so empty of substance that they can’t meaningfully conclude anything about reality.
- The simulation argument shares many uncomfortable parallels with religious arguments defending the existence of God. Inasmuch as you are unconvinced by the latter, you shouldn’t be convinced by the former.
- The simulation argument violates Occam’s razor.
I’ve been pretty unrelenting here in my criticism of the simulation argument, but I’d like to close by making it clear I am not against this type of speculation. Indeed, I devour the speculation one finds in futuristic science fiction novels and movies, and relish the philosophical questions they raise and the insights they yield. The problem I have with the simulation argument is that it has crossed the line from genuinely interesting philosophical speculation into pseudo-scientific, ungrounded assertion. Rather than positing the interesting notion that we might be living in a simulation and exploring what this could mean for us, it appeals to tenuous, unscientific, unphilosophical reasoning to declare that we almost certainly are. The latter gets you more media attention, but any gains there are outweighed by what it loses in intellectual respectability.