Do we live in a simulation? The simulation argument, formally set out by Professor Nick Bostrom, is a startling philosophical disjunctive argument which has quite profound meta-physical and epistemological implications (Bostrom, 2003). A disjunctive argument is one which has several mutually exclusive premises (disjuncts) and one of these premises must be true. One of these disjuncts suggests that it is almost certain that we are living in a simulation. I don’t think the argument affects how we should behave, but it does update our prior beliefs about our species’ future prospects. It is erroneous to draw parallels between Bostrom’s argument and sci-fi films like The Matrix Trilogy, but the paper raises the profile and underlines the importance of philosophy and existential risk research. I’m first going to set out what the simulation argument is and its underlying assumption. I’ll then discuss why the argument is well supported and present some of its implications.
What is the Simulation Argument?
As Bostrom lays out in his paper, there is a distinction between the simulation argument and the simulation hypothesis. The simulation argument is a disjunctive argument which holds that one of the following three must be true:
- Any intelligent species is very likely to go extinct before reaching an “advanced” state.
- “Advanced” intelligent species do not want to run simulations of their (or any) evolutionary history.
- We are almost certainly living in a computer simulation.
Descriptively, the argument goes like this. If any intelligent species continues to advance technologically and it doesn’t go extinct, then at some point it should have the capability to run ancestor simulations (simulations resembling the simulator’s own evolutionary history). If that intelligent species would like to run one “ancestor simulation”, then it is expected that they would run more than one. In fact, they would likely produce billions of such simulations. Therefore, it is far more likely that we are in this intelligent species’ simulated reality than base reality.
The simulation hypothesis asserts that (3) is true – that we are almost certainly living in a simulation.
Assumption that consciousness is substrate independent
The only assumption required to validate the simulation argument is that consciousness is substrate independent – that is, whatever consciousness is, there is no law of physics which confines it to carbon-based forms and forbids it from emerging upon a silicon (or other) based substrate.
This view usually assumes that consciousness is a functional property. For thousands of years, people believed that only by exactly replicating birds could humans fly. However, in 1903, the Wright brothers proved that flying is one such functional property. Today, both an albatross and a Boeing 747 can fly thousands of miles in a single trip, but they “fly” using very different mechanisms and materials. In my view, the arguments why consciousness is a functional property, which arises due to information processing alone, are very strong, but there are some objections.
For example, Christoph Koch’s theory of consciousness (Koch & Tononi, 2015), built upon Giulio Tononi’s mathematical framework, asserts that it takes a particular type of hardware to instantiate what we perceive as consciousness (“I think, therefore I am” et al). He uses the example of computer simulations displaying a star being swallowed by a black hole. Here, the computer is simulating everything contextually meaningful about a black hole from Einstein’s field equations, but the computer that runs this program won’t itself get swallowed by that black hole. Therefore, in the same way that one requires the physical as well as the informational make up of a black hole to truly instantiate it, Koch’s theory says we’d need the same cause-and-effect structure of biological neural networks to create artificial consciousness. However, other philosophers and neuroscientists draw different conclusions from Tononi’s framework and think that if we were to create an identical clone of all the connections in our brain within a digital (logic based) medium, then that clone would indeed “experience something to be” that clone – a general qualitative description of consciousness (Nagel, 1974).
An “advanced” intelligent species is one that has the capability to run such simulations. The species must be intelligent to devise computers to be able to run such programs. In fact, they’d have to be more “intelligent” than we are today – perhaps by augmenting with machine superintelligence? However, the question of how an intelligent species become “advanced” is a redundant one. Taking our species, if one assumes any rate of technological progress at all, then it doesn’t matter where we place the flag – 100, 1,000, or 10,000 years from now – at some point, if the substrate independence of consciousness holds, we should expect our species to be able to run ancestor simulations.
Let’s look at the history of video games. 50 years ago, we had the game Pong. This just involved a paddle hitting a ball back and forth. Today, we have incredibly complex, high-definition video games with 3D graphics and enormous real time multi-player gameplay. That’s in 50 years! It doesn’t require much of a creative leap to envisage photo-realistic VR games that are almost indistinguishable from reality in the near-future. Think what might be possible 1,000 years from now? Even this rather naïve argument leads one to believe that “ancestor simulations” are probably inevitable (if we don’t go extinct).
In fact, in the paper, Bostrom shows, using back-of-the-envelope calculations (and present computational architecture), that a planetary‐mass computer “could simulate the entire mental history of humankind by using less than one-millionth of its processing power for one second”. Furthermore, our conscious experience is firmly planted in the macroscopic world and at specific levels of abstraction. This reduces the complexity of the simulation required. For instance, when I focus on my keyboard’s letter ‘a’, I don’t observe any of the other details about my keyboard nor do I observe the atoms that are making up that key. In a video game, the game-engine only renders what the user perceives (the designer doesn’t need to compute a rainbow from first principles QFT!). Similarly, the simulator doesn’t need to render everything down to a quantum level, only what is required for the simulated observer to be fooled. When someone looks through a microscope, takes a photo, or makes a quantum measurement the simulation can render the requisite detail on an ad hoc basis. In short, computational limits do not refute the simulation argument.
For the simulation argument, we only need to appeal to post-humans (our species in the future) but considering life elsewhere in the universe only strengthens the argument. There exists about a trillion galaxies in the observable universe and a typical galaxy hosts around 100 billion stars. Therefore, there are around 1×1023 solar systems that might harbour intelligent life in the observable universe. Here, the existence of a “Great Filter”, which explains the Fermi Paradox (why have we not seen aliens) by saying that intelligent life is incredibly unlikely, is swamped by the 1023 factor. In fact, for the simulation argument, we should consider the universe as an entirety (which might be infinite) and not just the observable part, making the multiplicative factor even larger. If homo-sapiens in their present form have been extant for only ~70,000 years and yet we’ve invented iPhones, the LHC, and MRI machines, then because the universe has existed for 13.8 billion years, we would expect intelligent life to have already reached the “advanced” stage somewhere else in the universe – provided that (1) is not true. Remember though, even if we find that the universe is finite, and we find that the “Great Filter” is so enormous that intelligent life has only developed once – on Earth – in the entire universe, the simulation argument still holds.
Why “ancestor simulations”?
Bostrom’s original paper is built around “ancestor simulations” (those resembling the simulator’s own evolutionary history) because he thinks it’s the type-of-simulation an advanced civilisation would most want to create. Most of the popular video games and films of today are those that most faithfully mirror reality. This is probably to do with ourstory-bias; we can relate more to games or films that are closer to what we experience in everyday life and so get more wrapped up in the story that it’s telling. This is why Sims, the popular video game, is about raising simulated humans not a family of dogs or aliens. I think for most of the ideas (a-g next section) I can think of, simulators would prefer to create simulations with beings that resemble themselves as much as possible.
However, Bostrom notes that the simulation hypothesis includes the possibility that we are simulated by aliens, but says that the distinction is irrelevant because whichever civilisation built the simulation we’re running in is our ‘home’ civilisation by definition.
The argument also includes the possibly that our universe is very different to that of our creators. Although most popular films are based in a “real” world, there are still popular animation and CGI films based in very different realities. As I’ve just said, I think there might be more simulations with physics and biology that is “close” to that in the simulator’s reality, but this claim could be wrong. In fact, when Elon Musk talks about the simulation argument, he thinks that a simulated world is likely to be more exciting than the simulator’s world. Just like a film is more exciting, more drama filled-etc than real life, he thinks a simulator is likely to want to distil only everything that is exciting about their reality into one they create. He therefore thinks that if we are simulated, then our reality is likely more exciting than that of our creators. I don’t agree with Elon on this. I tend to think that “excitement” increases with time. Following the canonical Stephen Pinker argument, if you were to choose any point in history to be alive, then you would always choose today. Because I think the main driver of this has been technological progress – which sparked the enlightenment etc – I think this will continue. I therefore think that if a civilisation has the capability to run ancestor simulations, then their everyday lives are likely much more exciting than ours.
Although I think simulators would want to create “beings” similar to themselves, that is not to say that we wouldn’t be different from our creators. The space of possible consciousnesses is enormous. I’m going to subscribe to Thomas Nagel’s (1974) description of consciousness which is that “there is something it is like to be a conscious organism”. There is something it is like to be me, there is something it is like to be you, and there is something it is like to be a dog. We are all conscious. There is not something it is like to be a toaster. A toaster is not conscious. There is a greater difference between my conscious experience and a dog’s conscious experience than between mine and yours – but our conscious experiences are still very different (a product of our genetics and our prior experiences). What the simulation argument assumes is that it is possible to create something in a computer where there exists “something it is like to be” that thing. Perhaps we have ‘lower’ forms of consciousness to that of our creators – we’re like a dog to them being the human. This is very possible.
Why would an advanced species want to run ancestor simulations?
If you had the capability to create a simulated world knowing it would end up like the one we have today (with lots of happiness, but also lots of unnecessary suffering), would you do so?
I don’t think there’s a ‘right’ answer to this yet – as it depends on your population axiology (which state of affairs is better than another?) All I would say is, even if you, the reader, or I might question the ethics of it, there are countless contemporary examples of people doing ethically questionable things just because they can (cloning etc), which provides the handful of people needed to falsify disjunct (2).
Here’s some of the possible future motives that I can think of:
a) A future historian wants to explore counterfactual histories.
b) Future economists/psychologists want to examine social interactions.
c) Just like the game Sims: people create the simulation for fun– remember our subjective experience of time could be millions of times slower than the objective flow of time in the simulator’s reality!
d) A scientist wants to tweak some physics parameters and see how that affects whether intelligent life does/doesn’t evolve.
e) It could be an art installation in a Louvre (or more likely – Tate modern) of the future.
f) It might be that philosophers of the future have convergently decided upon absolute moral truths and that the ‘right’ answer to ‘should we be creating simulations?’ is yes – perhaps they subscribe to total Utilitarianism where one thinks each conscious life has net positive welfare, therefore it is “a good thing” to create as many conscious lives as possible in order to maximise total welfare.
g) Someone, with a ridiculous amount of available compute, is bored one day, stumbles across the simulation argument and thinks, “wouldn’t it be cool if I make a simulated universe?”
The Simulation Hypothesis
How much credence should we have in disjunct (3)? Should we believe in the simulation hypothesis? Because the simulation argument is disjunctive, any evidence which increases our belief in (1) or (2) should decrease our belief in (3) by the same amount. Bostrom believes that (1) is not unlikely which is why he established the Future of Humanity institute to research existential risk. I also (sadly) think that (1) might be quite likely – in fact, super-intelligent AI, could be “the black-ball” technology that an advanced intelligent species invariably pulls out of the “urn of inventions” which leads to that species’ extinction. My AI PhD will be hoping to reduce this possibility!
Although (2) sounds implausible considering today’s norms – there are many people that, given the capability to do so tomorrow, would want to run ancestor simulations – it is difficult to comprehend the psychology of an “advanced” intelligent species. This is because, if our species reaches the “advanced” state (a point when creating simulator ancestors is possible), we should expect our species to be living lives very different to those we live today. We will very likely have populated thousands of planets and asteroids, our capabilities will be augmented by nootropics and brain-machine interfaces, and we may well be functionally “immortal”.
The simulation argument has several interesting implications. I’ll consider five here. First, if in a thousand years a group of software engineers have just created the infrastructure and written and compiled a program that will run an ancestor simulation. Immediately before they press the “start” button, they themselves will know that disjuncts (1) and (2) are false meaning that they should have high belief that they are themselves in an ancestor simulation.
Second, as the first implication eludes to, if the simulation hypothesis is true, then it can be repeated recursively to generate many layers of reality. There are some constraints from information theory and entropy that establish how many layers of depth can exist. However, for it to be believable for those in the simulation, the simulators need not simulate an entire universe down to the quantum level. For example, our species won’t be exploring anything beyond our solar system for a while! Therefore, the simulation only needs to create macroscopic holograms of anything beyond – we can’t precisely measure anything about stellar objects beyond macro-level properties and so the simulators don’t need to precisely compute every quark and lepton in Alpha Centauri.
Third, if there is a simulator, one has every reason to expect that they can and would tweak parts of our simulation to meet the objectives for “our” simulation. David Pearce commented that this “is perhaps the first interesting argument for the existence of a Creator in 2000 years.” However, remember that the simulation hypothesis would still hold that our simulator is bound by their level of reality’s naturalistic laws. Also, although by construction this would be a deistic position, the simulation hypothesis remains agnostic about any theistic claims.
Taking this further, I think it’s interesting to think about the classic theological question – why would an omnipotent, omniscient and perfectly benevolent god create a world with unnecessary evil in it? – in the context of the simulation hypothesis. I’m not theologically educated enough to know exactly how Christians, Muslims etc surmount this problem, but I think it’s quite difficult to do so without giving up one of the three: omniscience, omnipotence or perfect benevolence? Anyhow, if disjuncts (1) and (2) are false and the simulation hypothesis is true, then we have to accept that there was a “designer/creator” for this universe. But, we would avoid this theological trap because, if the creator was psychologically similar to us, we certainly have a huge amount of moral imperfections – so already we know we’re not a perfectly benevolent creator.
Fourth, remember for disjunct (2) to be true (arguably the most optimistic one), it has to be the case that ALL beings in all advanced species choose not to run simulations of their evolutionary history. If even a handful do (out of the possible trillions and trillions of people with the capability – as an advanced civilisation has probably populated a galaxy), then disjunct (2) is false and we are left with (1) or (3). This is because said handful with no ethical qualms will still (in the long-term) have enough time and computing power to run lots of simulations. So, again, we’re more likely to be in a simulated reality than base reality.
Finally, disjunct (3) isn’t necessarily as depressing as it might seem. It might well be that our rate of technological progress slows significantly, such that yes we are destined to go extinct before reaching the “advanced” stage, but that we exist for millions of years at a technological state (with lots of great stuff!) just before this (I think some rate of progress is guaranteed, but it could slow with time, like a log(x) graph).
I agree with David Chalmers’ position on the simulation hypothesis, which is that the meta-physical debate over whether we are in a simulation shouldn’t affect how we view anything or how we behave. Theoretical physics currently has quarks and leptons as the most fundamental constituents of nature – we know that this framework (the standard model) is incomplete and requires a more primitive theory, perhaps one based on “strings”. Originally, people thought “atoms” were the indivisible building blocks of nature. When, in 1911, Rutherford discovered that atoms were themselves made from protons, neutrons and electrons that didn’t change how people viewed the beauty of mountains, a tree, or other human beings. Likewise, physicists now know that protons and neutrons are made up of three quarks, but this doesn’t affect how a physicist sees their partner or their children; a physicist can still marvel at the beauty of a rainbow, a Da Vinci painting, or a rocket despite knowing that, at the micro-level, all of these are emergent macro-level phenomena governed simply by the interactions between point-like particles. For the same reason, if quarks and leptons are “strings” or if, instead, they are computer programs running on a simulated substrate, this doesn’t affect how “real” I view a chair to be.
I’ve argued that if we assume consciousness to be substrate independent, then the simulation argument appears to be true. I think the empirical and technological arguments I’ve employed in this essay aid our intuitive understanding about the simulation argument. I am also forced therefore to take the simulation hypothesis very seriously. With that, if I hope (1) is not true and I believe (2) to be unlikely, then I must have a high degree of belief that we are living in a simulation.
Bostrom, N. (2003). Are You Living in a Simulation? Philosophical Quarterly, Vol. 53, No. 211, 243-255.
Bostrom, N. The Vulnerable World Hypothesis. Working Paper.
Koch, C., & Tononi, G. (2015). Consciousness: here, there and everywhere? Philosophical Transactions of the Royal Society B: Biological Sciences 370.1668.
Nagel, T. (1974). What is it like to be a bat? Philosophical Review. LXXXIII (4), 435-450.