“Are we living in a simulation?” is one of those questions that sounds like late-night dorm-room talk and then turns out to have a serious literature behind it.
The honest short answer to “how likely” is that nobody knows, and that the question may not even have a clean numerical answer. But that is not a reason to wave it away. The reasons we cannot confidently put a number on it are themselves interesting, and they tell us something real about the limits of probability, the nature of consciousness, and what counts as science.
So this is not a post that will hand you a percentage. It is a post about why the percentage is so slippery, and what I actually think once you account for that.
TL;DR
- The simulation argument (Nick Bostrom, 2003) is a trilemma, not a claim that we are simulated: either civilizations die before technological maturity, or mature ones do not run ancestor simulations, or we are almost certainly living in one.
- The case for “quite likely” rests on a counting argument - if simulated minds vastly outnumber biological ones, a randomly chosen observer is probably simulated.
- That argument leans on shaky premises: that computation alone produces consciousness, that simulations are cheap enough to run in huge numbers, and that we can meaningfully count observers at all. The measure problem breaks the counting down, especially in an infinite cosmos.
- A perfect simulation is unfalsifiable by design, which pushes the question toward metaphysics. Narrow tests, such as hunting for a spacetime lattice, can only rule out unimaginative versions of it.
- David Kipping’s 2020 Bayesian analysis landed near 50-50, but that figure is a confession of ignorance dressed in arithmetic, not a measurement.
- My view: as a request for a probability, the question is currently malformed. It is worth thinking about clearly, but far too uncertain to deserve a number.
What the simulation argument actually says
The first thing to clear up is that the famous version of this idea does not claim we are probably in a simulation. It claims something more careful.
In 2003, the philosopher Nick Bostrom published a paper called Are You Living in a Computer Simulation?. It does not argue for the conclusion that we are simulated. It argues that at least one of three propositions must be true:
- Almost all civilizations at our stage go extinct before reaching technological maturity, the point where they could run enormous numbers of detailed simulations of their own ancestors.
- Technologically mature civilizations almost never bother to run such “ancestor simulations,” perhaps through lack of interest, ethics, or law.
- We are almost certainly living in a computer simulation.
This is a trilemma, not a conclusion. Bostrom’s own position, as far as I can tell, is roughly even credence spread across the three, not a confident vote for the third. The argument’s real achievement is the structure: if you reject the first two, the third follows, and that is a genuinely uncomfortable place to be pushed.
The logic of the third branch is a counting argument. If a mature civilization can simulate conscious minds, and if it runs even a modest number of historical simulations, then simulated minds vastly outnumber the original biological ones. Pick a conscious observer at random and they are overwhelmingly likely to be one of the simulated many rather than one of the original few. You have no special reason to think you are an exception. So, the argument goes, you should assign high probability to being simulated.
It is elegant. It is also, I think, much more fragile than it first appears.
The case for “quite likely”
Take the argument at full strength for a moment.
We already build simulations, and they improve fast. Step back and the timescale is startling. Anatomically modern humans have been around for roughly 300,000 years, and we have had electronic computers for less than a century of that, a sliver of a sliver of our existence. In that sliver we have gone from a bouncing white square in Pong to open virtual worlds, photorealistic game engines, and physically detailed scientific models. A still frame from a modern game is already hard to tell apart from a photograph, and the gap is closing in motion too. Extrapolate that curve even modestly and the moment when a simulated world becomes genuinely indistinguishable from lived experience stops looking like distant fantasy and starts looking like a question of when, not if. A species that has been technological for an eyeblink is already this close. It is hard to insist that a civilization with a few thousand more years of practice, let alone a few million, could not close the rest of the distance.
If you also believe that consciousness is substrate independent, meaning that what matters is the pattern of information processing and not the specific stuff doing it, then a sufficiently detailed simulation of a brain would not just model a mind, it would be one. Combine that with the counting argument and the conclusion is hard to dodge. Most minds like ours would be simulated, so you probably are too.
There is also the question of motive, which is Bostrom’s second branch wearing different clothes. Granting that a civilization could run ancestor simulations, would it actually want to? I find it hard to believe the answer is a flat no. If I were in base reality and had that power, I am fairly sure I would run a simulation of myself and the people I know, just to watch how we coped with situations we never actually had to face. Plain curiosity would be enough. And if the hand on the controls were not human but alien, the same reasoning carries over. An advanced species might run enormous numbers of simulated worlds precisely to see what those worlds come up with, harvesting the ideas, designs, and solutions that minds invent under pressure and keeping whichever turn out best. A civilization could even run a vast, ever-expanding library of simulations as a kind of innovation engine, ours among them. Once you grant the capability, total indifference to it starts to look like the strange assumption, not the safe one.
The counting argument is also more robust than it looks to how the cosmos is actually built. Some cosmologists suspect our Big Bang was not a single event but one of an unbounded series, a picture associated with eternal inflation, and several lines of physics point toward a vast or infinite ensemble of parallel universes. Run the simulation argument under either setup. If reality contains endlessly many universes, then it contains endlessly many base realities and, provided even a tiny fraction of mature civilizations build simulations, far more simulated ones, nested layer upon layer. If instead there was exactly one Big Bang and one universe, the squeeze is arguably tighter still: that single base reality could spin up a far larger, perhaps unbounded number of simulations within itself, so a mind picked at random from that one universe is overwhelmingly likely to be one of the simulated crowd rather than one of the base-reality few. Whichever way the cosmos is arranged, the same pressure remains. Base realities look rare and simulations look common.
This is why some prominent voices put the odds very high. Elon Musk has said the chance we are in base reality is “one in billions.” The pull of the argument is real, and if you accept its premises the third branch is not a wild leap. It is close to forced.
The whole question, then, is whether those premises hold.
Where the argument starts to crack
The consciousness assumption
The counting argument needs simulated minds to actually be conscious. If they are not, they are not observers, and they do not belong in the count at all. A trillion simulated philosophical zombies do not make it more likely that you, a being who genuinely experiences things, are simulated.
Substrate independence is a reasonable bet, and many people in the field hold it. But it is a bet. We do not have a theory of consciousness that tells us what is sufficient to produce experience. We do not know whether running the right computation on silicon yields an inner life, or merely a very good description of one. Until that is settled, the simulation argument is resting on an unproven assumption about the deepest unsolved problem in science. If consciousness turns out to need something biological, or quantum, or otherwise hard to replicate, the argument collapses straight into branch one or branch two.
The cost objection, and its clever rebuttal
A simulation detailed enough to fool physicists looks staggeringly expensive. Simulating a universe down to quantum behavior plausibly costs as much as, or more than, the universe being simulated. Worse, quantum mechanics is fundamentally probabilistic, so even a perfect simulator cannot simply precompute everything.
The standard rebuttal is that a smart simulation would not render everything. It would do what video games do: compute things in detail only where someone is looking, and keep the rest as cheap approximation until observed. Lazy evaluation on a cosmic scale.
I find this rebuttal genuinely clever and also genuinely double-edged. To save on cost by rendering on demand, the simulation has to track every observation, keep the books perfectly consistent across all observers, and patch the approximation back into full detail the instant a scientist points an instrument at it. The accounting required to never get caught may itself be enormous. The cost objection does not kill the argument, but it does dent the confident assumption that mature civilizations could run simulations cheaply and therefore would run a great many.
The counting problem
The argument says most observers like us are simulated. But “most” requires counting, and counting requires a well-defined set to count over. How many simulations are there? Across what span of time, what set of civilizations, what possibly infinite space of outcomes? To say a randomly chosen observer is “probably simulated” you need a probability distribution over all observers, and we have no principled way to write one down. This is a version of what cosmologists call the measure problem, and it is not a footnote. An infinite cosmos makes this worse rather than better. If there are infinitely many base realities and infinitely many simulated ones, the ratio the counting argument relies on becomes infinity divided by infinity, a quantity with no fixed value. Reaching for an unbounded multiverse does not secure the conclusion that we are probably simulated; it removes the ground the conclusion was standing on. Without a measure, the phrase “almost certainly” in branch three is doing work it has not earned.
Is it even falsifiable?
A perfect simulation is, by construction, indistinguishable from base reality. If every possible observation comes out the same either way, then the hypothesis predicts nothing, forbids nothing, and explains nothing extra. By Occam’s razor, piling realities on top of realities to explain a world that already looks self-consistent is a move you should resist unless something forces it.
That does not make the idea meaningless. But it does push it toward metaphysics rather than physics, which matters a lot when someone asks you for a probability.
Can we actually test it?
Sometimes, a little, for certain kinds of simulation.
In 2012, the physicists Silas Beane, Zohreh Davoudi, and Martin Savage published Constraints on the Universe as a Numerical Simulation. Their idea was sharp: physicists already simulate tiny patches of the universe by laying down a discrete spacetime grid, or lattice. If our universe were that kind of simulation, the underlying lattice might leave fingerprints, such as a slight directional bias in how the highest-energy cosmic rays travel, or a ceiling on their energy.
It is a lovely piece of work, and it shows the question is not entirely beyond physics. But notice how narrow it is. It can only constrain a specific type of simulation, one built on a lattice resembling the ones our own physicists use. A simulation running on different principles, or one clever enough to hide its grid, would sail straight past the test. You cannot rule out the general hypothesis this way. You can only rule out unimaginative versions of it.
What “likely” even means here
Now we can return to the actual question. How likely is it?
In 2020, the Columbia astronomer David Kipping ran a Bayesian analysis, written up in Scientific American under the headline “Chances Are about 50-50.” He collapsed Bostrom’s trilemma into a dilemma, since branches one and two both end with “no simulations exist,” and weighed that against the simulation hypothesis. With a neutral prior, the result landed a hair under 50 percent. Slightly more likely that we are in base reality than not.
I think the precise figure matters far less than what the exercise reveals. A Bayesian answer is only as good as its prior, and here the prior is close to a guess. We are assigning probabilities to civilizational futures we have never observed, to a theory of consciousness we do not have, and to a reference class of “observers” we cannot define. Garbage in is too harsh, because the reasoning is careful. But uncertain in, uncertain out is exactly right. A number like 50 percent here is not a measurement. It is a confession of ignorance, dressed in arithmetic.
Kipping himself added a sharp caveat: the day a civilization actually builds a conscious simulation, the odds lurch hard toward “simulated.” So the honest probability is not a fixed quantity. It is contingent on a fact about technology that we do not yet know.
An old question wearing new clothes
It is worth stepping back. The simulation hypothesis feels like a product of the computer age, but the doubt underneath it is ancient.
Plato had the prisoners in the cave mistaking shadows for the world. Zhuangzi dreamed he was a butterfly and woke unsure which one was the dream. Several philosophical and religious traditions describe ordinary reality as a kind of veil or illusion. Descartes imagined an “evil demon” feeding him a complete and convincing falsehood, and twentieth-century philosophers updated that to the brain in a vat. The simulation hypothesis is the same skeptical worry running on newer hardware: maybe the world you experience is not the world as it fundamentally is. The Stanford Encyclopedia of Philosophy’s entry on skepticism traces how stubbornly this problem has resisted a knockdown answer for centuries.
What the computational version adds is a mechanism and a motive. It does not just say “you could be deceived.” It offers a specific story about who would do the deceiving and why, and a counting argument for why it might be probable rather than merely possible. That is a real contribution. But the core move, doubting that appearances reach all the way down, is not new, and it has never been refuted, because by design it cannot be.
There is also a quieter response worth knowing. In Reality+ (2022), the philosopher David Chalmers argues that even if we are in a simulation, our world is not therefore fake. A simulated tree is still a real pattern; a simulated life still contains real joy and real loss. On this view the discovery would change the substrate of reality, not its value. It would be a fact about the basement, not a verdict on the house.
My own view
I do not think we are likely to be living in a simulation, and I do not think we are likely not to be. I think the question, as a request for a probability, is currently malformed.
To produce an honest number you would need at least three things we do not have: a theory of consciousness that says whether computation alone can produce experience, a sense of whether mature civilizations actually run ancestor simulations, and a way to count observers that does not fall apart under scrutiny. Each of those is an open problem. Multiply three deep uncertainties together and you do not get 50 percent. You get “the question is not yet answerable,” which 50 percent is really just a polite way of saying.
What I do take from the argument is its discipline. Bostrom’s trilemma is genuinely valuable, not because branch three is scary, but because branches one and two are easy to overlook. If you find yourself confident we are not simulated, the argument quietly asks you to commit: do you believe civilizations like ours usually fail before maturity, or that mature ones lose all interest in simulating their past? Those are substantive claims about the future of intelligence, and most people who dismiss the simulation idea have never actually picked one. The argument’s gift is that it makes the cost of disagreement explicit.
And there is one more thing I keep coming back to. Even if we are in a simulation, the sunlight is still warm, grief still hurts, a proof is still elegant, and the night sky is still staggering. If this reality is computed, then computation is capable of all of that, which is its own kind of wonder. If it is not, we are a temporary arrangement of atoms that woke up able to ask whether it was real. Either way, the strangeness is not reduced. It only moves.
So how likely is it that we are living in a simulation? Likely enough to be worth thinking about clearly. Far too uncertain to be worth a number. And, in the end, not the question that decides whether any of this matters.
Closer To Truth on this question
If you want to hear physicists and philosophers actually argue this out, the Closer To Truth YouTube channel is the best place I know. Robert Lawrence Kuhn has interviewed a wide range of thinkers on simulated reality, and what comes through is the same thing you find with most deep questions: even the experts do not converge on an answer. They converge on the fact that the question is real, and that the easy confident takes in either direction usually have not done the work.
A few interviews worth your time: