We are living inside a computer simulation

Bostrom's trilemma

The modern simulation argument comes from a single paper: “Are You Living in a Computer Simulation?”, published by the Swedish philosopher Nick Bostrom, then at Oxford, in the academic journal Philosophical Quarterly in 2003. It is worth taking seriously on its own terms, because it is more careful than its popular reputation suggests: Bostrom is not claiming we live in a simulation. He is claiming that at least one of three statements must be true, and inviting the reader to work out which:

(1) almost every civilization at roughly our level of technological development goes extinct before it becomes capable of running high-fidelity ancestor simulations, detailed recreations of an evolutionary history, populated by minds convincing enough to believe they are the originals; or (2) almost every civilization that does become capable of building such simulations chooses, for whatever reason, never to run a significant number of them; or (3) we are almost certainly living in a simulation right now.

The logic behind the trilemma is a numbers game. If even one posthuman civilization ever ran a large number of ancestor simulations, and if each of those simulations contained billions of simulated people who believed themselves to be the original, unsimulated species, then simulated observers would vastly outnumber non-simulated ones. At that point, Bostrom argues, an average observer picked at random has no particular reason to assume they are one of the rare originals rather than one of the overwhelming majority of copies. The argument leans on substrate-independence: the assumption, widely held in philosophy of mind and cognitive science, that consciousness is a matter of the right kind of information processing, not the particular biological material (carbon neurons versus silicon chips) that does the processing. If that assumption holds, there is no principled reason a sufficiently powerful computer couldn't host a mind as real, to itself, as any of ours.

None of this requires exotic physics, and Bostrom is specific about what “very large computers” would actually mean. Reasoning from the physical limits of computation rather than from any particular product cycle, he estimates that a computer built from already-understood nanotechnology, with a mass on the order of a planet, could perform on the order of 10⁴² operations per second; a computer exploiting quantum computing more fully could in principle do vastly more still. Simulating a human brain in real time, on his figures, costs somewhere in the range of 10¹⁴ to 10¹⁷ operations per second, and reproducing the entire mental history of humankind (every person who has lived, in the kind of detail an ancestor simulation would require) costs on the order of 10³³ to 10³⁶ operations in total. Set against a single planetary-mass computer, Bostrom calculates, that entire history could be run “using less than one millionth of its processing power for one second.” On these numbers, the bottleneck was never going to be computation; it would be whether a civilization survives long enough, and stays curious enough, to bother.

The same intuition has also traveled well outside philosophy departments. In June 2016, at Recode's Code Conference, Elon Musk offered his own version of the argument from the stage: forty years ago video games were “two rectangles and a dot” (Pong) and within four decades they had become photorealistic, with millions of people playing simultaneously online. Extrapolate that trajectory forward, he argued, even at a small fraction of today's pace of improvement, and games indistinguishable from reality stop looking like fiction. Once such games exist and can run on billions of ordinary devices, he concluded, “the odds we're in base reality is one in billions.” Whatever one makes of the number itself, the underlying observation, that the distance between a video game and a lived reality has been closing quickly, with no obvious wall in sight, is the same one that gives the trilemma's first two branches their bite: it is not hard to imagine a civilization eventually building the very large computers Bostrom describes, because a much smaller version of that trend is already visible from an ordinary living room.

The argument's reach has extended into physics as well as popular culture. In 2012, physicists Silas Beane, Zohreh Davoudi, and Martin Savage took the hypothesis seriously enough to ask what it would predict, publishing “Constraints on the Universe as a Numerical Simulation.” Reasoning from a real computational technique (physicists already simulate small volumes of space by placing particles on a discrete lattice, because continuous space is too expensive to compute exactly), they showed that a simulated universe built the same way would very faintly break perfect rotational symmetry, in a way that could in principle leave a trace in the distribution of the highest-energy cosmic rays ever recorded. Whatever one concludes from it, the paper is a sign of how far the trilemma pushed the conversation: physicists with no particular stake in the idea nonetheless found a legitimate, publishable question hiding inside it: what a simulated cosmos would look like, if its underlying architecture leaked at all.

None of it requires the third branch of the trilemma to be true, and Bostrom has never claimed that it is. His own view, stated plainly on the essay's companion site, is that he does not know which of the three propositions holds, only that believing all three false at once looks harder to justify than most people assume.

The problem of proof

The central weakness of the simulation hypothesis is not that it has been tested and failed: it is that almost no one agrees it can be tested at all, which is a more fundamental problem for any claim that aspires to be scientific rather than purely philosophical. The philosopher of science Karl Popper's longstanding criterion for distinguishing science from non-science is falsifiability: a hypothesis earns scientific standing only if there is some observation that could, in principle, prove it wrong. Physicist Sabine Hossenfelder and others have argued the simulation hypothesis fails this test outright, since any experiment that seemed to rule it out could simply be reinterpreted as a limitation the simulators built in on purpose.

Computer scientist Scott Aaronson has made the point with particular force. Reviewing a physics paper that claimed to identify computational limits a simulated universe couldn't exceed, he noted that every such argument can be escaped by simply assuming a more powerful simulator: if classical computers can't fake quantum mechanics efficiently, “why not just imagine that the universe is being simulated on a quantum computer?” And if even a quantum computer runs into a hard limit, there is nothing stopping a simulator with access to enough time from simply running the calculation slower, spending, in his words, a trillion years of its own time to simulate one second of ours. Because the hypothesis can always retreat to a more powerful hypothetical simulator, no finite experiment can close it off.

The one serious attempt at a physical test only sharpens the problem. In 2012, physicists Silas Beane, Zohreh Davoudi, and Martin Savage published “Constraints on the Universe as a Numerical Simulation,” reasoning from an existing real-world technique: physicists already simulate small patches of the universe by placing subatomic particles on a discrete cubic lattice, because continuous space is too expensive to compute directly. If our own universe were built the same way, the underlying grid should very faintly break perfect rotational symmetry, and that break should show up as a subtle preferred direction in the highest- energy cosmic rays. Using the observed cosmic-ray spectrum, they derived a bound: if such a lattice exists, its spacing must be smaller than roughly 10^-11 GeV^-1 (a technical way of saying: finer than anything current instruments can rule out). No lattice signature has been detected. But even the authors are careful to frame this as a speculative, order-of- magnitude exercise, not a genuine test of the simulation hypothesis in general: a universe built on some other computational architecture, or with no lattice at all, would sail through the constraint untouched.

Astronomer David Kipping attacked the question from a different angle in 2020, re-running Bostrom's trilemma as a formal Bayesian calculation rather than an intuitive numbers game. His conclusion was not that the simulation hypothesis is false, but that the earlier argument carried more certainty than the math strictly supports: once you account for our genuine uncertainty about whether large-scale ancestor simulation is even possible, the odds that we live in one come out close to fifty-fifty, essentially even, far more modest than the trilemma's popular reputation suggests. Kipping has also been blunt about the falsifiability problem in his own right, asking, if the claim can't be tested either way, “how can you claim it's really science?”

That gap between the rigorous argument and its popular retelling shows up clearly in how the hypothesis actually circulates outside academic papers. Pointing at decades of video-game graphics and extrapolating to “indistinguishable from reality” (the version of the argument Elon Musk offered from the stage at Recode's Code Conference in June 2016, concluding that “the odds we're in base reality is one in billions”) answers a narrower question than the one Bostrom's trilemma actually poses. Rendering a convincing scene for a human's eyes and ears means fooling a handful of senses; simulating a universe a physicist could study means reproducing the behavior of every particle in it, a vastly larger computational problem that a few decades of game graphics say nothing about one way or the other. Bostrom's own paper does not actually lean on extrapolating current technology at all; it leans on the physical limits of computation at a hypothetical posthuman level, a more abstract and more modest claim than “games keep getting better.” Stating the conclusion as a near-certain number also skips past the hedge Bostrom was careful to keep in place: the 2003 paper declines to say which branch of the trilemma is true, and “one in billions” is a far bigger claim than anything the trilemma itself asserts.

There is a philosophical objection underneath the physics, too. The whole trilemma assumes that a computer, however powerful, could host actual subjective experience: that being simulated in enough detail is the same as being conscious. That premise is contested territory in philosophy of mind, not a settled fact, and if it is false the entire argument loses its footing regardless of what any cosmic-ray survey finds.

Why a probability argument feels like proof

Part of the simulation hypothesis's pull is simply timing. It was formalized in 2003, just as home computing power, video game graphics, and later virtual reality were visibly improving year over year: it is easy to look at the jump from blocky 1980s game sprites to today's photorealistic rendering and extrapolate the curve forward until “indistinguishable from reality” feels like the next obvious stop, rather than an enormous, untested leap.

The idea also does something religious and existential traditions have long done (offer an account of why anything exists and whether it means something) but frames it in the vocabulary of computer science and probability rather than theology, which makes it feel accessible to people who find traditional metaphysics unpersuasive. That it arrived through a peer-reviewed philosophy paper, with formal logic and named propositions, adds a veneer of proof that a casual argument wouldn't carry; a trilemma summarized in a headline reads very differently from a trilemma read in full, with its careful hedges about what remains unknown.

Finally, prominent technologists like Elon Musk have mused publicly about the hypothesis being plausible or even likely, which lends it a kind of borrowed authority, even though, as the actual debate among physicists and philosophers shows, serious engagement with the idea has produced disagreement and unresolved falsifiability problems, not consensus.