The Simulation Hypothesis

← Encyclopedia

The simulation hypothesis is the proposition that the world is not the bottom of reality: that what presents itself as the physical universe is a computation running on hardware elsewhere, and that human beings are among the things being computed. As a picture this is very old. As an argument it is very recent — what turned a science-fiction premise into a live philosophical position was a short paper published in 2003, unusual among such ideas in appealing to no demon, no dream, and no doubt, only to assumptions about computing power and a piece of probability.

Nick Bostrom’s “Are You Living in a Computer Simulation?” appeared in the Philosophical Quarterly that year, and its conclusion is a disjunction — a fact nearly every popular retelling gets wrong. The paper argues that at least one of the following is true: “(1) the human species is very likely to go extinct before reaching a ‘posthuman’ stage; (2) any posthuman civilization is extremely unlikely to run a significant number of simulations of their evolutionary history (or variations thereof); (3) we are almost certainly living in a computer simulation.” It does not say which. Bostrom’s published advice was to divide one’s credence roughly evenly among the three, and when a critic later saddled him with the belief that we are simulated, he corrected the record in print: “I believe that we are probably not simulated.” The argument is a fork, not a destination. It is uncomfortable enough as a fork, since the first prong is extinction.

The load-bearing premise comes from the philosophy of mind. Substrate-independence is the assumption that conscious states can arise in any of a broad class of physical substrates — silicon as well as carbon — provided the right computational structures run there; Bostrom needs only a modest version, pitched at the grain of the synapse. Grant that, and the rest is arithmetic. By his estimates, a single computer with the mass of a planet, built from nanotechnology designs already published, could simulate the entire mental history of humankind using under a millionth of its capacity for one second. A technologically mature civilization curious about its own past could therefore run “ancestor-simulations” in enormous numbers, and economically: distant astronomy compressed, microscopic detail rendered only when someone looks, the rare brain that notices an anomaly quietly edited. If such civilizations exist and care to, simulated minds vastly outnumber original ones — and an observer with no way of telling which it is should, Bostrom argues, set its odds by the proportions.

That last step is the contested one. It runs through what Bostrom calls a bland indifference principle — if a known fraction of all beings with human-type experiences are simulated, one’s credence in being among them should equal that fraction. Brian Weatherson, in the same journal, distinguished four ways of reading it and argued that none does the work required; remaining confident of being flesh and blood, he concluded, is rationally permissible. Bostrom replied at length. In 2011 he and Marcin Kulczycki conceded a genuine flaw in the original formula — under certain demographic assumptions all three prongs could fail together — and published a patch rescuing the conclusion from weaker premises. Physicists have been blunter. Sean Carroll observed that the reasoning tends to undermine itself — it treats humans as typical observers, yet humans demonstrably cannot run such simulations — and that the laws physics actually finds contain hidden structure doing no observable work, which a simulator would have no reason to compute. Sabine Hossenfelder filed the hypothesis under religion rather than physics.

There have been attempts to drag it toward the laboratory. John Barrow suggested that simulators with incomplete knowledge would have to patch their worlds as they ran, so a simulated universe might betray itself in occasional glitches and a slow drift in the constants of nature. In 2012 the physicists Beane, Davoudi, and Savage worked out what would follow if the universe ran on a cubic spacetime lattice of the kind used in early quantum-chromodynamics calculations: the highest-energy cosmic rays would arrive with a telltale directional bias, and the observed cutoff in their spectrum already bounds the lattice spacing. The limits of these tests are conceded on every side. Nothing obliges a simulator to use a crude uniform grid; better numerical methods erase the signature; and a simulator could in any case feed false readings to every instrument involved. The hypothesis touches evidence without being falsifiable from inside. Preston Greene added a prudential coda: even given a decisive test, running it might be unwise, since a simulation that discovers what it is may be a simulation that gets switched off.

What the argument is not, its principals insist, is skepticism. Bostrom distinguishes it sharply from Descartes’s demon and the brain in a vat: it starts not from doubt but from empirical premises about technology, and it aims to say something about the world rather than about how little can be known of it. David Chalmers pushed the point furthest. In “The Matrix as Metaphysics” he argued that the hypothesis of lifelong simulation is not a skeptical scenario but a metaphysical one — a claim about what physical things are made of, not about whether they exist. If it were true, tables and chairs would still be real; their deepest nature would be computational, which is no stranger than what physics already says of them. “We can think of the Matrix Hypothesis as a creation myth for the information age,” he wrote, and his Reality+ of 2022 generalizes the verdict, arguing that virtual worlds are genuine realities and that the simulation idea answers Descartes’s old puzzle rather than sharpening it.

The hypothesis is usually given a long ancestry, and the ancestry deserves more care than it usually receives. Zhuangzi, in fourth-century-BCE China, dreamt he was a butterfly and woke unable to say which of the two was dreaming the other; in context the parable urges a mild perspectival humility — any awakening may sit inside a larger dream — not a doctrine that the world is manufactured. The māyā of Advaita Vedānta makes the world of plurality a superimposition upon Brahman, neither wholly real, since knowledge dissolves it, nor wholly unreal, since it is experienced; its business is liberation, not engineering. Descartes’s malicious demon of 1641 was a methodological instrument, the far end of a deliberate doubt built to discover what could survive it. Berkeley abolished mind-independent matter altogether: to be is to be perceived, and for him there was no veil to tear, because appearance simply was the reality. Putnam’s brain in a vat became the modern textbook skeptical scenario — though Putnam’s own purpose was to argue that the scenario refutes itself. The resemblances between these and the simulation hypothesis are real, and worth following. They are not the same argument. Each was built for its own ends, and none of them reasons from the economics of computation.

Few technical philosophy papers acquire a public career; this one did. The fiction came first — Daniel Galouye’s Simulacron-3 in 1964, Fassbinder’s World on a Wire, The Thirteenth Floor, and above all The Matrix in 1999, whose texture scholars read at once as Gnostic and Buddhist: ignorance as the problem, a false world mistaken for the real one, awakening by a guide. Then, in 2016, Elon Musk told a conference audience that the odds we live in base reality are about one in a billion, reasoning from Pong to photorealistic games in forty years — adding that he and his brother had banned the topic from hot tubs. Bostrom’s standing FAQ now fields the question of what follows if one happens to be Elon Musk — simulators might oversample world-historical figures — along with his practical counsel should the third prong be true: “I recommend brushing your teeth, getting enough sleep, and being kind to other people and animals.”

An older vocabulary fits all of this almost too well, and noticing the fit is interpretation, not history. The visible world as a made image — fashioned by a power that is not the highest reality, dependent at every moment on a reality beyond it — is among the oldest cosmological intuitions on record, and the simulator is a demiurge in the precise sense: potent, possibly fallible, not ultimate, working from a reality on which the image-world wholly depends. The moderns keep drawing the lines themselves. Bostrom ends his own paper with what he calls a naturalistic theogony, in which simulators stand to the simulated as gods — intervening at will, seeing everything, themselves answerable to whatever runs the level beneath them. Barrow wrote that in such cosmologies “the gods reappear in unlimited numbers in the guise of the simulators.” Dainton matched Bostrom’s stacked virtual worlds to the 365 nested heavens of the Gnostic teacher Basilides — ours the lowest — and proposed in earnest that natural evil may be the work of simulators rather than of God, while Steinhart judged the argument’s relations with Neoplatonism friendly. The open dispute over the simulator’s character — careful experimenter, negligent gamer, buffer between God and the world’s pain — replays the old quarrel between the good craftsman of Plato’s Timaeus and the blind maker of the Gnostics. An echo is not an identity, and rhyme is not derivation. What is genuinely new is only the direction of travel: the Hermetica reasoned from the cosmos as a made image to reverence for its maker, while Bostrom reasons from trends in chip fabrication to a probability that humanity is the image. The conclusion is ancient; only the premises are new.

→ Related: Gnosis · Hard Problem Of Consciousness · Kardashev Scale