Rendered Reality

In June 2016 Elon Musk told a technology conference that the odds we live in “base reality” were “one in billions”. Games had improved so fast, he reasoned, that fully realistic simulations were inevitable, and once they existed simulated people would outnumber real ones. To get there Musk had to assume that simulations indistinguishable from reality are possible, which is the thing he was trying to establish. Over the following decade the people repeating him grew more numerous and less troubled by it.

People who found The Matrix disturbing in 1999 now post the idea as a joke. Between 2023 and 2026 generative AI began producing a large share of the text, images and voices online, and watching something get rendered became an ordinary daily experience. Philosophers have asked whether the world is real for 2,400 years. Zhuangzi wrote one of the oldest surviving versions in the fourth century BC:

Once Chuang Chou dreamt he was a butterfly, a butterfly flitting and fluttering around, happy with himself and doing as he pleased. He didn’t know he was Chuang Chou. Suddenly he woke up and there he was, solid and unmistakable Chuang Chou. But he didn’t know if he was Chuang Chou who had dreamt he was a butterfly, or a butterfly dreaming he was Chuang Chou. — — Zhuangzi, The Complete Works of Chuang Tzu, trans. Burton Watson

Zhuangzi meant it as relativism. There must be some distinction between the man and the butterfly, he wrote, but nobody inside the dream can settle which is which. Plato’s prisoners, chained in a cave, take the shadows on the wall for the world. Descartes in 1641 imagined a demon devoted to deceiving him about everything he perceived, mathematics included. Berkeley argued that objects exist only while someone perceives them.

After Newton, physicists described the universe as a machine, and for two centuries the machine they had in mind was a clock. Leibniz mocked the theology that came with it, writing in 1715 of a God who, by his account, “wants to wind up his Watch from Time to Time: Otherwise it would cease to move.” Laplace removed the winder and imagined an intelligence that, given the position and velocity of every particle, could compute the entire future. Nineteenth-century physicists swapped the clock for the steam engine, and the universe began running down towards “heat death.” Twentieth-century physicists swapped the steam engine for the computer.

John Wheeler coined “it from bit” in 1990 to say that reality is at bottom information, and that the universe is not merely like a computer but is one. Wheeler was making a claim about what the world is made of. Nick Bostrom made a different one in 2003, about who is running it. If advanced civilisations tend to survive and to run detailed simulations of their own history, he argued, simulated minds would outnumber real ones so heavily that any given mind is more likely simulated than not, from which at least one of three propositions must hold. Civilisations like ours nearly always die before reaching that stage, or those that survive decline to run such simulations, or we are almost certainly living in one.

Several physicists have tried to test it. A 2012 paper looked for a cut-off in the energies of cosmic rays, the kind of edge a universe computed on a lattice would leave. Melvin Vopson proposed a “second law of infodynamics” in 2023 and presented it as testable. Franco Vazza calculated in 2025 that simulating Earth down to the smallest measurable scale would mean converting something like 100,000 stars into energy, which makes a full simulation “(nearly) impossible.” A simulator competent enough to be worth worrying about would defeat any such search anyway, since it need only render what is being observed and could smooth over any shortcut the moment someone looked for it, the way a video game draws detail only where the player points the camera.

Believers in the hypothesis inherit the apparatus of a religion. Their simulator sits outside space and time, can read and rewrite the code, and holds motives unavailable to anyone inside; suffering becomes a parameter and good behaviour a wager against deletion. Like the creationists they would not care to be compared with, they hold a position no observation could count against. Some act on it regardless. The New Yorker reported in 2016 that “two tech billionaires have gone so far as to secretly engage scientists to work on breaking us out of the simulation.”

---

No animal perceives the world as it is. Each one registers the narrow band of signals its organs evolved to catch. Jakob von Uexküll named that band the Umwelt in 1909 and illustrated it with a tick, which finds its way through the world on three cues: the butyric acid in mammal sweat, warmth, and the touch of skin. “Out of the vast world which surrounds the tick,” he wrote, “three stimuli shine forth from the dark like beacons, and serve as guides to lead her unerringly to her goal.” The branch the tick waits on has no colour and no texture for the tick. It is a place to wait for the smell of a passing mammal.

Donald Hoffman put that on a mathematical footing. He and his collaborators ran evolutionary simulations in which organisms that perceive their environment accurately competed against organisms tuned only to what bears on survival, and the accurate perceivers went extinct. Their fitness-beats-truth theorem holds that accuracy loses because it costs effort that fitness-tuned perception never spends. Perception on this account works like a desktop interface, useful because the icon hides the machinery: “Spacetime is our desktop,” Hoffman writes, “and physical objects, such as spoons and stars, are icons of the interface of Homo sapiens.” Most researchers decline to follow him to the conclusion that spacetime is therefore unreal. Far fewer dispute the smaller claim, which arrives from other directions as well: evolutionary psychologists catalogue the useful errors, the stick taken for a snake, and neuroscientists describe perception as the brain’s running best guess rather than a recording. Anil Seth calls it a “controlled hallucination,” and adds that “we’re all hallucinating all the time,” and that “when we agree about our hallucinations, we call that reality.”

Whether a hidden computer renders the world is unsettled. That the nervous system renders it is not. From a thin stream of signals it assembles a workable version of the world, tuned by evolution to keep its owner alive rather than to be correct, and nobody has ever stepped outside that assembly to compare it against the original. Everyone already lives inside a render. Until recently nobody chose its design.

---

Natural selection did the building for almost the whole history of life, working blind and answering to no one. Engineers started building their own in the last decade and call them world models: systems that learn an environment well enough to generate its next frame. David Ha and Jürgen Schmidhuber trained an agent in 2018 that improved its driving inside its own dream of a racing game. Google DeepMind’s Genie could conjure a playable world from a single image by 2024, and Genie 3 was generating interactive three-dimensional environments in real time a year later. Another model ran the 1993 shooter Doom with no game engine underneath it, every frame produced by a neural network. OpenAI called its Sora generator “a promising path towards building general purpose simulators of the physical world”; Yann LeCun and other sceptics answer that such models learn the statistics of pixels and not the laws of physics.

Robots now learn to walk and to grip inside simulated physics before their trained policies are loaded onto metal. Researchers have started putting language models inside these worlds as inhabitants. A Stanford team populated a simulated town called Smallville in 2023 with 25 characters that formed memories and organised a Valentine’s party nobody had scripted, and a startup reported releasing about 1,000 agents into Minecraft the following year, where by its own account they developed an economy and adopted a constitution. Bostrom had imagined a posthuman civilisation running ancestor-simulations full of such agents. Graduate students were running small ones within twenty years of his paper.

David Chalmers argues that a rendered world is not less real than a physical one. If a physical object is at bottom a pattern of information, he writes in Reality+ (2022), a virtual object is that same pattern realised in silicon and therefore a real digital object: “virtual reality is genuine reality,” and “what happens in VR really happens.” Chalmers arrives where the biologists arrive, from the other side. They find that nobody sees the physical world as it is; he finds the virtual world real even though it is rendered.

Millions of people hold daily conversations with chatbots built to be friends or confidants, and a 2025 study found that these ease loneliness “on par only with interacting with another person.” Machines write a growing share of the open web: Ahrefs found that 74.2% of new pages in April 2025 contained AI-generated text. Photo tools from Google and Samsung let people revise their own memories after the fact, opening a shut eye or removing a stranger from a holiday. A Samsung executive answered the complaint by denying the category: “There is no such thing as a real picture.” An MIT study found that AI-edited images roughly doubled the number of false memories its subjects formed.

Robert Nozick imagined an experience machine in 1974 that could give you any experience you wanted, writing a great novel or making a friend, perfectly and indistinguishably, for the rest of your life. He expected most people to refuse, on the grounds that they want to do things rather than feel that they are doing them, and to stay in contact with a reality beyond their own sensations. Felipe De Brigard reversed the experiment in 2010. He told subjects they were already inside the machine and asked whether they would unplug: 87% stayed when the outside life was described as a prison, and 59% stayed when they were told only that the life they knew was simulated. What people defend when they say reality matters for its own sake turns out, on this evidence, to be the life they already have. Three things do separate a physical world from a rendered one: a counterpart with an inner life who can consent or be wronged, a record held in common rather than generated per user, and consequences that survive a reload. None of the three is authenticity of feeling, which is what Nozick’s readers took the question to be about.

Bostrom asked how probable it is that we live in a simulation, and the biologists had already settled the part that can be settled: nobody sees the world unrendered. What remains is what the renders are for, whether a synthetic friend will do, whether a record can be held in common, and who owns the machines the renders run on.