If the World Can Be Computed, Could We Be Living in a Simulation?

Suppose AI solves one of the Millennium Prize Problems. What happens next?

This is a thought experiment. As of September 14, 2026, the Clay Mathematics Institute lists only the Poincaré conjecture, solved by Grigori Perelman, under solved problems.^[1]

If AI produced a proof that survived scrutiny, we would have a long-awaited answer. We might also reconsider the problems that still resist us: is the world too difficult to understand, or have our abilities simply not been sufficient yet?

If AI kept uncovering deeper mathematical relationships and physical laws, we might predict more accurately what we could previously only observe. Take the thought further: what if the whole world, including us, could be computed?

Elon Musk has put the chance that we live in “base reality” at “one in billions.”^[2] Base reality means the level of reality that is not itself being simulated by another world.

It is easy to connect the ideas. As AI understands more of the world, the world might begin to resemble a program that can be run. If it can be run, could we already be inside one of those runs?

But why would being able to compute a world establish that we live in one?

How far is one solved problem from computing a world?

One Millennium Prize Problem concerns moving water. The Navier–Stokes equations describe fluid motion. The existence and smoothness problem asks whether suitable smooth solutions continue to exist under specified conditions, including three dimensions and incompressibility, or whether a case of breakdown can be established.^[3]

If AI solved it, we would understand these equations better. To predict every part of a particular river tomorrow, however, we would still need information about its channel, initial flow and external influences, a usable computational method, and grounds for applying the model to that situation.

The proof would answer a mathematical question. The conditions of that particular river would still need to be measured and checked.

For the whole world, far more information would be needed. We would have to know its governing equations and current state, then finish the calculation in a finite time. Computing devices have physical limits too: energy and other physical conditions constrain processing speed and information capacity.^[4] AI could improve the methods without automatically gaining the missing information and resources.

What would “the world can be computed” mean? An approximation over a limited range requires different capabilities from a complete reproduction of every process. Giving the probability of an outcome is also different from predicting what will happen in each individual event. The broad phrase can hide those differences.

Solving one problem would tell us more about that problem. Whether the whole universe is computable would remain open.

Still, suppose future intelligence overcame those obstacles. A computation really could fully implement a world like ours, with observers inside it. Would Musk's conclusion follow then?

Where does Musk's probability come from?

At the 2016 Code Conference, Musk started with the progress of games: from Pong to increasingly realistic multiplayer games, then to future simulations potentially indistinguishable from reality. He imagined vast numbers of such simulations and concluded that our chance of occupying base reality was tiny.^[2]

The probability depends on how many observers like us exist inside simulations. Increasingly realistic games give us a starting point for considering whether simulations could become indistinguishable from reality. To say we are probably inside one also requires large numbers of simulated observers.

Consider a smaller version: a million observers have experiences similar to mine, but only one lives in base reality. The others are simulated. If I have no other clues, and accept an equal chance of being any one of those observers, then it makes sense to think I am more likely to be simulated.

But we put those million observers into the thought experiment. The resulting fraction follows from the numbers we assumed; it is not yet a measurement of the universe.

Nick Bostrom's simulation argument makes the conditions more explicit. Under assumptions including that suitable computation can produce consciousness, he considers several possibilities: civilizations rarely reach the necessary technological stage; those that do rarely run many such simulations; or most observers like us are simulated. Moving from the fraction of observers to a probability about our own position also requires a self-location assumption: how, without further clues, should we use those proportions to judge where we are?^[5]

It becomes clearer what AI might change. If it makes the technology needed to create such worlds more feasible, we have reason to reassess that premise. Whether civilizations get that far, want to bear the costs, or create subjects with experiences would need separate investigation.

Before accepting or rejecting Musk's conclusion, I want to know which of these conditions has gained new support.

Once we can build a world, we may begin to question our own

Even if we could not calculate the probability, creating a world could make us look at our own differently.

Suppose we really created such a world. Its observers conduct experiments and discover stable relationships governing motion, a limit on information transmission, and values that remain constant however often they are measured. They call these the laws of nature.

We know that some of them come from code and parameters. We can even change the settings and watch what happens.

Looking back from that position, our own world begins to seem less familiar. Might the physical constants we take for granted have a source we have not seen? Could some deeper mechanism determine what can happen and what never can?

As the imagined creators, we could study the laws inside while seeing how the world was implemented outside. Studying our own world, we have to look for clues from within.

Creating a world whose inhabitants had experiences would at least establish that such worlds can be made. Whether we also lived in one would still require evidence about our own situation. The world we made could not answer that for us.

“Virtual” can suggest that everything is fake. But if those observers truly had experiences, their pain, relationships and passage of time would not become empty appearances because their world had an external implementation. We would need to explain what supports those experiences; the experiences would still be there.

That source is what interests me. Does this world operate as it does simply because that is how it is, or because something at a deeper level makes it so?

Even if there is a designer, how much does it decide?

A program brings a programmer to mind. A setting suggests someone who chose it for a purpose. We know that way of working and naturally use it to make sense of what we do not yet understand.

The source of the laws need not resemble a person. It might be a more fundamental mechanism, or something we do not yet understand. I use “higher dimension” here loosely, to mean a deeper level we cannot currently observe directly. It is not an established claim that an entity inhabits an extra spatial dimension.

To call it a will, we would also have to explain how it chooses, what it prefers and why it makes something happen. Causing a result does not establish that it wanted the result.

Even a hypothetical designer could participate in different ways: write only the rules, or choose the initial conditions too; leave the world alone after starting it, or keep watching and making changes. How much it decides depends on which of these it does.

Suppose we ran a model simply to see what a set of rules would produce. Every collision could follow from those rules and conditions without being an outcome we specifically wanted. Even if the complete conditions determined the model's trajectory, each consequence would not necessarily correspond to a separate wish.

“The world may be simulated” therefore leaves open whether everything that happens is intentionally arranged by some will. That further question adds purpose, choice and intervention. If we merge them, we may write the designer's work log before finding the designer.

Could AI help us discover what is outside?

If AI could eventually solve difficult mathematical problems and understand physical laws far more precisely than we do, could it find where those laws come from?

I hope it could notice differences we missed, design new experiments and rule out explanations. But however strong the reasoning, it still needs information that distinguishes the possibilities.

For example, the same world might run on two different external systems. If every experiment available inside yielded the same distribution of results in both cases, those results alone could not tell us which system was being used. Studying the internal laws more precisely would not add information that distinguished them.

Our actual situation may be different. Perhaps there is something we could observe, and we have not yet thought of how. AI might help us find a way, turning the broad question of whether our world is simulated into specific comparisons.

If a particular simulation scheme would leave a specific deviation, we could look for it and compare the result with other physical explanations. An anomaly would first call for checks of our measurements and models. Supporting the simulation explanation would require more successful predictions; claiming an anomaly after it appeared would tell us little.

If stable laws mean the design is good, anomalies mean someone is making edits, and finding nothing means the designer hides well, the story can always continue. But if it accommodates whatever we observe, observation gives us little help in deciding whether it is right.

What I can say for now is that such mathematical breakthroughs would advance our understanding of those problems and might make simulated worlds more feasible. They could advance some premises while leaving consciousness, observer numbers and our own situation unresolved. Musk's claim still needs answers to those questions.

If AI produces that proof and it survives scrutiny, I want to see which testable questions it can help us ask next. We might find deeper laws, perhaps even clues to an external implementation.

And the original question might be waiting one level further on: what makes that deeper world run?