None of this is real: we are NPCs building the next simulation

The uncomfortable odds#

Nick Bostrom’s simulation argument is not mysticism. It is three doors, and only one of them has to be true:

1. Civilisations almost never reach the point where they can run ancestor simulations.
2. Civilisations that can almost never choose to run them.
3. We are almost certainly already inside one.

People skip past the third door because it sounds like science fiction. The first two doors require you to believe that technological progress stops, or that curiosity dies, right before the interesting part. Look around. Neither of those feels like a safe bet.

If a post-human stack can run one high-fidelity world, it can run millions. Your lived experience becomes one draw from a very large urn. Base reality is the special case. The rest is render.

I am not saying I have proof. I am saying the argument is tighter than the shrug people give it at dinner. And the more we build with AI, the less abstract the shrug becomes.

It behaves like a game#

Call it a simulation if you want academic cover. Day to day, it feels like a game.

There are rules you did not write. Gravity, latency, death, taxes. There is a spawn condition you did not choose: parents, country, decade, body. There is a narrow field of view. You do not get the server logs. You get a first-person camera and a story your brain invents to keep you moving.

Games teach the same lesson. The map feels infinite until you hit the invisible wall. NPCs feel alive until you ask the wrong question twice. Physics feels absolute until a patch note changes the jump height. Nothing about that experience feels fake from the inside. That is the point. A good simulation does not announce itself. It just runs.

Glitches are optional. Déjà vu, synchronicities, the sense that the loot table of your life is strangely curated. You do not need any of that for the claim to hold. You only need the possibility that a higher stack exists, and that it is cheaper to simulate you than to be you.

We are already nesting worlds#

Here is the part that is not philosophy.

On this plane of existence, we are building simulations constantly. Not someday. Now.

Video games stopped being toys and became persistent economies with laws, status, and grief. Digital twins of factories, cities, and power grids let operators rehearse futures without burning the real plant. Climate models, traffic models, epidemic models: all of them are thin slices of reality spun forward under assumed rules. Autonomous driving stacks train in synthetic worlds because the real road is too expensive a teacher.

Then AI arrived and collapsed the cost of inventing the rest.

Large models do not just answer questions. They instantiate characters, rooms, motives, and histories that never happened. Multi-agent systems argue, plan, and fail together inside sandboxes. Generative video and world models are racing toward environments you can walk through and poke. We are not only describing reality. We are authoring runnable approximations of it, then dropping agents inside them to see what emerges.

That is nesting. A civilisation inside a possible simulation, building simulations of its own, and getting better at it every quarter. If Bostrom’s ladder is real, we are not spectators at the bottom. We are mid-level players shipping the next layer.

AI is the engine room#

Before AI, a convincing world needed armies of artists, level designers, and hand-authored dialogue trees. After AI, the bottleneck is compute and taste.

You can spin up a town that never existed, fill it with agents that remember last Tuesday, and watch them form habits you did not hard-code. You can train robots in a hallucinated warehouse until the policy transfers to metal. You can let a language model role-play a customer, a attacker, a patient, a voter, and stress-test systems that used to need real humans to break.

The philosophical sting is not that these worlds are perfect. They are not. The sting is that perfection is not required. A simulation only has to be good enough that the agents inside it cannot reliably prove otherwise. We already accept that standard for entertainment. We are approaching it for science, training, and social rehearsal.

So when someone says “none of this is real,” they are not only gesturing at the stars. They are describing the product roadmap. We are industrialising unreality, then plugging it back into reality through decisions, policies, and machines that learned in fake places.

What changes if it is true#

If this layer is a game, meaning does not vanish. It relocates.

You still feel pain. You still lose people. You still have to choose whether to be useful or cruel inside the rules you were given. Calling the map a simulation does not refund the cost of a bad run. It does change how you relate to certainty. Dogma looks sillier. Curiosity looks smarter. Hubris looks like a player who thinks the HUD is the universe.

It also changes how you look at AI work. Every digital twin, every agent sandbox, every synthetic dataset is practice for the thing the argument says someone above us already did. We are not waiting for the future to become weird. We are the weird, compiling.

I do not know if we are base reality. I do know we are behaving exactly like a species that would populate a universe with nested games the moment the hardware allowed it. That is either a coincidence, or a clue.

Play well either way. And notice what you are rendering.