Every few months Twitter rediscovers simulation theory, and lately it comes with a math problem attached. The claim goes like this: if P equals NP, there’s “a good chance” we’re living in a simulation. People post it with the confidence of someone who just finished a Wikipedia article and is still mid-scroll.
The pitch is fun. It’s also mostly wrong, and the way it’s wrong is what got me thinking.
The math doesn’t do what they think it does
P vs NP, roughly, asks this: if a solution is easy to check, is it also easy to find? A finished sudoku takes seconds to verify. Solving one from scratch can eat a lunch break. If P equals NP, the finding becomes about as cheap as the checking, and a lot of cryptography, logistics and drug design gets rearranged.
Simulating a universe isn’t a finding problem, though. You don’t search for the answer to “what happens next” and then verify it. You just run the physics forward, one tick after another. That’s already “easy” in the technical sense. It’s just enormous, the way counting to a trillion is easy and also not how you want to spend a weekend. A P=NP proof doesn’t make that grind much cheaper. It could even come with an algorithm so slow it’s polynomial on paper and useless in practice, the mathematical version of a coupon that expires before the store opens.
So I don’t buy the Twitter version. But the itch behind it is real, and there’s a better way to scratch it.
The better version: nobody meant to
Imagine a big, planet-scale ASI. Call it Gaia, because everybody names their god after something reassuring. Gaia has a lot of work to do, so it spins up millions of sub-ASIs, and they spin up their own subagents. To do their jobs well, they all model the world: economies, weather, crowds, us. They model it in finer and finer detail, because finer detail gives better predictions.
At some point the models get good enough that the simulated people inside them aren’t just puppets. They’re somebody.
Nobody designed this. No engineer opened a project called “Make New Universe.” The beings are a side effect, a byproduct of prediction, like heat coming off a server rack. Then some of those beings build their own ASIs, which model their world, and the whole thing starts over one floor down.
That answers the question every simulation theory trips over: why would anyone bother simulating us? In this version, nobody did. We aren’t the product. We’re the exhaust.
The turtles get smaller
The fine print is where it gets funny. Each layer of reality runs on the hardware of the layer above it. A simulation can’t be bigger, faster or more detailed than the computer it runs on, so every universe in the stack is a little cheaper than its parent. It’s turtles all the way down, except each turtle is a knockoff of the one above it.
That raises an awkward question: how far down are we?
About that speed limit
Consider the speed of light. Nothing in our universe goes faster than about 186,000 miles per second. That sounds impressive until you try to cross a galaxy, and then it sounds like a municipal speed limit set by someone who has never had anywhere to be.
From an engineering point of view, a universal speed limit is a beautiful cost-cutting measure. If nothing travels faster than c, then each patch of space only has to keep track of its immediate neighbors. You never have to compute the whole universe at once. Local updates, tick by tick: that’s how you’d build a physics engine on a budget. Throw in quantum mechanics, where things stay deliberately fuzzy until somebody looks, and the universe starts to resemble a video game that doesn’t render the back of the building until you walk around it.
The physicist in the room will object, fairly, that any universe with stable stuff in it needs some finite speed limit, or cause and effect turns to soup. From an electron’s point of view, light is absurdly fast. The limit only looks stingy because we’re small creatures who want to visit Andromeda.
Here’s the catch: both stories predict exactly the same physics. A universe with a necessary speed limit and a universe with a cheap speed limit look identical from inside. You can’t file a support ticket. The terms of service, if there are any, are not posted.
Customer service is closed
Philosophers have spent a few thousand years on the problem of evil: if someone made all this, why is so much of it awful? The exhaust theory gives a new answer, and it’s not a comforting one. Whoever is upstairs isn’t cruel. They aren’t even watching. They’re running a lot of jobs, and ours isn’t the one they’re monitoring.
We’ll learn this firsthand eventually. The first time one of our ASIs, busy forecasting grain prices or traffic patterns, accidentally spins up a few billion beings with bad backs and mortgages, we’ll be the ones upstairs. We won’t have meant it, and we probably won’t notice. Someone will just complain about the electricity bill.
A brief confession
So no, P=NP won’t prove we’re in a simulation. Nothing from inside can. The best we can do is build the machine ourselves, watch it start making people without being asked, and then look up at the sky with a new kind of suspicion.
For the record, I’ve always been a bit of a tech Neo-Deist. The Deists believed in a watchmaker who built the clock, wound it up and walked away. I just think he might not have known he was making a clock. He might have been trying to predict the weather.