If This Is a Simulation, Someone Is Paying the Electric Bill

Look at the screen you're reading this on. Somewhere inside it, a chip is flipping billions of tiny switches every second to draw these letters. Every one of those switches costs energy, and the device grows slightly warm because of it.

Now consider the possibility, taken seriously by some philosophers and physicists, that the room around you, your hands and the light you are reading by are drawn the same way, by a computer running somewhere outside everything you can observe.

If that is true, someone is paying for it. Every particle, every collision and every photon would have a running cost. And physics has a precise rule for what the cheapest possible computation costs.

I want to give the simulation argument a fair hearing, then follow the bill: the argument itself, the physical price of information, the problem with simulations inside simulations, why no one has seen a glitch, and one detail about simulations that I think most people overlook.

That last detail was the first thing I wrote down when the idea started bothering me:

"It's so interesting how that has actually come from the fact that simulations themselves are modeled after reality and a lot of people forget that."

We'll come back to it at the end. First, the argument, because it is stronger than its reputation.

The Argument Is Better Than It Sounds

In 2003 the philosopher Nick Bostrom published a short paper that did not claim we live in a simulation. It claimed something more careful.

Suppose a civilization eventually gains enormous computing power. Suppose it is interested in running detailed simulations of its own ancestors, and runs many of them. Then simulated minds would vastly outnumber original ones, and any given mind, including yours, would most likely be one of the simulated ones.

So, Bostrom argued, at least one of three things must be true. Civilizations almost never reach that stage, or they almost never choose to run such simulations, or we are almost certainly simulated. He did not say which. The force of the argument is that rejecting the third option means accepting one of the first two.

When I first met it, I could not find the flaw. It rests on probability and arithmetic, not on mysticism. So I went looking for the cost.

Every Bit Has a Price

In 1961 the physicist Rolf Landauer, working at IBM, showed that computation is physical. Erasing a single bit of information, resetting a memory cell so it can be reused, must release a minimum amount of heat into the surroundings. At room temperature that minimum is about 3 × 10⁻²¹ joules per bit. Tiny, but never zero. Experiments in 2012 measured heat release approaching that limit.

That gave my unease a name:

"Due to the cost of computation principle that I forget the name of, isn't it unlikely that simulation theory aka turtles all the way down is valid?"

Any simulation that keeps running has to process and overwrite information, so it has to pay Landauer's price. A simulation of a whole universe, tracking every particle, would need a power source and somewhere to dump the heat.

Simulations All the Way Down

The bill grows when you follow the argument to its natural conclusion.

If we are simulated, and our own descendants eventually run simulations, then our simulation contains simulations, which may contain more. Every level is paid for by the level above. Each layer adds its own energy demand to the one beneath it.

The honest counterpoint is that the top level might be far richer than ours, with physics we cannot imagine, and could afford what looks impossible from inside. That is fair. But it moves the question rather than answering it. Something at the top has to be real, not simulated, and it has to pay for everything below.

Why No Glitches?

There is a second, more ordinary problem:

"In that case how come we have never experienced any single computation glitches in our universe?"

The usual reply is that a pause in the outer computer would be invisible from inside, since our clocks would pause too. That is correct. But it makes the hypothesis very hard to test, and a claim that can't be tested is hard to weigh.

There is also a practical clue. In 1981 Richard Feynman pointed out that ordinary computers struggle to simulate quantum systems, because the resources needed grow explosively with size. His conclusion was blunt: "…nature isn't classical, dammit, and if you want to make a simulation of nature, you'd better make it quantum mechanical."

We now have early quantum computers. Meanwhile, a landmark 2015 simulation by the Blue Brain Project modeled a small piece of rat cortex with about 31,000 neurons. Nature at its smallest scale turns out to be extraordinarily expensive to copy, and even a sliver of a living brain is a major project.

The Copy and the Original

That brings us back to the line I started with.

Every simulation we have ever built is a model of something: weather, aircraft, galaxies, economies. Simulations are copies, and copies come after originals. Then we turn around and suggest that the original, reality itself, might be a copy.

To me that looks circular. The idea of a simulation only makes sense because there is a real world to model. So the hypothesis can never remove reality. At most it moves reality up one level, to wherever the computer runs, where the bill still has to be paid and the same question waits: what is that level made of, and why does it exist at all?

So here is the question I'd leave you with: if every simulation has to be a copy of something real, what do you think this one would be a copy of?