This mini-essay is a refutation of the ideas expressed in Nick Bostrom's Are You Living In A Simulation? essay. If you haven't read that essay, this will make little sense to you. It is far from a rigorous or complete paper, and should not be taken as such.
For the record, I consider this to be the weaker of the two arguments presented here.
Mr. Bostrom seems to have drastically under-estimated the computing power required to run an ancestor simulation, by including only the minds of the humans in the simulation in the processing power required. This is patently ridiculous: humans interact with their environment, so the environment itself must be simulated.
For sufficiently primitive human societies, I suppose that all the extra stuff (the earth, the stars, the moon, the atmostphere, the animals, the plants, the oceans, and on and on and on) could be fudged badly enough to make the extra processing power negligible, but if one is going to theorize that we are in a computer simulation, standards are much higher. The simulation must account for everything that human scientists have been able to observe, which is a tall order indeed.
There are essentially two ways to go about this: one is to have the simulating computer run an AI which is capable of observing all the humans in the simulation at once and feeding them only the data they need (and keeping it all consistent, one hopes). We can't even imagine how much computing power such an AI would have to have, so I won't even try. I will suggest, however, that such an AI's computing needs would dwarf the computing needs of the human brains in the simulations by dozens of orders of magnitude.
Someone pointed out that I present no evidence for this view. There's a simple reason: I think it's a ridiculous thing to speculate on. He could be right, I could be right, we could both be incredibly, mind-numblingly wrong in one direction or another.
I simply think that an intelligence with the capacity to keep the state of billions of other intelligences in its head (and collate and manage those states and the states of those intelligences' environments and so on and so on) would be many, many times more complicated than all of the intelligences it is managing combined.
I have no evidence for this. I don't think evidence one way or the other on this issue is even *remotely* possible. It's a belief, nothing more, and I'm reluctant to pretend otherwise by attempting to argue it.
The much simpler way is to actually directly simulate the physics of the earth (one can probably fudge the stars pretty badly until the simulated humans leave earth, so I'll ignore that).
These assumptions are patently wrong, in such a way as to make the simulation much less computationally intensive than it would really have to be to account for modern science. They are also based on rather cursory research.
Some of the numbers above are being ignored, being added mostly to make it clear that this is an under-estimated.
The computer that Mr. Bostrom uses for his calculations can run 10E42 instructions per second. With our highest term of 10E55 calculations required to run an ancestor simulation, that's 10E13 seconds per simulation. That's approximately 317 thousand years per ancestor simulation.
Note, please, that the despite Mr. Bostrom's insistence that "The microscopic structure of the inside of the Earth can be safely omitted", that is not the determining factor here. The determining factor is actually the interactions of atoms and photons on the earth's surface and in its atmosphere, which we must simulate to provide any kind of ancestor simulation at all!
This makes it very, very unlikely that simulated people outnumber their real counterparts by any substantial margin.
I actually consider this part to be a much more compelling argument, as one can always handwave future computing technology to an arbitrary extent. It has, however, been pointed out to me that Mr. Bostrom's essay does not theorize a future with ancestor simulations, only the results of such a future. So this section, at least in some sense, isn't an argument against his paper, except in as much as I think the underlying scenario his paper is based in is unlikely.
In a future where ancestor simulations are possible, one would hope that moral, ethical, and legal protections have developed for sentient beings other than physical humans. In particular, one would hope that such protections exist for simulated and artificial intelligences.
Given that, an ancestory simulation is almost without question morally intolerable and would certainly be illegal.
Think about it: if we are in a simulation, our creators are responsible for every pain and discomfort that every person in this simulation has ever experienced. Hitler. Vesuvius. The Mongol Hordes. World Wars I and II. Starvation. Rape. Murder. If we are in a simulation, our creators are directly repsonsible for allowing those things to occur to sentient beings. This is a problem for exactly the same reasons that a psychologist can't raise a child without exposure to speech to "study the development of feral children". Or torture a person to see what people do under torture.
And even if it was legally possible to consent to such things in our current legal system, I sure a hell didn't consent to being created in a world filled with such immense amount of pain and suffering.
Supposing, for a moment, that such things were considered legally and ethically acceptable for some reason, bear in mind that you would never be able to turn the simulation off. Ever. Or at least, not until every sentient being had died. While a legal system might accept simulations that are not pardisical, it is unlikely to accept mass murder of billions of sentients. This makes things rather more complicated.
Any future sentients that wouldn't mind doing these things cannot be considered, in any way, to be human, and if they are all of "humanity" that is left, that must be considered the result of a whimper scenario, at best.
I found Mr. Bostrom's article perplexing and intruiging when I first read it, but it had always nagged at me that something seemed wrong. Having figured that out and written this page, I find that I can no longer put any stock at all in the idea of future cultures running vast arrays of ancestor simulations.