Being the title of a paper by one Dr. Nick Bostrum of Yale University
, due to be published in the Philosophical Quarterly. The full paper is available at www.simulation-argument.com/classic.html, but here's the argument "in my own words" (both because the original in my opinion didn't put the focus where I would like it and digressed overly, and because (of course) cut and paste writeups will die. And also because I'm not sure the original actually works, or at the least I think it jumps over technicalities I'd rather deal with. I advise you to read the original as well as this).
The argument aims to show that this thing you call reality is most likely a simulation. Which of course is a ridiculous notion crazy enough that only philosophers and Hollywood2 would consider it, but hear this out. More precisely, the argument shows that at least one of the following holds
- 1 - Humanity will almost certainly die out before reaching a stage of sufficient technological sophistication to be able to conduct ancestor simulations
- 2 - Humanity will reach this stage, but for some reason will with a high degree of probability choose not to conduct such simulations
- 3 - We are, with little doubt, living in an ancestor simulation.
Well, first of all, what's an "ancestor simulation"? Nick defines it as "a simulation of the entire mental history of humankind", which is a bit loose, but basically it's a simulation of the ancestors of our future super-teched descendents ("posthumans" being the transhumanist term he uses), ourselves included (assuming this is reality), simulating mental states fully and the environment sufficiently that it would seem real enough not to affect those mental states. That this is possible is something which the article goes to great lengths to demonstrate, but suffice it to say that as long as you accept that human minds can be simulated exactly through information processing with the need for neither actual neurons nor "souls", then assuming a whole lot of knowledge and ability on the parts of our posthumans, it can. Moreover, they should be relatively easy to run. According to his calculations, a planetary-mass computer could run an entire ancestor simulation in 10^-7 seconds.
So assuming our posthumans, or least some amongst them, feel like running such simulations, very very many will be run. It wouldn't take a coherent effort on behalf of the civilisation, either - he's expecting a whole lot of these planetary computers, so even if such simulations are considered frivolous or immoral, someone somewhere is likely to run the odd billion. Consider how just how many hours of valuable mainframe time and processing power have been spent running Nethack3... So anyway, now we're at the stage of having rejected (1) and (2), and we want to demonstrate (3).
This is where my analysis parts company with Nick's. Here's how my version runs - Suppose that this is either reality or an ancestor simulation. Now by ancestor simulation, I mean an approximately faithful simulation of the past of reality's future, faithful in particular in that similar evidence is available pointing to the likelihood of ancestor simulations being run in what appears to be the future. So whether this is reality or such a simulation, we have in either case that many ancestor simulations will be run in reality's future (as compared with either now, if this is reality, or else the moment of reality which is being simulated).
Now, we are hypothesising that the simulations are faithful enough that we can't differentiate between reality and simulation, so if we invoke some version of the anthropic principle (or possibly just the principle of indifference - I'm not too sure on this point but I think the fact that the simulations are only similar to reality and not identical means indifference is insufficient) we find we should reason as if we were chosen randomly from all observers in reality and simulation. And since we have many many simulations and just the one reality, that gives a very high probability of us being in a simulation and a very low one of this being reality.
So we've got that if we assume this is either a reality or an ancestor simulation, then the second is far more likely. Which seems, intuitively, to imply that the probability of this being reality is very low full stop, though I'm not sure if this technically counts as a proof... any ideas here much welcomed. The problem is that although this seems intuitionally reasonable, causality means that this situation never actually arises in the real world so we have no good reason to trust our intuition on it. I think.
So, QED, kindof. Sorry to belabour the above points so - the paper pretty much glosses over them but my feeling is that they are important. Personally, despite initial misgivings, I became quite close to being convinced by the argument, before deciding it was probably bollocks after all. Even if it did work, I'd suggest (1) as the most likely of the alternatives.
- The paper at www.simulation-argument.com
- Nick's book "Anthropic Bias - observation selection effects in science and philosophy", the first 5 chapters of which are freely available for all your anthropic principle related confusion needs at www.anthropic-principle.com
- A New Scientist article on the paper, 27 July 2002
1 - Soon (at the time of writing) moving to Oxford University
2 - Please don't talk to me about the Matrix. I have never seen it, and don't care. Thank you.
3 - So yes, we could quite plausibly be the result of computer science geek's time-wasting activities. Which means, of course, that we monitor monitors are worshipping our all-powerful God with every tap and click. A better excuse for noding I'd like to see.