Built on Facts

There’s an interesting post over at Sentient Developments about the simulation argument.

The SA essentially states that, given the potential for posthumans to create a vast number of ancestor simulations, we should probabilistically conclude that we are in a simulation rather than the deepest reality.

Most people give a little chuckle when they hear this argument for the first time. I’ve explained it to enough people now that I’ve come to expect it. The chuckle doesn’t come about on account of the absurdity of the suggestion, it’s more a chuckle of logical acknowledgment — a reaction to the realization that it may actually be true.

In a little more detail, the idea is this: right now we humans run simulations in our computers. Sometimes they’re very precise models of scientific phenomena, designed to study the universe or help engineer complicated pieces of equipment like airplane engines or submarine propellers. Some simulations are less complicated and we run them for fun, like The Sims, Sim City, or Roller Coaster Tycoon. Eventually – or so the argument goes – our understanding of biology and our computing power will grow to the point that we can run neuron-by-neuron or atom-by-atom simulations of people who are thus indistinguishable from you and me.

And nothing stops those people from building their own computers in their own simulated world, simulating their own sub-Sim people as well. And so on.

Given all these simulations nested like Russian dolls, what are the odds that we’re the people on top – the “real” people in the real world? Therefore we’re probably actually simulated beings being processed on a computer in somebody’s PC.

Ok, sounds plausible enough. Looking deeper though, there’s some good reasons to think it’s not true. The biggest is our own experiences with simulation. In living creatures a lot of very complicated chemical processes are ongoing all the time. To simulate these processes with any degree of precision – and not even from first principles – requires a vast, vast amount of computing power. The Folding@Home project uses the services of millions of linked computers just to adequately simulate the motions of individual protein molecules. To simulate the atoms in an entire human being would require an improvement in computational power by a factor of Avogadro’s number. Multiply by 6 billion for the rest of the people and you’re still nowhere close to done. You still have an entire planet or more to simulate in order for them to have a place to live.

Or maybe you’re willing to cut corners and only simulate things on a neuronal level. That doesn’t help all that much. Not only is neuron simulation very difficult, but you still have to simulate the external world in great detail in order to be able to feed those neurons some sensory data. And you have to coordinate between all six billion neuronal simulations and pretty soon you’re having the same problems as a full atomic simulation. Now there’s been some success simulating smaller neural systems, and that will certainly lead to success with larger systems. But there’s a big difference between simulating a slice of rat cortex and keeping track of the vastly, vastly more complicated human cerebellum – to say nothing of the fact that you still have to simulate a world for your Sims to live in and build their own Sims. A simulation of the neurons alone won’t work.

And once your simulations build computers and start simulating more people, things get out of hand. It’s not at all far-fetched for those six billion to want to simulate their own worlds, so you’ve squared the computational capacity you need. Keep going and you further exponentiate. The top-level computers are going to start smoking and sputtering pretty soon.

If indeed the first-level simulation is even possible itself. The speed of light, the granularity of atoms, and the fuzziness of quantum mechanics have already begun to stall out our own ability to make computers faster. There’s a reason clock speeds have been stuck around 3GHz for a while now, and though they become more efficient and are tied together in parallel the speed of light puts hard limits on how fast they can transfer information between each other. Now there’s no requirement the simulation be in real time, you might take years to simulate a second in your SimUniverse. But you’re going to run out of seconds eventually, the universe isn’t going to last forever.

Well, maybe higher-order Sims who are simulating us have physical laws better suited for building computers. Maybe, but at that point it starts to sound less like the Simulation Argument and more like the Simulation Late-Night Stoner Discussion.

And furthermore all this assumes that conscious experience can be simulated in the first place. I’m in a minority in my view, but I suspect it can’t be for philosophical reasons not far removed from the zombie argument. Which itself has no lack of legitimate criticisms, but that’s a topic for another day.

Back to the article. What are the consequences of the truth or falsehood of the simulation argument?

If it’s true:

If we were ever to prove that we exist inside a simulation, it would be proof that the transhumanist assumption is correct — that the transition from a human to a posthuman condition is in fact possible. But that will be of little solace to us measly sims! The simulation — er, our world — could be shut down at any time. Or, the variables that make up our modal reality could be altered in undesirable ways (e.g. our world could be turned into a Hell realm).

Pretty typical stuff. And though the article’s next paragraph skirts the idea, I note that accepting the simulation argument isn’t all that far removed in concept from accepting the ontological-style arguments for God. Skepticism about God should (in my opinion, though it probably doesn’t in practice) imply skepticism about nested simulations. For that matter belief in God should imply skepticism about nested simulations as well, since an omniscient being wouldn’t need to simulate anything in the first place.

If not true:

Now, on the other side of the virtual coin, should we ever prove that we are not in a simulation, that would also be bad. It would be potential evidence that the transition to a posthuman condition may not be possible…

Similarly, we should be disturbed that we are not in a simulation because it may imply that we don’t have a very bright future — that civilizations destroy themselves before developing the capacity to create simulations. Otherwise, we have to take on a exceptionally optimistic frame and assume that we’ll survive the Singularity and be that special first civilization that spawns simulations. Again, a probabilistically unsatisfactory proposition.

Or as I said, it could mean merely that large-scale simulation is not feasible. You can still have a bright future without that kind of computational capacity. It does mean the rapture-of-the-nerds style Singularity is less plausible, but I don’t see that as a great loss. Technology doesn’t need to be magic to be cool and useful.

So that’s something to think about. It’s not observable, testable, or repeatable and so it’s not really worth worrying about too much. But I’d say it’s still interesting to think about.


  1. #1 D
    January 19, 2009

    “I note that accepting the simulation argument isn’t all that far removed in concept from accepting the ontological-style arguments for God.”

    What’s the connection?

  2. #2 Alex Besogonov
    January 19, 2009

    A couple of interesting thoughts:

    1) It’s too hard to emulate everything ab initio. But it’s probably much less hard to have special-case emulation for humans and animals. I always liked an argument that the observer effect in quantum mechanics is just an artifact of emulation – usually we are emulated using classical mechanics, but our ’emulator’ has to low down to the level of QM effects when we try to look deeper.

    2) Possibility of hypercomputers. Maybe we’ll someday be able to create hypercomputers which will have infinite speed and memory capacity.

  3. #3 mad the swine
    January 19, 2009

    The SA essentially states that, given the potential for posthumans to create a vast number of ancestor simulations, we should probabilistically conclude that we are in a simulation rather than the deepest reality.

    The probability of two specific, unrelated individuals (my mother and my father) meeting and hooking up, and mixing their genes in such a way as to create a person with my particular DNA, is very, very small indeed. Therefore, probabilistically, I don’t exist.

    See the problem there? 🙂

  4. #4 Epicanis
    January 19, 2009

    Functionally, I don’t see this as being any different than Berkeley’s assertion that we’re all thoughts in the mind of God™.

    It’s about as useful, too, given that in either case there’s no way to actually verify it. It’d be like trying to directly observe the back of your own head (which, as Terry Pratchett pointed out, is only possible in extremely small universes…)

    Unless of course we are simulations, and the programmers left some very obvious “debug” statements in there that we might find.

  5. #5 Romeo Vitelli
    January 19, 2009

    I suppose I could always kick a rock and say “I refute it thus” the way that Samuel Johnson reportedly did when to disprove Berkeley’s idealism. I’d rather not break a toe over this though. Just consider that rock kicked.

  6. #6 Matt Springer
    January 19, 2009

    “What’s the connection?”

    The argument in both cases goes something like this:

    1. It’s possible.
    2. [handwaving]
    3. Therefore it’s true.

    Which isn’t to say that the handwaving part isn’t perfectly legitimate or convincing (no less than Godel argued for it re: God), just that it’s not going to convince anyone who isn’t already convinced.

  7. #7 andy.s
    January 19, 2009

    We don’t really have to simulate every cellular process or virtual particle exchange. We get an enormous increase in throughput by only using a full blown fine grained simulation when one of you Sims happens to looking in a microscope or running a particle accelerator.

    The rest of the time we just make do with a coarse grained simulation with random perturbations. Which I believe you Sims call “Quantum Indeterminacy”. We really thought the discover of that would have tipped you off, but, oh well.

    And we think it’s just adorable the way you believe that we’re bound by the same speed-of-light and complexity limitations of your universe. I was part of the commitee that designated the speed of light and also the +— Minkowski signature of your space-time. (I actually wanted a Euclidean metric but was voted down).

    Of course, now that I’ve told you, I’ll have to reboot you.

  8. #8 Matthias
    January 19, 2009

    The simulations don’t have to run in real time (and unless we/they model them quite after ourself, there won’t be any meaningful “real time”), which may make them a little more feasible.

    This also leads to possibilities like “ok, let’s set their speed of light lower, so the observable universe is a lot smaller than ours, so our superduper galaxy-sized computer can handle it quantum by quantum”.

    I actually don’t really care whether the universe is “real”, a sneeze of the Great Green Arkleseizure, the dream of a lonely Boltzmann Brain or whatever. It’s interesting and a helluva lot to explore anyway.

  9. #9 D
    January 19, 2009

    – 3. isn’t what the simulation argument concludes. What it concludes rather, is that simulation is likely or plausible or something…
    – Isn’t the fascination with the ontological argument because it claims that for a certain kind of entity, possible existence implies necessary existence?
    – I think you’re assuming the simulating universe, so to speak, has the same laws of nature as the one it simulates (ours). This seems unmotivated.
    – You left out 4. ??? and 5. Profit!

  10. #10 Matt Springer
    January 19, 2009

    In regard to the “different physics” argument for the necessary computational capacity which several people have mentioned, here’s some more detail on why I think it’s a non-starter.

    The entire point of the simulation argument is to create a logical argument for the probability of us being simulated. I.e. “We can simulate, therefore others can simulate.” But if we can’t in fact simulate adequately, the argument fails. If you have to invoke new and unobservable physics to save it, you might as well invoke (as #8 put it) “a sneeze of the Great Green Arkleseizure, the dream of a lonely Boltzmann Brain or whatever.” With different physics, you’ve lost any reason to prefer nested simulations above any other bizarre metaphysical speculation.

  11. #11 andy.s
    January 19, 2009

    There actually isn’t a series of nested simulations, it’s just you.

    The real universe is a 196,884 dimensional entity with gauge fields based on representations of the Monster Group. Your little SU(3)xSU(2)xU(1) space is just a toy model built for a science fair.

    Bizarre metaphysical speculation, indeed. Coming from a universe with only three generations of leptons, that’s rich!

  12. #12 complex field
    January 19, 2009

    To expand upon the idea about “real time” execution, who remembers when a garden-variety vector “cartoon” depicting a planet’s motion took a week to compute and render on a Cray, just so its full length of about 15 seconds could be shown on NOVA?

    We, as simulations, would have no idea as to whether we were executing in real time or in fits and starts on Deep Thought or some hyper-dimensional microsoft-crashing-all-the-time PC at a science fair.

    Since we would have no awareness of how we were executing, the question then becomes moot. We then are left to the variables provided to us, according to the programmer’s code.

    So what? So, sit back and execute. We have a bunch of cool stuff to explore, regardless of why we are here.

  13. #13 kermit
    January 19, 2009

    Subroutines in a virtual world; a brain in a tank fed with fake input; a psychotic in a padded room somewhere; living in a 6000 year-old universe, made to *look older by a trickster god. These are all alike in these ways: they are illusions, they are conceivable, they are untestable, and we have no empirical reason to think any of them are true.

  14. #14 Joshua Zelinsky
    January 19, 2009

    There are ways we could test limited versions of this hypothesis. With some ideas about the architecture of the simulating machine we could guess what errors in the computation would look like. Unfortunately, there’s no reasonable way to get any information about the architecture or to get any idea what the computational error rate is. (How many protons can this simulated universe have before there’s an overflow error and we end up with a negative total?)

  15. #15 beautox
    January 19, 2009

    This occurred to me; if we are Sims then it’s likely (or possible) that the beings that simulate us would have conquered many if not all of the problems that plague us, like limited lifespans, conflict, suffering and so on.

    Thus these beings might have created us so that they can experience these very things. (Living forever could get boring without diversions)

    The saying goes “If god exists, why did he create all this suffering”. Well, this argument suggests that it was all done on purpose.

  16. #16 Matthias
    January 19, 2009

    @#14 maybe like this?

  17. #17 Frasque
    January 19, 2009

    All I can say if someone is simulating my life, they’re pretty goddamn evil. Like those folks who go to great lengths to build death traps for their sims.

  18. #18 Lord
    January 19, 2009

    Any reason a simulation couldn’t be a reality as well?

  19. #19 Richard Eis
    January 20, 2009

    But isn’t a perfect simulation, no longer a simulation but the object itself?

  20. #20 Stephen
    January 20, 2009

    Occam’s Razor.

    It’s simpler to believe that we’re running in an imperfect simulation than it is to believe in quantum physics, dark energy, dark matter.

    Hate to throw the new physics thingy back at you, but I reckon the parent universe is continuous, so has infinite memory and computing resources available at all scales.

  21. #21 Dunc
    January 20, 2009

    Even if we accept that the required level of simulation is possible, that still doesn’t mean that it’s worthwhile. You can run a weather forecasting model years out into the future if you want, but it’s not going to give you useful results that far out – even if your model is perfectly detailed. As far as I can see, ancestor simulation can’t actually tell you anything useful that you don’t already know. I suspect that anyone with the ability to do such a thing isn’t going to waste their time on it.

  22. #22 Cannonball Jones
    January 20, 2009

    I’m always suspect of anyone arguing from what is logically possible to what is probable. There are just far too many hidden assumptions in the argument to make it hold water i.e. is it actually possible to make such simulations, would we notice a difference, etc, etc. I love the argument, it awakens my ex-hippy student side, but it’s really no more than a discussion point for late-night pot sessions…

  23. #23 catgirl
    January 20, 2009

    I wonder what happens when The Programmer who simulated us logs out of the game or turns off his/her/its/? computer. I guess we just wouldn’t notice and things would pick up at the next second when we’re turned back on.

    Also, what if we aren’t a simulation, and are actually someone’s dream? I think that possibility is just as likely.

  24. #24 complex field
    January 20, 2009

    @ #23 Am I a man dreaming he is a butterfly or a butterfly dreaming that I am a man dreaming that he is a butterfly?

  25. #25 Thras
    January 20, 2009

    It’s a bit easier than that. You don’t actually have to simulate neurons. You just have to simulate a reasoning creature that thinks it’s made of neurons. And you don’t have to simulate all the complex phenomenon it sees, you just need to make it think or remember it’s seen a bunch of complex stuff.

    It’s possible that the core computational complexity of a reasoning mind could run on a 286 — it’s just that we have no idea how that core part really works.

  26. #26 Bob Hawkins
    January 20, 2009

    Not only are we a sim, but it’s being run on inadequate hardware, on software written by a mediocre at best programmer. Look around you.

    I mean, the Arizona Cardinals in the Super Bowl? Obvious glitch. The same constant, pi, being used for all sorts of unrelated purposes? Sheer laziness. Gilbert Gottfried has a career? Give me a break!

  27. #27 Bob Hawkins
    January 20, 2009

    “It’s possible that the core computational complexity of a reasoning mind could run on a 286 — it’s just that we have no idea how that core part really works.”

    Heck, I have relatives whose mental processes can be simulated under DOS by setting the prompt to “HUH>”.

  28. #28 greg
    January 20, 2009

    Note writer’s (improper)use of 6.023 X 10*23 in 5th para. Ron

  29. #29 Miguelito ... aka seahunt
    January 20, 2009

    I like, I like, but I think there may be critical assumptions you made that may be flawed.
    First when looking at this people seem to skip the context of object programming. I think it might be able to overcome much of the problem.
    More importantly you mention the problem of limitation of computational power. There is a way that could be removed and there are a number of hints to it in the literature of history. It’s a problem of a being, you or I, choosing to be part of a simulation. Take Second Life. It is limited, but if each entity in Second Life had their own processing power within the simulation it would change things.
    This refers to some work of my own. I think it applies to your essay. You might find what I wrote to be interesting. Educate yourself on “Lobster Hunting” and you might find some thoughts you have never seen.

  30. #30 Anonymous
    January 20, 2009

    It’s a sim, but only of one concious entity – me. You are just some data input to my sim, sorry!

  31. #31 Alan Kellogg
    January 21, 2009

    Who says the simulator has to live up to our capabilities? What we know is not all there is to know. What we can do is not all there is we could do. Life is a long, unending process of learning new things, and how to do new things.

    According to what we now know the possibility we are a simulatin isn’t likely. Additional knowledge could change that conclusion. We don’t know, yet.

    If we remember nothing else, let us remember that we are not all that and a plate of chocolate chip cookies too.

  32. #32 Chris
    January 27, 2009

    I think an immediately prior post beat me to this. You don’t have to simulate a universe, or even 6 billion people.

    You only have to simulate me. Everything else can be eye candy. The rest of the model doesn’t even have to be 100% reliable as long as I’m simulated to accept the occasional glit the occasional glitches as something unnoteworthy.

    You can even take this a step further. You don’t have to simulate me from birth. You could start my simulation a few minutes ago as long as I have the necessary “memories” of an earlier period. Again these “memories” can be quite sketchy — how many of us give much thought to what we did in HS, much less wonder how reliable those memories are. Or remember what we ate for dinner two nights ago?

  33. #33 Bob Smith
    February 11, 2009

    “The entire point of the simulation argument is to create a logical argument for the probability of us being simulated. I.e. “We can simulate, therefore others can simulate.” But if we can’t in fact simulate adequately, the argument fails.”

    This has been dealt with at great length on the Wikipedia discussion board for “simulated reality”. The dream argument addresses your point.

    Dreaming is a much better example of a simulation than theoretical future simulations. And since we all have a shared experience of being fooled by dreams we don’t need to contest it.

    Here is a link:


New comments have been disabled.