Am I A Sim?

The simulation argument says that IF you:

  1. expect a substantial chance Q of our civilization surviving a long time,
  2. given survival, expect a vast number V of subjective years of experience by future descendants, in total across the whole future,
  3. expect them to spend a substantial fraction F of their per-person-subjective-year resources R running simulations of their distant past, for entertainment, research, or other purposes,
  4. expect a fraction A of such sim resources to go to sim this our current year, 2010, on Earth, and
  5. expect a fraction B of these 2010 sim resources to go to sim fully conscious humans unaware they are in a sim, where each such creature costs on average C per subjective year of experience.

THEN you should expect there will be on average of N = A*B*Q*V*F*R/C sim creatures who assume they are humans living in 2010. If your N >> 1010 (current human population), then unless there is some particular reason to think your life is much less likely than average to be the sort of life that a sim lives, you should strongly expect that you are such a creature; you are a sim.  (Of course you should then question how much you know about the sort of universe you live in, which may change your N estimate. Even so that probably won’t drastically reduce your estimated chance you are a sim.)

For example, if Q = 10-2, V = 1030, F = 10-4, A = 10-9, B = 10-2, R/C = 101, then N = 1014; you are a sim.

While I published back in ’01 on how to live your life differently if you might be a sim, it took Charlie Stross pondering the topic recently to remind me that I’ve never fully engaged the argument, by trying to come up with my own best estimate of N.  So what do I think?

Let’s break it down by purpose.  First, consider entertainment.  Even compared to other humans, we today spend record large fractions of our income on tv, movies and video games; we are in the process of reacting to the development of unprecedented hyper-stimuli.  Humans in general are also clearly unusual compared to other animals, who spend almost nothing on anything sim-like.  And humans are mainly interested in simulations of other humans; we hardly have movies or games about monkey life.

So if our descendants become better adapted to their new environment, they are likely to evolve to become rather different from us, so that they spend much less of their income on sim-like stories and games, and what sims they do like should be overwhelmingly of creatures much like them, which we just aren’t. Furthermore, if such creatures have near subsistence income, and if a fully conscious sim creature costs nearly as much to support as future creatures cost, entertainment sims containing fully conscious folks should be rather rare.

Now, consider academic study of history.  Once economic growth and tech innovation slows to a near halt, I expect far less interest in learning new things, which includes learning new history.  The little learning that remains should mostly be done to signal future folk good features, and so they’ll much prefer to pay one future person to think carefully about the past, rather than spend similar resources to sim one person from the distant past.

Full scale simulations of the entire Earth over many years should be very rare, and perhaps non-existent. Similar understanding would come much cheaper from sims that only model a few people in enough detail to make them fully conscious.  Modeling a few such folks in detail and then having most other modeled folks just act in ways that are statistically similar to those few detailed folks is probably good enough for most purposes. Perhaps there will be other reasons to run sims containing fully conscious creatures, but I expect those to be even more rare.

Bottom line: I expect R/C near one, even if Q=1, and expect A*B*F to be smaller than 1010/V. So, I’m probably not a sim.

GD Star Rating
Tagged as: ,
Trackback URL:
  • You assume our future descendants will operate under evolutionary pressures (and that because of that, they’ll be rather different from us). I think it more likely, that if we have any sort of positive future at all, then we’ll have (possibly sentient, possibly not) machines doing all the work for us, while we optimize ourselves for having fun, not for being competitive with beings much more efficient than us.

    I for one certainly don’t intend to modify myself to be hyper-efficient and competitive. I’d rather buy a robot that’s smarter and more competent than me to do all the boring stuff.

    Also, the arguments here are pretty supportive of what I said here:

    • Doug

      In some futures there might be an overclass of beings solely dedicated to “hedonism” but their fraction of resources will probably be small. Most intelligence will be directed towards growth rather than hedonism. Intelligence takes resources, so diverting too much intelligence towards hedonism rather than growth will give systems a competitive disadvantage against systems with less hedonism.

      They will either be lag behind so hedonistic systems are tiny compared to growth systems (such that almost all “global” intelligence is directed towards growth) or even directly consumed by growth systems with those hedonistic resources re-diverted towards growth.

      This doesn’t mean that the long-term hedonism levels will necessarily be zero, just small. Systems may find that changing the meta-rules to remove all hedonistic components destabilize the system. You can see this with modern capitalist states. Even though it’s a pretty growth oriented system it tolerates resources diverted towards pure hedonism (e.g. wealthy heirs who hang on their yachts in the Mediterranean and do drugs all day). It appears as if we could increase growth by redirecting these resources towards another purpose.

      However property rights are strongly protected in capitalistic systems for all parties. Changing the meta-rules to weaken property rights for the unproductive very often leads to vast portions of the economy dedicated to redistribution and rent seeking rather than production. The gains from redirecting the hedonistic resources aren’t worth the possible destabilization to the system. But if a substantial portion of resources today were being wasted on the unproductive we would be more likely to change the meta rules.

      If in there are beings in the future that are dedicated to pure hedonism expect them to be pretty similar to wealthy heirs today. A small portion of the total resources of the system, a low priority, and tolerated but not encouraged by the system.

      • Sounds like you’d benefit from reading the link I provided in the comment you’re responding to.

        We might want to explicitly set up a permanent overarching power structure of one sort or another to prevent certain evolutionarily default outcomes. (I won’t go into details of the argument, since that’s done well behind the link.)

      • Fun-loving types seem as though they are not paying the proper level of attention to preparing for meeting expansionist aliens – and as a result might eventually wind up not having any fun at all.

    • I’m not conditioning on whether the future is “positive” by your lights. I do estimate a competitive future to be quite a bit more likely. Non-competitive futures most likely are not dominated by people like you doing what you think is “fun.”

    • Fructose

      You assume our future descendants will operate under evolutionary pressures

      He also assumes that the laws of physics won’t magically change, and that 2+2 will continue to equal 4. Evolutionary pressures don’t just disappear, they apply to intelligent agents as much as they do to bacteria.

      • Evolutionary pressures don’t just disappear, they apply to intelligent agents as much as they do to bacteria.

        Only in the absence of a permanent overarching power structure, that doesn’t allow any changes that are deemed unwanted.

        Currently there isn’t such a power structure, but we’ll be very silly if in the future we don’t set up one. For more, see:

  • Will Newsome

    I’m sorry if you’ve answered this elsewhere, but what’s your take on the mathematical universe hypothesis?

  • Pingback: Tweets that mention Overcoming Bias : Am I A Sim? --

  • Doug

    A good argument for why being a sim or not doesn’t matter.

    Let’s assume first that sim creatures cannot tell the difference from real creatures (i.e. no soul, etc.). So let’s say your life is being run in a sim by a guy using a Turing tape machine.

    Logically I would say that Turing machines would not tell the difference if
    1) For redundancy the program was being run simultaneously on multiple tapes (i.e. you wouldn’t have double the experience)
    2) The tape was running slow or fast, time to you would pass at the same rate
    3) The tape arbitrarily stopped for periods of time, like a second, a century, or a billion years, it wouldn’t seem like your life stopped running.

    All of these extrapolated to the limit imply that even if [i]no tape[/i] is running that the sim’s experience is still “real.”
    1) Let’s say the Turing tape master is running 1000 simultaneously tapes, if he removes 1 from the redudance that doesn’t matter. If he goes from 100 to 99, 3 to 2 or 2 to 1 it also doesn’t matter. So why should 1 to 0 be different? There’s nothing any more special about Tape 1 then there is Tape 1000, and if the end of the latter didn’t “annihilate” the sim why should the former?

    2) Let’s say that the tape keeps running slower and slower over time (bad machine maintenance). It runs at a rate of 1 for the first unit of tape time, then 1/2 for the next unit of tape time, then 1/4, etc. The tape will never go past 2 units of sim time even if run for eternity, yet before we said the rate at which the tape is processed doesn’t matter. Would the be “annihilated” at 2 unit sim time the same as if the sim stopped running?

    3) Pretend that every time the tape jams up, it stops for a random amount of tape time. If the stop time distribution follows the exponential distribution then any sim time will complete in a given tape time with defined probability. If instead the jam time follows a exponential-Cauchy distribution the same cannot be said. There is some probability that even in infinite tape time a given sim time will not be reached. Yet I don’t see why the sims on the exponential-Normal tape would feel less “real” than the sims on the exponential-Cauchy tape. And if they did make it to the end of the Cauchy sim in finite time would they be “more real” than those that didn’t? Would this go back in (tape) time and effect whether the sims were real at the beginning of the sim?

    I think these arguments are evidence in support of all mathematically (or at least computable) objects as being “real.” Whatever life you live is real simply because it is mathematically possible. It may be “instantiated” physically a single time, many times (in multiple sims or different parts of the random universe), infinite times or maybe even no times, but experience and existence I would argue are unitary regardless.

  • Carl Shulman


    1. You can be simulated by creatures that aren’t descendants of entities in our position, i.e. sims don’t need to be ancestor sims. You need to argue for general convergence across civilizations, including human civilizations that already have a world government by the time they get to uploads (probably more than 1 in 1000 human civs are like that, due to lucky geography or the development of nukes turning out differently, etc).

    2. You claim to buy the Self-Indication Assumption, so if you think that there is a 1 in 1 million chance of billions of sims of you in your current mental state per unsimulated version being made, with only unsimulated versions otherwise, you should be quite confident you’re a sim.

  • Carl Shulman

    The “1 in 1 million chance” is obviously the estimate before applying SIA.

  • Doug, I already said why it matters if you are a sim here.

    Carl, yes in principle I might be a sim of aliens who had no idea what Earth creatures are like. Seems unlikely. Yes a 10-6 chance of 1010 sims of me makes me likely a sim; the question is: what are the actual numbers.

    • Carl Shulman

      Robin, then the apparent approach in the post, filling in point estimates (e.g. for R/C) from your favored scenarios, is wrong.

    • Khoth

      Why unlikely? I’d expect the vast majority of sims to be of things that don’t actually exist in reality.

    • *Why* does it seem unlikely? Humans simulate creatures far-inferior to themselves all the time – for example in genetic algorithms. This is the “optimisationverse” scenario.

  • I’m generally very skeptical of any claim about residing in a simulation of the laws of physics of the outside world are at all similar to our own. Simulating a large object is most easily done by simply having an actual copy of that. Unless the sim is coarse, the simplest simulation of any given object is just that object. Now, if one had lots of resources (say Jupiter brains) then one might be able to reasonably simulate maybe a small planet like Earth, be coarse about the internal details, and use a rough simulation for the stars and planets. But even that seems unlikely, since we can actually observe planets around other stars which suggest that if we are part of a simulation a lot of detail has gone into it.

    • Check with the physics simulations in our games and movies.

    • Simulating a large object is most easily done by simply having an actual copy of that.

      Let me propose a theorem: simulations are NEVER perfect. I have spent my life simulating the living crap out of superconducting devices, radio systems performance, and position determination systems. At MANY steps in each sim we simplify, replacing a complex but (we hope) irrelevant detail with an on-off sim, the effect of 100s of other radio signals with spectrally shaped random noise. In MANY other places in the sim we simplify without realizing it, we forget that bloch waves are approximate for what the electron does, neglecting defects in crystals, non-point-source charges, interactions with other electrons, and secondary interactions where electrons influence the crystal which then defines the electronic wave functions. This last effect GIVES you superconductivity, a perfectly adequate simulation of a metallic conductor will not predict superconductivity.

      Strong AI idea if it is true can be extended to suggests that there is no difference in our world, in our experience, whether we are “real” or a PERFECT sim running on some other computational substrate.

      I think strong AI is wrong. I think a simulation of a nuclear attack is way different from a nuclear attack, not the least because no people are killed, no cities are destroyed, but also because in the structures left standing (in the sim) you don’t find the shadow of a human on a wall, clever pieces of jewelry and children’s toys mysteriously preserved in the rubble, and so on. I KNOW having written sims that they are gigantically less complex than the actualities.

      Am I guilty of a crime when I kill a hooker with my bare fists in Grand Theft Auto (video game)? I don’t think so. Sure, some subroutine is using rules plus random number generators to separately create this creature in the Nintendo processor. I don’t think that makes her alive, or even sentient, or even human. Among other things, I’m pretty sure the coders didn’t put pain and fear and hopes and dreams in each creature in the sim, not just because it would have slowed down the game, but also because they don’t have a clue how to do that.

      The probability that the humans in a sim will be sentient I think got left out of Robin’s original equation. Even if it is possible to make a creature in a sim sentient, we know that many sims are made without sentient humans now. Like probably ALL of them. Lets say I see 10 simulated non-sentient humans a minute while playing. Suppose I have played 2000 hours in my life. Thats over a million non-sentient simmed humans per my one me, a sentient human. So probabilities above get multiplied, I think, by at most 1/1000000.

  • arch1


    I need a little help w/ your assumption 4, whose time- (but not space-) granularity seems somewhat arbitrary and also material to the final answer..

    If in assumption #4 I change the year 2010 to the month August 2010, (or alternatively, to the decade centered on 2010), my estimated probability of being a sim just went down (up) by a factor of 10, without on the face of it affecting the validity of the argument.

    (My first thought on this was that your choice of time unit in assumption 4 is not arbitrary, because it must be identical to those in assumptions 2, 3, and 5. But on second thought, it appears that assumption 4’s time unit is independent of that used in the other assumptions.)

    • If you change the time units of both R and C, the changes cancel.

  • Are you going to update the probability for that scenario here?

  • Another variable is “how ‘happy’ am I?” I guess this is similar to the “why does God allow suffering?” argument.

    I presume that the creators of simulations would generally try to make their subjects as happy as possible. Now, even if happiness is largely relative, this shouldn’t pose too much of a problem: create simulations where the lowest status people are automatons and the highest status person or people are conscious. Thus, the fact that there is considerable suffering in Africa does not necessarily reduce the chance of this world being a simulation, if in fact those people were automatons created solely to make myself feel better by comparison. However, if [i]I/i] felt I was suffering, I would be much less inclined to “believe”.

    • Steven Ehrbar

      I presume that the creators of simulations would generally try to make their subjects as happy as possible.

      Why? There are plenty of people who, when playing games like The Sims, deliberately torture and kill their Sims. And when modern scientists run things like Tierra, there’s no effort to make things nice for the simulated life.

  • MichaelG

    1. If we had the ability to simulate worlds, we might simulate the origins of life, and do it many times with different initial conditions.

    By analogy, if future AIs are interested in the origins of the first true AI, they might simulate our world many times with different conditions to see what emerges. It would not simply be for entertainment purposes.

    2. This would also explain why our particular time is being simulated — we are close to the emergence of true AI. You think you have a past, but only the last few years before the Singularity are ever simulated. Once AI emerges, the simulation ends and the people are discarded.

    3. Also, I don’t think your assumption of a super-competitive future stands up. Look at nature outside of humans now. You would think from general principles that all animals would be in intense competition and spend all their resources (and all their time) trying to increase their advantages.

    Yet, lions spend most of their day lying around. It’s not a harsh competitive world except for the minority of their time they spend hunting. If they were intelligent enough to do it, they could spend most of their days doing art and science and entertaining themselves.

    • Doug

      Biological creatures today are in extremely intense competition. I think you’re judging animals by the standards of civilization where nutrition is abundant relative to free time. However for the majority of creatures feeding opportunities occur infrequently so it is an optimization to spend most of their time “doing nothing.”

      You’re looking at lions, and thinking “doesn’t look hard to me, sleep 20 hours a day then eat, a lot easier than my 10+ hours of work/commuting/errands.” In reality the mortality rate among lions (especially cubs) is astronomical. They spend all that time resting because viable prey only wanders into their territory very infrequently and their thin, sleek, muscular bodies burn a tremendous amount of calories moving around. They can only hold tiny fat reserves otherwise they become too slow to catch prey, as soon as they use up their fat they start losing muscle and a vicious cycle to death occurs. They’re not relaxing, they’re conserving their very meagre energy reserves in the hope of not dying.

      Also it’s assuming a very big change to say if lions were intelligent enough to do art/science. The reason they aren’t is blind luck, it’s because the caloric demands of a big brain keep the vast majority of animals as stupid as they possibly can be to survive.

      Virtually all creatures spend all their active time feeding, hunting, recovering from injury, or mate seeking. If their opportunities for these are limited they are probably in a state of near total shutdown. This down time is more analogous to the time your computer spends in suspend mode, rather than the self-actualization free-time that wealthy humans enjoy.

      • MichaelG

        I see what you are saying with respect to lions, but doesn’t that line of argument mean primates with social life and playtime could never have evolved? After all, monkeys have lots of competition. How can they also have the luxury of large brains?

      • MichaelG: mammals are rare compared to the average insect or something; primates are pretty rare among the mammals; *social* primates aren’t even a majority, I suspect (consider the entire prosimian grouping).

        As it is, they have a small niche in a few areas and tend to be going extinct or be under pressure. If they were as common and successful as your average insect species, then that’d be a serious problem.

        How do they survive? They probably fit in a niche where a group is powerful enough to kill any hyena or lion that attacks it but small enough to survive mostly off fruits and vegetables in denser forest stands; and the group requires sociality. (I’m not a biologist, I’m just guessing here and going off memories of how chimps can easily kill men and baboons hyenas.)

  • Scholastic

    Here’s an argument against the Simulation Argument:

    1. Either we live in a simulation or we do not.
    2. If we live in a simulation, there is no reason for thinking that perception is veridical.
    3. If we don’t live in a simulation, we don’t live in a simulation.
    4. So, either there is no reason for thinking that perception is veridical, or we don’t live in a simulation.
    5. Now, if there is no reason for thinking that perception is veridical, the Simulation Argument fails.
    6, So, either the Simulation Argument fails, or we don’t live in a simulation.
    7. If we don’t live in a simulation, the Simulation Argument fails.
    8. So, the Simulation Argument fails.

    • Cyrus

      If I grant (5) at the level of generality your argument requires, then it applies against itself at least as strongly as it does against the simulation argument.

      • Scholastic

        Why? Sims are (at least) one extra step removed epistemically from the real world than are inhabitants of reality. Unless one takes radical skepticism seriously, it’s sensible to hold that a perceiver in the real world has reasons for regarding perception as veridical. Similarly, it may be reasonably claimed that a sim has reliable perceptual access to the simulated reality. Yet as far as I can see, there are no reasons for thinking that sim perception yields knowledge not only of the simulation, but also of reality itself.

        I should clarify, incidentally, that SA fails if perception is not veridical because the argument relies on knowledge about technological trends, computational limits and the nature of consciousness which can only be acquired perceptually.

  • mjgeddes

    I’m not convinced that there is anything to reality other than pure knowledge representation. That is, I strongly suspect reality is all map, and no territory. (I view the idea of a ‘territory’ as similar to the idea of an ‘aether’ pre-relatively, the notion is redundant). This can make sense provide reality is separated into multi-level maps aka the movie ‘Inception’.

    The simulation we are in is a self-simulation generated by the universe itself. The space-time block universe is the deepest ‘layer’ of the simulation. This idea is consistent with my very strong suspicion that Bayesian Induction is merely a special case of categorization (knowledge represenation, analogy, clustering).

  • Brian Basham

    Just had to chime in about the math. Your N is the expected value of the number of sims of 2010-people. However, calculating 10^10/N isn’t your probability of being a non-sim. Your probability of being a sim is clearly less than Q, the probability that your civilization lasts “a long time”. If you assume Q is 10^-2, then it doesn’t matter how many sims could exist 1% of the time, what matters is that we have estimated that no sims exist at all with a probability of 99%.

    Similarly, if I shoot one person in the world at random with 10^20 bullets, the expected number of bullets per person is much greater than one. However, your probability of dying is still very small.

    • I’m talking about your prob of being a sim conditional on seeing you exist. That is different that the global prob of their being many sims.

    • arch1


      Thanks, your objection got me thinking (and I can imagine that much discussion has occurred on this point over the years). But I think Robin’s model is correct in this regard. Here is a separate scenario which helped me convince myself of this:

      On arrival in your first visit to the very populous country Slobbovia, you learn of the following Slobbovian custom:
      1) the capricious and fabulously wealthy ruling dynasty, on its annual Judgement Day, randomly selects 10 people to work in the Slobbovian salt mines for the rest of their lives, “just for chuckles.”
      2) Additionally, with a probability of 1% (i.e. on average once per century, randomly), the dynasty randomly selects 10,000 other people to receive $1 million each (tax-free, lump sum, of course).
      3) Each prospective salt miner, and each prospective millionaire, is notified on Judgement Day that he or she has been Chosen – and is then held in isolation for the next 24 hours without knowing his or her fate.
      4) At the end of the isolation period, each Chosen one is carted off to the salt mines, or presented with a $1 million check, as appropriate.

      You also learn that, as fate would have it, today is Judgement Day – and you have been Chosen. As is the custom, you are immediately clapped into isolation to ponder your fate.

      Your first reaction is that you have roughly a 99% chance of spending the rest of your life in a Slobbovian salt mine, since 99% of Judgement Days are millionaire-free. On further consideration, however, you come to a significantly different conclusion, reasoning as follows:

      1) On average, over the course of a century, 10*100 = 1000 salt miners, and 1*10,000 = 10,000 millionaires, are Chosen; so on average 10,000/11,000, or 91%, of the Chosen are future millionaires.
      2) There is no reason to believe you are more or less likely than a typical Chosen one to be destined for the mines
      3) Therefore, you have a 91% chance of becoming a millionaire (and only a 9% chance of wielding a pickaxe for the remainder of your days).

      I find the second argument convincing, and think that its analogue is implicit in Robin’s Sim model above.

  • jb

    I suspect you don’t play many games, Robin, so you don’t appreciate just how robust the current state-of-the-art is for simulations. I’m going to rattle off a bunch of different ideas about what I would do with infinite computing power that would represent a form of “sim”.

    1) The game Spore simulates the evolution of a microbe into a galactic empire. More realistic physics and smarter enemies would be awesome. Simulating an entire galaxy will eventually (~150 years) be dirt-cheap, kids will be playing it on their equivalent XBox.

    2) Economists could model the past Earth (~100 years from now) and then change various events in order to see how economic policies worked or did not work. Sufficient simulations leads to statistically significant results re: Keynes vs Hayek vs Futarchy, etc

    3) Physicists could be modeling universes with different constants, to see how they develop. (~200 years).

    4) Historians could model the past Earth (~120 years from now) altering various specific events and seeing how that change cascaded forwards in time.

    5) Similar to Spore, the game “The Sims” in the future could have a much more fully realized past words. (~50 years) – not far enough out that we’re like “monkeys”. See such games as: Medieval: Total War, Rome: Total War, Civilization, Red Dead Redemption, Pirates!, etc

    6) Future children/families could go back on a “simulated” experience of what it’s like to live in the past – i.e. their minds are injected into avatars so they can “rough it” in the disease- and poverty-ridden 21st century. ~200 years – also Renaissance festivals. Assume time dilation for the visitors

    7) Disobedient future children could be injected into a 21st century simulation to help teach them to be more appreciative of the wonders of their actual universe. Assume time dilation.

    8) As part of a standard education, future children could be injected into a 21st century simulation to teach them what the past was like. Again, assume time dilation.

    You said “why not use mathematical models instead of expensive, fully-realized sims?” – but have you watched the way video game companies compete – they compete on fun, yes, but also on graphics quality, advanced AI, bigger worlds, more realistic worlds, destructible worlds, etc, etc. Competition between them would _easily_ explain the race towards a fully simulated town/state/country/planet/solar system/galaxy/universe – just so they can make their game more attractive than their rival’s.

    If you’ve never played the game Spore, you start by evolving a microbal life form into a fully-formed human-equivalent alien physiology. Then you build space ships and explore the entire galaxy. And this galaxy’s habitable planets are populated with the alien physiologies created by the other players of Spore.

    We may be hopelessly primitive monkeys, but to some alien future species we might just be adorable and cuddly. Or ferocious badasses. Or intellectual giants. It doesn’t really matter.

    In 20 years, the video games will be far more realistic, far bigger, far more physics-accurate than today. The AI will be more advanced as well, the non-player-characters will be more fully realized, the enemy strategists will be smarter, and so forth. And this trend will only continue into the future, because eventually the graphics can’t get any better, but the AI and the physics and the game-scope can get better/bigger for decades or centuries to come.

  • William H. Stoddard

    Isn’t the whole “I am a sim” argument basically a new version of Descartes’ evil demon?

    • Indeed! I actually pulled up the wiki page on Descartes’ meditations and was looking through them as I read some of these entries and thought the same thing.

      We also have mjgeddes above denying an external reality. IIRC, that was Bishop Berkeley’s position (esse est percipi). In the same course I learned Descartes and Berkeley, we ended with Kant, who seemed to solomonaically split the baby: there is an external reality (noumena) but we can’t perceive it, what we perceive is something else (phenomena). I remember thinking if I could understand Kant that I’d have it knocked, but that seminar was the end of that, I went on to study physics.

      As part of my philosophical education I studied Popper who I interpret as saying (among many other things) “a difference that makes no difference is no difference.” That is, if you think you have a theory that is different from another theory, you must be able to design an experiment that has a different outcome between the two theories. If you can’t, you are just rearranging the deck chairs on the Titanic.

      So what are the properties of a Sim such that we could not tell we are in a sim? Such a sim, if it existed, IS reality.

      Thought experiment: why create expensive new fast computers? Lets just write a simulation of one, and then we can get results really fast by running them on the simulated computer.


  • A probable reason for our descendants running simulations of eras relatively near to the origin of life would be to understand the forms that *other* civilisations might take – in order to help deal with any aliens we might meet. That may not become boring for a very long time – since the survival of our entire civilisation may depend on getting at those answers – to best avoid assimilation.

  • David C

    “Similar understanding would come much cheaper from sims that only model a few people in enough detail to make them fully conscious. Modeling a few such folks in detail and then having most other modeled folks just act in ways that are statistically similar to those few detailed folks is probably good enough for most purposes.” – Robin Hanson

    You don’t actually know that that hasn’t been done. The statistically similar individuals would still need to be different enough that the modeled individuals did not realize they were in a simulation. For all you know, the majority of people on this planet aren’t nearly as detailed as you think they are.

    I also agree with Carl Shulman’s argument.

  • matt

    What if we’re just a simulation than got left on by accident? Like someone in some universe left their version of Conway’s Game of Life on and forgot to turn it off.

  • anomdebus

    Is this just a roundabout way of saying the Drake equation is absurd?

  • Philo

    For historical and entertainment purposes *virtual* similation will be cheaper and easier than physically realized similation. And the “hypothesis” *that I am a “virtual being”* is incoherent.

  • Vlad

    The argument only addresses people. But what about the entire observable universe? And even if we consider astronomy to be merely an illusion crafted for our mental comfort (but, really, billions of galaxies?!) and suppose that only Earth is truly simulated, why bother to simulate humans (and all the animals etc.) in all their biophysical detail, all the way down to elementary particles?!

    I’m thinking that if I were to set up a giant simulation of humanity I would basically build an agent based model, each agent designed with a set of algorithms driving its behavior (and I guess consciousness, and mind reading in general, needs to be among these algorithms such that agents react properly to one another). Perhaps it’s just me, but, if I were interested in simulating humanity, I wouldn’t build the sims from atoms and wait for them to emerge (who knows when, whether and in what form). So, is all our knowledge about the structure of matter etc. is also supposed to be illusion?

    I think that the more stuff you need to assume to be illusions rather than actually simulated, the less convincing the simulation hypothesis becomes. And if you assume that everything science discovered is part of the simulation than this does not look anymore like a simulation focused on us. So, basically, the main skeptical arguments about the existence of God (i.e. the fact that the universe doesn’t seem to be designed with us in mind) are also replicated about the simulation hypothesis.

    Anyway, it seems to me that the reason why the simulation hypothesis is interesting and attracts attention is that it implies that some stuff is illusory. But what exactly is supposed to be illusory? Astronomy? What else?

    • Robert Koslover

      Vlad, I think you are right. It seems to me that to assert that “we are in a sim” fails the test of Occam’s razor: “entities must not be multiplied beyond necessity.” Obviously, Occam’s razor could be wrong in this case, but I confess to liking it and I intend to continue using it.


  • Patrick L

    We can probably guess a simulation of the universe would be imperfect at some level. We shouldn’t be able to see past a certain level of detail, there would be arbitrary boundaries, and certain things wouldn’t make sense or would break. I’d expect to be seeing lots of shortcuts in the universe.

    Is there anything in our universe that seems like it meets these conditions?

    • mjgeddes

      Yes, there are two big give-aways. I can clearly see that physics appears be separated into three levels, showing that its a ‘cut and paste’ job. You can see the boundaries between the different levels of reality if you examine the physics clearly:

      Boundary 1:

      Relativity- Quantum physics (Space-Time/Forces)

      Quantum mechanics is not consistent with general relativity. The base level of the simulation is the space-time block universe and there is a discontinuity between the space-time background and the quantum mechnical forces.

      Boundary 2:

      Quantum physics – Classical physics (Wave functions/Objects)

      The second boundary shows an inconsistency between the QM picture and the classical picture of concrete objects. There is no explanation of how to convert wave functions into the concrete objects we actually observe (Born probabilities are an arbitrary rule).

      The Inception planted in our minds is that all this is real. We need ‘kicks’ to wake up. Fail to awaken and you will remain in limbo 😉

      • Robert Koslover

        Your reasoning appears to, more-or-less, parallel the “giveaways” arguments that have long been claimed as “proofs” of the existence of God (see Ref. 1). To a purist physicist, the observable fact the world is mysterious, complex, and difficult to understand proves only that it is mysterious, complex, and difficult to understand. It doesn’t “prove” that we were created by God and/or are merely a sim created by some more advanced technological beings. On a related note, I wonder if Robin has a similar formula for predicting the probability that God exists? Hmmm. And if I am a sim, is my “creator,” for all intents and purposes to me, my God? If so, perhaps we should establish a “First Church of the Sim God(s)” or something like that, to more formally pray to him/her/it/them for mercy.

        And many, many more…

      • mjgeddes

        First, I don’t claim there is any intelligent design at work (see my earlier post, the simulation we are in could have many causes, and even be a by-product of natural proccess).

        Second, I am pointing to very clear and specific discontinuities based on modern knowledge – (1) the very clear discontinuity between the space-time continuum of relativity, and quantum mechnical forces, the field cannot be quantizied. The discontinuity clearly separates out two different levels – the sharp and specific boundary suggests it’s a feature of reality, not in our minds. (2) the very clear discontinuity between the QM wave function and the world of classical concrete objects , again, a very sharp separation between the classical and quantum worlds, an abitrary unexplained rule (for generating Born probabilities) and clear inconsistencies (need for non-locality), suggests the confusion is with reality itself, not in our minds.

      • I don’t believe there are any known inconsistencies between General Relativity, Quantum Mechanics, and Classical mechanics. That is to say, I don’t believe there are some situations where QM gives the right answer and GR gives the wrong one, or vice versa.

        If there are that you know of, do tell!

    • Maybe “Game” or “PUA” only works on the simpler women, the ones that are not fully simulated. To tie another common theme from Overcoming Bias into this post.

  • MattJ

    The possibility that we all may be living in a simulation explains a lot, really.

    Tom Cruise is clearly one of the people from the future playing in our simulation. That would explain why he acts so insane, how he keeps getting work, and how he’s managed to snag a succession of beautiful women.

    He knows the cheat codes.

  • Sigivald

    I expect B=0.

    (And I agree it doesn’t matter in any case.)

  • Chris

    If we are all Sims, that’s bad news. Think about the games we play – mostly about the unique, and often, difficult times. There are way more WWII simulator games than peaceful, boring simulators. So, unless we’re lucky enough to be in 2357’s version of Farmville, bad things are coming…

  • “Full scale simulations of the entire Earth over many years should be very rare, and perhaps non-existent.”

    I’ll disagree with this point. Who says the planet is populated by more than one observer? Let’s assume it’s just me and nobody else. I’m the avatar, you’re all NPCs (so to speak). I’m fairly educated (MS) and yet I have never personally tested quantum physics, measured the universe, tested the speed of light, etc. The simulation, for me, merely has to be the little slice of time/space I’m in right now and of sufficient detail that I can’t detect a fake (assuming one can even do so, given all I know is what I’ve observed). That level of detail (basic physics, senses) seems within grasp of simulation technology that will appear in my lifetime.

  • Wg

    Believing you are in a sim is not distinguishable from believing in an omnipotent god (of any type).

    • Checking with the profiles of the adherents suggests that there may indeed be some differences.

    • Believing you are in a sim is not distinguishable from believing in an omnipotent god (of any type).

      I love it! I’m going to use this as a quote at then end of my emails, at least for a while. (Probably not all the way until the sim ends. Maybe just to the next save.)

  • William H. Stoddard

    I basically sympathize with Vlad’s argument. But there’s an amusing response to his comment that “why bother to simulate humans (and all the animals etc.) in all their biophysical detail, all the way down to elementary particles?”: What if the biophysical detail, all the way down to elementary particles, is a vastly simplified model of the physics of the real world inhabited by real human beings?

    • John

      The speed-of-light limit could be taken as evidence in favor. A world with no preferred frame of reference is a world which can be parallelized.

      • xxd

        Yes. The question of “Are we in a sim?” depends on whether we’re in a finite computational space type sim or whether we’re in a non-finite computational space (i.e. running on a quantum computer).

        In a finite sim where we want it to be as realistic as possible, having a high speed of light would mean we’d have to simulate other civilizations too, thus reducing the achievable complexity of the sim on average.

        In a finite sim with a low speed of light we only need to simulate locality with any kind of high degree of fidelity.

        In an infinite sim, there could conceivably be no detectable difference between base reality and the sim.

        Speculating on whether we’re in a finite sim, I’d expect to see lots of attempts to compress unneccesarily detailed information and I’d also expect to see level of detail reduced when we’re not looking at it. I’d also expect to see non locality happening faster than the speed of light but not necessarily explained by the physics of the sim.

  • Roy

    “Once economic growth and tech innovation slows to a near halt, I expect far less interest in learning new things, which includes learning new history”

    Why exactly will economic groth or technical innovation ever slow to a near halt?

    When exactly has this happened?

    • Konvkistador

      You’re not that familiar with Hanson’s writing are you?

      At the risk of doing him great injustice I will venture to try and sum some of his ideas on this up (anyone please correct me if I’m mistaken):

      – the speed of light limit ensures Malthus has the last laugh.
      We can’t expand faster than the speed of light to absorb more matter. We can only get so much value or simulate so much virtual space with the matter we have.

      – we are exhausting the space of useful technology that is vastly more useful than previously available technology (compared to current and near future expected rates of advance). This note is not the same as exhausting all technology or even all useful technology and he explicitly (I think) argues against *that* possibility. Economic growth due to innovation will therefore essentially end.

  • Alexei Turchin

    Robin, where are several reasons why you are a sim:

    1) You live interesting life and you a special person, but 70 per cent of all humans ever lived were agricaltural workers. Our movies more often depict heroes then ordinary people.
    2) If you live in simple cheap simulation, you can not dictinguish it. (It is easy to prove that you are most likely in the cheapest of simulations, because they are majority)
    3) Future Super AI need simulations not only for fan but it is simulating different courses of development of diffrent civilizations in order to calculate probability distribution of different AI types in the universe.
    4) Super AI will be especialy interested in the period right before its creation, so 2010 is most likely year to be simulated. Also it is the funniest year.
    5) Super AI will be especially interested in simulating persons, who influenced the development of AI. So you are.

    And at least some simulations have after-life, where dead sims gets their own paradise and immortality. I think that the share of such simulations is not small, because sim owners could be morally obligated to compensate suffering of sims. ( Becuse they are affraid that they are also in simulation)

    Your agent Smith 🙂

    • Robert Koslover

      Finally! A logical argument that supports believing in life after death, without requiring any religious faith! Now, if only I could develop a strong faith in your argument, but… well, once a skeptic, always a skeptic, I suppose. Of course, that might simply be the nature of my pre-assigned sim personality.

      • Alexei Turchin

        Yes rationality denies itself because we must conclude that we most likely live in the simulation with miracles and after-life. 🙂

  • The simulation argument gets the level-crossings wrong. There may be a stack of simulated realities with ur-reality at the top. Is our medium grain reality actually at the top? The simulation argument proposes that mankind will run many simulations in the future and tries counting the beings in coarse grained realities one level below ours. They may wish to guess whether they are in a simulation or not. They need to know whether their coarse grained reality could exist independently as an ur-reality, a question which we are poorly placed to answer.

    For our part we need to know about the beings in the level above ours. They live in a fine grained reality and are simulating us in our medium grained reality, perhaps giving rise to very many simulated beings at our medium grain level.

    To guess the probability that we are in a simulation we need to know about the multiverse: what is the ratio of ur-realities at medium grain to ur-realities at fine grain. This is a key number, if ur-realities at fine grain are more common the simulation argument is stronger in proportion. But we have nothing to go on. Finer grained ur-realities might not exist at all and we are definitely not in a simulation. Medium grained ur-realities might not exist at all and we definitely are in a simulation. Lacking this key number we know nothing and the simulation argument is idle speculation

  • Alan, you said:
    “To guess the probability that we are in a simulation we need to know about the multiverse: what is the ratio of ur-realities at medium grain to ur-realities at fine grain. ”

    This is exactly the core of the original simulation argument. The key is it doesn’t matter the exact number of different types of universes (or multiverses) in the ensemble, only the ratios matter.

    The entire point of the simulation argument is to show:
    1. that every universe somewhat like ours eventually spawns N child sub-universes (of somewhat lower fidelity)
    2. that N is very large

    thats all you need to show. The simulation argument is only false if N is much less than 1.

    We will know for sure that we live in a simulation when we start creating our simulated universes complete with sentient entities who don’t know they are in a simulation. It looks highly likely that we will be doing that this century (if we don’t die out first).

    • “We will know for sure that we live in a simulation when we start creating our simulated universes complete with sentient entities who don’t know they are in a simulation.” We’ve already done that, if you adopt a reasonably low definition of sentient.

      To play devil’s advocate, though, I’d expect more simulations within simulations N deep, if we were in a simulation that was more than 1 level removed from actual reality ourselves.

  • Laurent

    Each modeler of each virtual universe with human simulated brains would do it almost perfectly taking into account the technology according to his psicological desires.
    So if for a moment we forget about the pure simulation model, and we think on the benefits or intentions the results of these virtual worlds would be different to what we have;
    this takes away the probability that we are in a virtual world with some clear almost perfect motive and therefore the chances that we are real if we discard the intentions of the modelers are almost 99% real.

  • tower

    What a joke. A probability for something that is not repeatable. Where is my high school stats book?

    Suppose your argument made sense. Anyone who believes something because it is likely or even nigh certain is quite clearly a fool. Surely you see this?

    • Cyan

      What a joke. A probability for something that is not repeatable.

      Here’s a link to help you get with the program.

  • Pingback: The Death Delusion « Kensho()

  • Pingback: Survival Of The Fittest? | HighStepping()

  • Ender

    I agree with whoever said that the simulations wouldn’t have to be simulations of historical events (you could argue that simulating a random possible universe would be unlikely to result in this particular universe, but I would shoot that argument down (since you can say that in any universe, provided that the universe exists))

    Also, what I took away from the original paper about simulations was not a belief either that we are or aren’t simulated. The point, as I see it, is that IF anyone simulates a universe (especially more than one) in the future, it means that we are probably in a simulation.

    Also, I have homework to do. The internet is ruining me.

  • Pingback: Overcoming Bias : Silence Suggests Sim?()

  • Pingback: The Quantum Super-PAC « Gödel’s Lost Letter and P=NP()