Oct 25th, 2019
Not a member of Pastebin yet? Sign Up, it unlocks many cool features!
  1. Is it likely that I live in a simulation?
  3. This is a question more and more people are asking themselves. As videogames become more realistic, it's not unreasonable to think that we could ourselves be characters in a very advanced computer game. If we could determine that our reality is in fact simulated, what consequences would that have?
  5. The idea has been explored in science fiction but hardly taken seriously before Nick Bostrom wrote about it[1].
  7. 1. Premises
  9. For it to be likely that I live in a simulation, following premises must be true:
  11. A. It is possible to create a consciousness (that exists in a virtual environment) in a computer.
  12. B. Someone advanced enough to do it, wants to do simulations containing conscious agents.
  13. C. There are many more simulated worlds (that I can exist in) than material worlds.
  15. 2. Definitions
  17. First of all we must define what we mean with a simulation. What most people think of, is the kind that is shown in the Matrix movies. In them, the main character still have his biological body but his brain is ”hooked up” to a computer simulated world. That is an advanced virtual reality, but not what this paper will discuss further, because in that scenario one material world corresponds to one virtual world (per person). So that doesn't fulfill premise C above.
  19. Another popular trope is mind uploading, where ones consciousness and memories are copied and one would live forever inside a computer or on the web. This doesn't apply here since I would remember if I had been uploaded.
  21. This paper define a simulated world as a world where both a conscious being, I, and the environment I experience, are rendered by a program that runs on a machine, created and programmed by another (more technologically advanced) conscious being. (These programmers can themselves be simulated beings)
  23. If someone has programmed my reality, they can do it in two main ways. The easy way is to just program me and render only the things I percieve with my senses. All other humans I interact with are like NPC:s (non-player characters) in a computer game – avatars that are programmed to act like they are conscious. This is a variant of the old idea of solipsism. My memories and all historical evidence can also be simulated so the simulation can have existed for only a short time.
  25. The other way is to simulate a huge universe from bottom up, let the world evolve and everything in it until conscious beings appear. This of course will require a mind-boggling amount of computation.
  27. A middle way between those two, is to simulate a limited number of conscious beings (for example the population of Earth) and their surroundings (and only when they are aware of them).
  29. Bostroms paper is mainly concerning ”ancestor simulations”, meaning recreations of an earlier period of ones history. That's because those who simulate us must obviously be more advanced than we are right now, but still somehow interested in us. That could be because our probable future is their present, and at least some of them are interested in history. That would also require less creativity, since they could use existing history, instead of having to compose the music, the first time I click on a Yuotube video of a Mozart symphony I haven't heard before.
  31. 3. Discussion of premise A
  33. a) Is it at all possible to make a computer program self aware?
  35. Unless consciousness require a supernatural soul, given by God, we must assume that it is ”sort of” programmed in biochemistry and electrical impulses on the hardware of our brains. Therefore, we theoretically should be able to make an emulation of a working brain in a computer, even if we have to render the quantum properties of every atom in it.
  37. The belief that a soul is needed for consciousness, can't be proven or refuted, so we must leave it at that.
  39. b) Is it physically possible to make as big a computer as needed?
  41. Even if the development of computers so far has been exponential, there are still physical limitations on how much we can improve their performance. It has been proposed that we soon will have problems with quantum effects, if we make the circuits smaller.
  43. You could say that we don't know anything about the universe that simulates us (size, age or even laws of physics), so we can't make any objections to the possibility of building the required computers. This is akin to saying we are a thought in God's mind. It doesn't give us any foundation for a discussion. To be able to make any guesses of the likelihood of us being in a simulation, we must assume that our programmers live in a somewhat similar world to ours, only more technologically advanced. For the same reason we can't assume any future tehcnology that would seem like magic to us.
  45. It seems obvious that we can't simulate the whole of a universe within a similar universe. You need more than one atom of a computer to simulate a virtual atom (even if you have perfect quantum computers), so the mass of the universe is not enough. But this is true only if we simulate a universe in real time. We could calculate all of the movements of a number of particles, in a computer smaller than all those particles put together, if we take a long time to do it. It could take a year to calculate a nanosecond of the virtual universe. And of course, inside the simulation, you can't know how fast the time goes outside of it.
  47. With that in mind, it still seems intuitive, that the whole universe can't be rendered down to the planck level by a civilisation, limited by the mass and laws in a universe similar to ours. Ringel and Kovrizhin[2] show that it's impossible to simulate the quantum state and entanglement of just a couple of hundreds electrons (unless you assume hypothetical technology or hypothetical laws of physics). That means that any simulation must be more or less ”fake” and only render the observed parts.
  49. Solipsistic simulations could be doable with future computers.
  51. The more of a bottom-up universe you do, the less initial programming is needed (just the fundamental laws of physics and the physical constants) but more hardware. The more solipsistic, the more creativity and interference is needed from the creator, but less computer capacity.
  53. To simulate all people on Earth plus everyones subjective reality (to a necessary resolution and with a limited history) and to keep the continuity of everything, you need a very big computer, and more importantly, you have to do an enormous amount of programming before you can even start the simulation (unless you start from when life began and let it evolve by itself).
  55. I'm not convinced by handwavy claims that this will be trivial in the future, by turning planets into computronium. It's seems more likely that future humans will create the capacity reqiured for specific needs and not more than that. So will there be a need for ancestor simulations as premise B states?
  57. c) Are there other limitations?
  59. If we were able to do these kinds of simulations right now, would we create conscious characters living as a prisoner in a gulag simulation? We could do the same simulation, but just without entities capable of feeling pain and fear.
  61. Some have suggested that advanced civilisations wouldn't do this for ethical reasons – that it would be immoral to create conscious, digital beings that could suffer and therefore it would be banned. You can discuss the morality of creating conscious beings (e. g. having kids) but even if that ban exists, we must assume that if there are many civilisations and many individuals that are capable of making simulations, some might do it anyway.
  63. Some would argue that it's better to exist than to not exist, even for digital minds, so it's a morally good thing to create artificial consciousness. If that's the case, you shouldn't use up computer capacity to try to fool them that they live in a bygone historical period, but use all computers to host as many minds as possible.
  65. d) Conclusion
  67. We must conclude that even if possible, it will at least require a big investment in time, material and manpower for building computers and programming them, to create a simulation of a consciousness and a world that it can't easily detect as a simulation. So that makes us ask ”why do it?”, and that leads us to premise B.
  69. 4. Discussion of premise B
  71. a) Why would an advanced civilisation want to simulate a less advanced?
  73. Our own simulations is either for entertainment or research, so let's look at them separately.
  75. b) Are we part of a video game or a virtual ant farm?
  77. The term ”ant farm” is here meant to represent something you can get some entertainment from by just looking at or tinkering with. It's the satisfaction you get from an aquarium or a garden.
  79. For most of us, our lives don't seem to be exciting enough, to be part of an advanced civilisations computer games or entertainment. We use games to experience much more exciting and fantastic situations than we usually do in our own boring lives. Shouldn't future generations do the same?
  81. Why waste computer time on simulating my life, when there are so many historical persons and an infinte number of made up characters with more edifying or entertaining lives? And why just look at a simulation from the outside instead of being an active, immersed part of a virtual reality (that could be a historic scenario, if that is what interest you)?
  83. c) Are we part of a science project?
  85. Simulations in general can be very useful to evaluate complex systems that we can't use simple formulas to predict. You can for example use vague formulas with many variables and tweak these until the simulation resembles reality and then you know more about the real laws of nature that governs that system.
  87. We can also simulate experiments that are too difficult or expensive to do in real life. As an example, we can imagine that in the future, scientists can simulate a human cell and all biological activities inside that, so they can introduce different molecules and see what happens. This virtual drug testing would enable a leap in medicin that would justify just about any amount of money used to develop it. Then you can start to make changes to the DNA and simulate how the fertilized egg would develop into a fully grown being. To simulate every cell in a body, is still less complicated than to simulate that body plus consciousness plus the world around us.
  89. After that, you start to make up completely new genes and study what new abilities we can evolve in the human body and what new species we can create. How hard is that? Let's look at an example. To examine all variants of proteins containing exactly 150 amino acids (the median size in human proteins is 431 amino acids) we need to check the gene expression and phenotype in more than 10^195 simulations. Compare that to the number of atoms in the universe: less than 10^90, or planck time units since Big bang: less than 10^61. This means we can use a matrioshka brain around every star in the universe during all of their lifetimes, and still not be able to examine every possible new protein.
  91. If my life was part of a science project, it would probably be a study of alternative histories of the creator's universe to see how different natural/poltical/economical/sociological conditions would have changed their history. This wouldn't need conscious agents, though, or the extreme detailed reality we live in. It would be enough to let every human be represented by a random number generator, and see over many iterations, how the decisions of people affect history.
  93. If the purpose is to reconstruct history, it seems more likely that more undocumented times than ours, should be the main focus.
  95. This can be interesting, but it's very hard to see how it could justify taking resources from the much more rewarding task of eliminating death and evolving into whatever we want.
  97. But after we have done that, we have all the time in the world. Wouldn't we then start to be interested in our own history? Maybe, but ancestor simulations isn't real history. It's made up history. And by that time (when they are at least Kardashev type II civilisation) why should they care so much about the very ancient history?
  99. Much more useful would be to do ”descendant simulations” - i. e. predict and test a lot of possible future scenarios.
  101. d) Conclusion
  103. I haven't read a strong case for what a technologically advanced civilisation would benefit from simulating present day Earth. The reasoning has mostly been; ”If they could, they probably would”.
  105. We can now think of many things we want to use advanced computers for, that are more rewarding and interesting than ancestor simulations, and in the future they must be able to come up with many more ideas.
  107. This reasoning is based on our descendants still being restricted by their brains and human psychology. If they are artificial super intelligences, it's much harder to speculate about motives.
  109. 5. Discussion of premise C
  111. a) How many virtual worlds are there?
  113. It's a common belief that there should be many civilisations in the universe, much older than us, based on how many stars and galaxies there are. The Fermi paradox puts doubt on that assessment. And it's hardly a paradox anymore since it's been dissolved[5]. We don't know if that is the case in the mother universe, though.
  115. Even a single civilisation could produce many virtual worlds. In a post scarcity world, each individual could be equipped with what we would describe as an extremely powerful computer, but would they be enough for realistic simulations? Post scarcity civilisations are still limited by the matter and energy that is provided by the universe.
  117. Some people think that we could create these kind of simulations ourselves within this century. If we are simulated, that will mean that we will create a simulation within a simulation. That simulation in turn, could repeat the feat, and so on in many levels.
  119. The problem with this idea, is that each level must compute all sublevels, in all the layers down from it. In the case were you simulate every particle in the universe, this doesn't add much complexity – you just rearrange atoms into computers and continue to render them there. (It will just take a long time for a new level to form, if you start every universe from the Big Bang and you can only simulate a smaller world than the computer or much slower than real time).
  121. But when you do more simplified variants, each new level or parallel simulation requires as much computations as the original one.
  123. b) Specifically; how many ancestor simulations are there?
  125. Even if there are many simulated worlds in each physical universe - that doesn't increase the likelihood of us being in one, unless many of them are ancestor simulations (and simulations containing me, specifically), wich takes us back to premise B.
  127. Ancestor simulations is a very specific and small part of the ever increasing space of possible uses for computers.
  129. Some would say that if you have near infinte computer capacity, even if ancestor simulations are a tiny fraction of the programs they will run – a tiny part of infinity is still a lot. But I would say that the matter in a solar system is not near infinite and therefore the amount of computing power is not near infinite. The set of things you can use a computer for, on the other hand, is near infinite, and ancestor simulations is only one subset.
  131. When a simulation starts to create its own simulations (or if it reaches the technological singularity), one could fear that this would strain the original program beyond its parameters and cause it to be shut down, and therefore one should avoid it. This anxiety should also affect physical simulators (since they can't know if they are simulated or not) and reduce the number of civilisations that do ancestor simulations at all.
  133. c) Conclusion
  135. Bostrom implies that the only reason no civilisation will do a great number of ancestor simulations, is if they converge on deciding not to do them. But they could also diverge so much that recreating history in a virtual world, isnt a common interest.
  137. The limitation of the number of simulated worlds, is probably the number of people wanting to program and pay attention to them, more than the amount of computer hardware.
  139. 6. Probabilities
  141. a) What are the chances that I don't actually have a physical body?
  143. Elon Musk thought it was a ”one in billions” chance that we live in ”base reality”[3]. Physicist Neil deGrasse Tyson says that it is "very likely" we live in a simulation.[4]
  145. These guesses are based on their belief that we will be able and willing to simulate a lot of digital worlds in the future. And if we can do it, others have probably done it before us, e. g. our programmers.
  147. b) What must happen before we are able to do this?
  149. We won't suddenly start to make ancestor simulations with conscious agents. First will simplified brain emulations come. These might evolve in two directions - one is to refine them to resemble the biological brain more and more, until consciousness emerges. The other route is to optimize with regard to intelligence.
  151. If you create a conscious being in a computer, you immediately run into to moral and legal questions. Wouldn't these AIs demand more digital space to live in and to procreate in and wouldn't that be their ”human” rights? That could quickly use up all available computer capacity.
  153. When you on the other hand, use brain emulations to solve problems, you can make the brains bigger and faster, and tweak their properties to be more effective problem solvers. (This won't necessary lead to consciousness).
  155. The first AIs will not be dumbed down to be only as smart as us humans, and they can improve themselves very qucikly and then we will have ASI:s. They can answer any questions we'll have. so there won't be any need for ancestor simulations to do research.
  157. Self-improving ASI:s will probably lead to the technological singularity happening. It's called a singularity because beyond that point, all our predictions and reasoning are futile.
  159. All of this above, will lead to a very different society, that makes it harder for us to guess if our descendants will be interested in ancestor simulations.
  161. d) Conclusion
  163. There is no compelling arguments that world simulations will be done on household computers, beside a blind faith in the never-ending exponential growth of computer power. If states, institutions and companies will decide how to use their supercomputers, I doubt ancestor simulations are at the top of any list.
  165. Realistic human behaviour can be simulated with some algorithms and statistical data. It's hard to see why you need to have actual self aware subjects for any scientific reason or entertaining purposes.
  167. But if you need consciousness, it will be harder to program a world with seven billion different individuals living together, than to iterate one simulated mind seven billions times with some variations, wich should give more accurate data than the full world simulation. Versions with a single conscious agent should outnumber the more complete simulations, because of how many more of them you can do with the same resources. So if you believe you are a simulation, you should also conclude that you are probably alone in your reality, living in a digital Potemkin village. The world is fake and only pretends to have substance when you interact with it.
  169. The boring conclusion is that we don't have any grounds for an estimate of probabilities. My personal feeling is that it's unlikely that I live in a simulation, based on that the limited computer capacity will be used for better things than simulating less advanced worlds to a ridiculously high resolution. We can now easily generate much more data than we can analyse or even store. This must be true in the future too.
  171. 7. Consequences
  173. a) Does it matter if I am made of atoms or bits?
  175. It is obvious that we can live our lives in the belief that the observed world is real (i. e. made of matter) so why shouldn't we continue with that? We can't affect our reality anyway or even scientifically determine if it is a simulation or not (unless it's a bad simulation).
  177. Some say that if we understand that our world is digitally constructed, we could start to figure out the reason why this program was created.
  179. b) Should we worry about what the programmers will do with us?
  181. If we were to discover that it is relatively easy to create virtual worlds inhabited with self aware agents, we would conclude that we probably also are simulations. Would that matter to the original creator (since he/she has gone to extreme lengths to trick us that we are physical)? If we suspect that this will ruin the experiment because we weren't meant to become aware of it, so he/she/it will reset and start over again, then we shouldn't pursue that kind of research. That's one more X-risk we have to consider.
  183. (The first programmer isn't the only one we have to worry about, because any one of the programmers on levels above us can turn off his simulation and then every sublevel disappears.)
  185. But we can never guess the motivations of any of the programmers above us, if they don't want to us to know what it is. To us, our programmer is like God, outside our universe and not bound by our logic or laws of nature. And his/her programmer is the same to him/her. Even the top programmer, who is actually material, can't know that.
  187. If our God/programmer doesn't want us to know why he/she/it created our world, we can't figure it out and we can live as if there are no God/programmer. That means that life doesn't have a purpose or meaning, because only the one who created us can have had a reason to do so and hoped we would do something. It also means that there are no absolute morals.
  189. If our Creator wants us to know what the meaning of life is, he/she must have made that knowledge available to us. If you are Christian, you believe that the Bible is that message from the Creator, but other than religious scriptures, it's hard to see an easily detectable manual for life. This seem extra important if there is a simulated afterlife that is depending on our actions in this life.
  191. But again, assuming we are in an ancestor simulation, our programmers should be somewhat similar to us and our psychology. so we could try to understand why we would make such a program. What would our objectives be? Against this, one can argue that a civilisation that can do this kind of simulations probably must be post singularity and therefore so different from us that we can't compare them with us.
  193. 8. Final thoughts
  195. This paper hasn't produced any strict logical or philosophical arguments that I am unlikely to be simulated on a machine. I still have tried to explain why my intuition says I'm probably not.
  197. The main reason is that I can't fathom why it is necessary to deliberately include conscious and self aware, independent agents in a simulation. If consciousness is an emergent property of an detailed simulation, the question is why they would need such a high resolution that would require orders of magnitude bigger computers, than more simplified simulations.
  199. Unless the Bible is a message from the Creator, we have no grounds for speculations about the reason why we are created/simulated. This kind of speculations can be fun, but never be the basis of any life philosophy or ideology. And that's my summation – this idea is an interesting but rather pointless thought experiment.
  203. References
  205. [1] ”Are you living in a computer simulation?”, Nick Bostrom, Philosophical Quarterly, 2003
  207. [2] ”Quantized gravitational responses, the sign problem, and quantum complexity”, Zohar Ringel and Dmitry L. Kovrizhin, Science Advances, 2017
  209. [3] Code Conference, 2 juni 2016
  211. [4] 2016 Isaac Asimov Memorial Debate
  213. [5] ”Dissolving the Fermi Paradox” Anders Sandberg, Eric Drexler and Toby Ord, Future of Humanity Institute, Oxford University, 2018
RAW Paste Data