I’m hoping to hear arguments against in particular (because it currently seems probable to me, though I haven’t read about it, I’ve just been thinking about it on my own), but all arguments are welcome of course (especially as others clicking onto this might be interested).
[Question] Does one have reason to believe the simulation hypothesis is probably true?
I think the question is underspecified. There is no single “simulation hypothesis” as I understand the term; there are multiple large classes of hypotheses that could be expressed as “the universe is a simulation.”
There exists a universe in which sapient beings have access to large amounts of computing power which they are dedicating towards simulating our civilization/universe. Or a class of such simulations. They have access to sufficient computing power that such simulations are cheap, and they run enough of them that most beings exist in a simulation rather than the base-level universe. This would include ancestor simulations, works of fiction/games where we happen to be NPCs, models for exploring alternate histories and futures, models of possible alien lifeforms, all kinds of purposes.
There exist (infinitely?) many layers of such simulations, nested, driving the probability of being in a simulation to ~1. (Time may flow at very different rates between levels, which might confuse the anthropic reasoning, I’m not sure.)
Base-level reality has physics allowing for infinite error-free computing power, so someone somewhere decided to run all possible programs in order, or some subset thereof.
There may not be a base-level physical reality at all. It’s turtles all the way up and all the way down.
Base reality is computable, infinite, and randomly initialized, such that it by default includes infinitely many instances of all possible simulations and programs running in parallel forever.
The idea of “simulation” is itself confused in some way. Reality is fundamentally mathematical and computational.
Alternatively, the idea of “existence” or “reality” as a property separate from mathematical consistency or some form of computability is in some way confused.
I am a Boltzmann brain, or something dreamed up within one, or equivalent (for example, as a fluctuation in a max-entropy infinite universe in what I would regard as the far future).
I think the term “simulation hypothesis” usually refers to (1) or (2) only? But I’m not sure if that’s universal, and I find the others just as interesting if not more so.
For me, a main argument for is that there are many simple sets of rules that could be the physics of universes containing vast computing power—much simpler than our own apparent physics. But the main argument against, from my POV, is that our own apparent physics is fairly computationally intensive to simulate as far as we know, and I can get a similar effect just by assuming our own universe is infinite in spatial extent and randomly initialized. I don’t have a sufficiently precise, unambiguous, and natural concept of simplicity that would guide me to one or the other conclusion in terms of priors.
That said, is there a particular reason you’re looking to be argued out of the idea?
Like any mystical explanation (that is, a model of underlying causality which cannot be tested and doesn’t have any observable or communicable effects), the reasons for believing are mostly about how it makes you feel, and how others react to your discussions.
If the simulation is good and the designer doesn’t want the sims to find out, they won’t find out.
this response is disappointing to me.
a model of underlying causality which cannot be tested and doesn’t have any observable or communicable effects
as a bayesian, you can reason about the probability of hypotheses you can’t test. for an example in this context, you could check Bostrom’s paper on this.
the reasons for believing are mostly about how it makes you feel, and how others react to your discussions
as far as i can tell, this is not true for me.
as a bayesian, you can reason about the probability of hypotheses you can’t test. for an example in this context, you could check Bostrom’s paper on this.
How do you know when your reasoning is faulty (as all human reasoning is) without experimental feedback? In absence of one, it is indeed about how it makes you feel.
Examples of ways to help know whether ones reasoning is sound in absence of experimental feedback include
Talking about it with others
Looking for opposing arguments, and then looking for counterarguments to any counterarguments you come up with
Converting ones reasoning into a syllogism, and checking it for logical fallacies, or having someone else check it
Asking oneself if one wants to believe a given conclusion for other reasons (i.e checking for motivated reasoning)
Reading a lot about avoiding biased reasoning, and setting a habit of automatically applying that to your own thoughts
In absence of [experimental feedback], it is indeed about how it makes you feel.
My low-credence belief that the simulation hypothesis is probable generally doesn’t make me feel anything. When it ever does, the feelings have not been good ones.
Furthermore, some important beliefs among many rationalists relate to future forecasting related to AI (e.g., ASI timelines, or beliefs about existential risks, or thoughts about anthropic shadow effects). If your intent is to argue that it is futile to reason in absence of the ability to test hypotheses, you may want to make a more robust argument and engage with rationalist literature on the subject first. (E.g., this is discussed in When Science Can’t Help; it may be discussed in other posts in the sequences too, but I can’t refer you to them because I don’t remember sequence post titles, sorry. It looks like there are also relevant posts under the “Practice and Philosophy of Science” tag)
These are all good ideas, but they are not a substitution for testing.
If your intent is to argue that it is futile to reason in absence of the ability to test hypotheses
The goal of reasoning is to eventually connect with experiment, i.e. to make accurate predictions. I have read the sequences (and helped name the book), but I strongly disagree with a lot of the posts. Specifically, When Science Can’t Help is very misleading. Sorry you got misled. The simulation hypothesis, is not a hypothesis, it’s a speculation with no way to connect to the real world, whatever it might be. I am not saying it’s wrong, it’s not even wrong. Focus on something that can be helpful to you and ignore this rubbish.
Tragically, you can’t connect to the real world using experiment alone. The whole problem is that the same experimental results can be predicted by different realities. The goal of reasoning as opposed to empirical observation is to correspond to reality.
Tragically, you can’t connect to the real world using experiment alone.
Yes, you absolutely need both. I don’t think anyone argues that point?
I’d be interested in reading about why you think it’s misleading. (Feel free to link relevant writing if any exists)
This is probably as far toward pure Bayesianism as is reasonable to go:
I read this. It seemed like it was very oriented towards an audience which had the impression that “unfalsifiable claims are not real science,” and it seemed to do a good job of explaining to that audience its objections.
It seems to try to differentiate hypotheses which are (a) hard to test, but meaningful and possibly true, from ones which (b) don’t really say anything at all and are compatible with however reality might truly be.
I think that the simulation hypothesis is of the former type.
First, it is saying something discrete about the nature of reality, which differentiates it from ideas which aren’t saying anything. And I agree that this quality is important.
The second point—being testable in principle, even if difficult or impossible for now—is less important to me; it seems to me like even if there were no way to empirically test an idea even in principle, we could still have some probability attached to it being true (just like we can in situations where an idea isn’t testable yet, but may be with future technology).
Still, I’ll note that we could in principle observe evidence which increases or decreases our probabilities in being simulated. For example, we could (in principle) discover a glitch in reality which increases the probability, or (again in principle) observe something really obvious like the sky tearing open to reveal text which reads “you’re in a simulation” (not to imply there would not be other probable explanations for that, were it to happen).
Similarly, there’s ways empirical evidence could decrease our probability in the hypothesis, though they’re a little more complex to imagine. E.g., maybe we search very hard for such glitches and don’t find any, and this decreases (at least slightly) the probability we’re in a simulation not meant to prevent discovery of this fact; or, for a more strong decrease, maybe an ASI solves physics, and finds some compelling reason to believe we’re in the base universe.
I appreciate your engaging with me. I understand how it’s probably frustrating to have a minority view and see people constantly say the opposite thing. If you have other writings or arguments as to why “When Science Can’t Help” is misleading, I’d still be open to reading them.
Here is my best argument against the simulation hypothesis:
Simulations typically have a reduction in dimensionality, at least the ones we have made so far. Which is to say: you cannot simulate the whole of the universe within itself. Our simulations are more granular and smaller than the real world, even if they become remarkably close. This reduction in detail can be temporal or spatial.
The simulation hypothesis supposes that the number of simulations would be astronomically high because of recursive simulations in simulations. This is the part that I take issue with. Assume that we live in simulation level X (some point in the chain of simulations). Anything that happens in the simulation X+1 (the simulations we make) would need to be computable by a set of operations in simulation X. This is similar to how a virtual machine ultimately runs as CPU instructions on the host machine.
This means that Simulation X bounds the complexity of every nested simulation (Simulation X+1...X+∞). If you run a virtual machine A inside a virtual machine B, on physical host C virtual machine A is still expressible as a computation of the host machine. You cannot create a virtual machine with higher specs and have this run in real-time, since the complexity of this chain is bounded by the root node.
Note that there is also overhead with every layer of simulation. Our entire universe is not interested in running simulations, only a small fraction of the compute under the control of intelligent life. A god-like civilisation that has conquered the universe may be able to dedicate a significant portion, but this necessarily cannot be 100%.
This means that the number of simulations should be much more finite than typical formulations of the simulation hypothesis depict.
This does not mean that I rule the possibility out. Even if the chain is short there may be many simulations which we could exist in. We might be the only simulation. But this possibility seems no more profound or likely than any other unprovable statement.
I don’t think this is essential for the argument. For example, we could run a sim where constructing computers is impossible. And according to Bostrom: “the simulation argument does not purport to show (and I do not believe) that the Sims outnumber the [non‐Sim] humans.”
This presumes that we have to fully simulate the whole universe in detail, without room for approximations, and that the physical laws of the outer universe are the same as ours.
Thanks for the response Lech
I was not aware of this as a common caveat to the simulation hypothesis. If it is, I still do not see a reason to believe that it is probable that conscious beings with our set of experiences should expect to exist in a simulation. It makes more sense to assume that this is the base reality if we expect those minds to be in the majority.
I agree with this, these approximations are related to the granularity I was describing. My belief is that there are only so many approximations you can make before the simulated universe is not detailed enough to express individual consciousness.
I am fully open to the possibility that a civilisation in a reality with more complicated physics is simulating ours, but I see no reason to believe this is probable, or more probable than many other possible claims (eg. brain in a vat).
The best response I’ve heard against the simulation hypothesis is “If we’re simulated, why aren’t we asked to do meaningful work?” Think about why we would want to simulate personalities in our near to medium term future: to answer questions, to help us work faster, to entertain us, to be our companions, etc.
So why aren’t we asked to complete the equivalent of Mechanical Turk questions? Is compute is so unfathomably cheap that they there’s no economic reason we would need to “pay rent” so to speak? Even if that’s the case, why don’t our creators want to interact with us directly? In other words, if we’re all video game characters, who is the player character? Why do I have consciousness, if I’m just an NPC?
And on and on. The crux is whether you believe that in the world that allows us to be simulated, it’s more likely that our creator just wants to view us through a thick pane of glass than they’ll want to play with us. My intuition is very much on the later.
I’ve seen this sentiment expressed in reverse: Isn’t it fascinating that we’re living in such a pivotal moment when AGI seems to be on the verge of emerging? If we are alone in the universe, how this unfolds might be the most significant event since the Big Bang.
This seems true if we anthropomorphize the simulator; another relevant question there might be “Why is there so much suffering?”. However, it seems to me that the vast majority of simulations would instead be initiated by intelligences which terminally value simulation. Even if such intelligences were very rare (i.e if we assume the chance of an intelligence having that value were as small as 1/n where n is all possible values), it may create disproportionately many simulations, compared to other intelligences which may not have reason to create any, or which may only create a few for instrumental reasons during the earlier stages of their existence. (And given simulators are developing in our universe, the chance may not be as small as 1/n)
I also think it may be plausible for stochastic physical interactions between matter in a lifeless universe to naturally end up in certain computational patterns, and for life to arise within them without any way of observing the nature of its ‘universe,’ but this idea is too complex for me to write about for now.
Thanks for responding btw!
I’ve been checking Joscha Bach’s blog these days, and I had found this blog post on the topic which I think goes to interesting depth on the question:
Do we live in a simulation?
Here’s an attempt:
(1) Maybe by default simulations aren’t conscious. You have to go out of your way to make a conscious being, e.g. maybe it has to be biological or something fancier than whatever the most efficient-yet-accurate simulation would be.
(2) Also, maybe there are pretty universally enforced ethical prohibitions against going out of your way to make conscious beings and then putting them in such a nasty horrible situation as 21st-century Earth. The number of entities with the will & means to violate this prohibition is vanishingly small, small enough that any conscious simulations created by them are outnumbered by the conscious non-simulations / evolved creatures like us.
(2) seems like a good point to me! I hadn’t considered that before, ‘whether the expected being is simulated’ as opposed to ‘whether the expected universe is simulated’.