Sabine Hossenfelder's video "The Simulation Hypothesis is Pseudoscience"
A physicist attempts philosophy and computer science and the result is a disaster
Sabine Hossenfelder is a theoretical physicist and science popularizer with a series of YouTube videos about physics, so I was excited to see that she posted a video about the Simulation Hypothesis entitled “The Simulation Hypothesis is Pseudoscience.”
Unfortunately, the video turns out to be a confusing mess.
Hossenfelder starts with her summary of Nick Bostrom’s argument:
“If there are a) many civilizations, and these civilizations b) build computers that run simulations of conscious beings, then c) there are many more simulated conscious beings than real ones, so you are likely to live in a simulation.”
This is poor framing. It suggests to viewers that we’re talking about right now and the universe we experience, while if the hypothesis is true, the universe and the time we experience is just a backdrop running in the simulation. While the Simulation Hypothesis can refer to a generalized version of the argument, Hossenfelder refers to this as Bostrom’s argument, which is incorrect. Bostrom specifically talks about post-human civilizations and ancestor simulations. Another somewhat misleading point is that Bostrom is not talking about whole civilizations being interested in running simulations but individuals within them, which increases the chances: “…all posthuman civilizations lack individuals who have sufficient resources and interest to run ancestor-simulations…”
Hossenfelder goes on:
“But before I explain why the simulation hypothesis is not a scientific argument, I have a general comment about the difference between religion and science. Take an example from Christian faith…”
Classic Poisoning the Well fallacy. The audience will be predisposed to compare the hypothesis to religion and not take it seriously before any argument is even made. There are many theories in theoretical physics where it is very difficult or impossible to come up with experiments to prove or disprove them, e.g. Everett’s Many-Worlds Interpretation or the String Theory. One might call these sorts of theories metaphysical, fine, but it doesn’t make them into disproven pseudoscience or religions. In fact, some tests of the Simulation Hypothesis have been proposed, e.g. https://ijqf.org/archives/4105.
“It’s a religious story because it doesn’t explain how the healing supposedly happens.”
That’s not what makes something a religious story at all. In fact, we have drugs with unknown mechanisms of action that are used in medicine every day, e.g. lithium.
“Maybe there aren’t any other civilizations, or they aren’t interested in simulations. That wouldn’t make the argument wrong of course, it would just mean that the conclusion can’t be draw. But I will leave aside the possibility that one of the premises is wrong because really I don’t think we have good evidence for one side or the other.”
This ignores the central point of the hypothesis. There is at least one civilization – our own. Unless there is a Great Filter in our future, our descendants should have access to a tremendous amount of cheap computing power and we’re starting with millions of people interested in history. We already have near human vision-fidelity VR and 3D audio and we can easily imagine how a decade of advancements in haptics and related technologies, not even including brain-computer interfaces, will be able to make the metaverse indistinguishable from reality. And that’s just the near future. (As an aside, it will likely always be cheaper to create a brain in a vat-type simulation than simulating the whole universe but that’s for another article).
“But nobody presently knows how to reproduce General Relativity and the Standard Model of particle physics from a computer algorithm running on some sort of machine.”
Here we get to Hossenfelder’s main objection to the Simulation Hypothesis. Four big problems are instantly obvious in this one sentence, revealing a serious lack of understanding of the hypothesis:
1. “presently” – but time in the simulation is not the outer time. The sim can run millions of times faster or the nature of time could be altogether different than we expect.
2. “nobody” – as in currently alive humans. Not what the simulation argument is about.
3. “machine” – outer universe (or our future descendants who might be AI or some human/AI hybrid) can have machines with profoundly different capabilities.
4. We only need to compute what conscious humans experience through their senses. There is no reason we can’t be fooled into thinking that the whole universe is out there and we’re not experiencing the results of computations. Even we ourselves can generate simulated universes that seem real to us like it’s not a big deal: our dreams.
Hossenfelder does not provide any proof that the Standard Model or General Relativity are uncomputable. Even with our weak understanding of the foundations of physics, we can already do things like numerically simulate evolving binary black hole systems: https://arxiv.org/abs/gr-qc/0511103 (2006).
“Indeed, physicists have looked for signs that natural laws really proceed step by step, like in a computer code, but their search has come up empty handed.”
Computers, of course, can calculate in parallel, not just sequentially, though it doesn’t change the number of computable functions. It means little that a few physicists haven’t found any signs that natural laws proceed step by step – and if they don’t then why wouldn’t we be able to design a computer that takes advantage of this? In general, in physics, we don’t yet have a deep understanding of any of the most basic questions about nature. We don’t even know if time is quantized.
“It also doesn’t help by the way if you assume that the simulation would run on a quantum computer. Quantum computers, as I have explained earlier, are special purpose machines. Nobody currently knows how to put General Relativity on a quantum computer.”
Again, it doesn’t mean much that nobody currently knows how to do it. No practical quantum computers exist yet, so obviously, nobody tried to run anything useful on them and more quantum algorithms will be found that run with lower time complexities than on classical computers.
“…for it to work, a civilization needs to be able to simulate a lot of conscious beings, and these conscious beings will themselves try to simulate conscious beings, and so on.”
We have no idea how to even approach the hard problem of consciousness, so making arguments based on consciousness makes little sense. It might not be necessary to simulate many conscious beings – most could be NPCs. Or the simulation does not need to be run until simulated beings can themselves simulate others. Or the sim can be turned off as needed. Or the laws of the simulated universe can prevent it. Etc.
“What algorithm can identify conscious subsystems and their intention and then quickly fill in the required information without ever producing an observable inconsistency. That’s a much more difficult issue than Bostrom seems to appreciate. You cannot in general just throw away physical processes on short distances and still get the long distances right.”
Your own brain produces simulations without observable inconsistencies every night (unless you’re able to have lucid dreams).
There has been tremendous progress in Machine Learning in the last decade. We now get superhuman performance in many games (with complete and incomplete information), language models that produce plausible English text, very accurate computer vision systems, and even computer-generated art. For example, some research shows that machine learning can simulate physical processes well without solving differential equations. What’s fascinating is that neural nets can do valuable things without the need for us to have a full understanding of their internal processing and representations. This increases the chances that we’d want to actually run them, analogously to how we might want to run ancestor simulations.
“Climate models are an excellent example. We don’t currently have the computational capacity to resolve distances below something like 10 kilometers or so. But you can’t just throw away all the physics below this scale.”
In a simulation, we only care about things we observe, and we can estimate the rest. Having some randomness is not a problem. If the beings in the simulated universe don’t see an issue, then it’s not a problem (and even if they do, this can just be something they won’t be able to explain or, more interestingly, an indicator to them that they are simulated). We do weather forecasts through simulations exactly because of how useful these simulations are, as opposed to trying to solve exact equations.
This concludes Hossenfelder’s video. It presents no valid arguments, gives no proof of anything, and does not cover any interesting Simulation Hypothesis issues.