Our Reality: A Simulation Run by a Paperclip Maximizer

Reality is whatever you should consider relevant [LW · GW]. Even exact simulations of your behavior can still be irrelevant (as considerations that should influence your thoughts and decisions, and consequently the thoughts and decisions of those simulations), similarly to someone you won't possibly interact with thinking about your behavior, or writing down a large natural number that encodes your mind, as it stands now or in an hour in response to some thought experiment.

So it's misleading to say that you exist primarily as the majority of your instances (somewhere in the bowels of an algorithmic prior), because you plausibly shouldn't care about what's happening with the majority of your instances (which is to say, those instances shouldn't care about what's happening to them), and so a more useful notion of where you exist won't be about them. We can still consider these other instances, but I'm objecting to framing their locations as the proper meaning of "our reality". My reality is the physical world, the base reality, because this is what seems to be the thing I should care about for now (at least until I can imagine other areas of concern more clearly, something that likely needs more than a human mind, and certainly needs a better understanding of agent foundations).

The problem is, when we simulate cars or airplanes in software, we don't do it at molecular level. There are big regularities that cut the cost by many orders of magnitude. So simulating the Earth with all its details, including butterflies and so on, seems too wasteful if the goal is just to figure out what kind of AI humans would create. The same resources could be used to run many orders of magnitude more simplified simulations, maybe without conscious beings at all, but sufficient to predict roughly what kind of AI would result.

I tweeted about something a lot like this

https://xcancel.com/robertskmiles/status/1877486270143934881

I find this far more convincing than any variant of the simulation argument I've heard before. They've lacked a reason that someone would want to simulate a reality like ours. I haven't heard a reason for simulating ancestors that's either strong enough to think an AGI or its biological creators would want to spend the resources, or explains the massive apparent suffering happening in this sim.

This is a reason. And if it's done in a computationally efficient manner, possibly needing little more compute than running the brains involved directly in the creation of AGI, this sounds all too plausible - perhaps even for an aligned AGI, since most of the suffering can be faked, since the people directly affecting AGI are arguably almost all leading net-positive-happiness lives. If what you care about is decisions, you can just simulate in enough detail to capture plausible decision-making processes, which could be quite efficient. See my other comment [LW(p) · GW(p)] for more on the efficiency argument.

I am left with a new concern: being shut down even if we succeed at alignment. This will be added to my many concerns about how easily we might get it wrong and experience extinction, or worse, suffering-then-extinction Fortunately, my psyche thus far seems to carry these concerns fairly lightly. Which is probably a coincidence, right?

I find some of the particular arguments' premises implausible, but I don't think they hurt the core plausibility argument. I've never found it very plausible that we're in a simulation. Now I do.

My alternative hypothesis is that we're being simulated by a civilization trying to solve philosophy, because they want to see how other civilizations might approach the problem of solving philosophy.

my most fun talk made a similar claim:

I find this argument fairly compelling. I also appreciate the fact that you've listed out some ways it could be wrong.

Your argument matches fairly closely with my own views as to why we exist, namely that we are computationally irreducible [LW(p) · GW(p)]

It's hard to know what to do with such a conclusion. On the one hand it's somewhat comforting because it suggests even if we fuck up, there are other simulations or base realities out there that will continue. On the other hand, the thought that our universe will be terminated once sufficient data has been gathered is pretty sad.

I don't know about "by a paperclip maximizer", but one thing that stands out to me:

If we're in a simulation, we could be in a simulation where the simulator did 1e100 rollouts from the big bang forward, and then collected statistics from all those runs.

But we could also be in a simulation where the simulator is doing importance sampling - that is, doing fewer rollouts from states that tend to have very similar trajectories given mild perturbations, and doing more rollouts from states that tend to have very different trajectories given mild perturbations.

If that's the case, we should find ourselves living in a world where events seem to be driven by coincidences and particularly by things which are downstream of chaotic dynamics and which had around a 50/50 chance of happening vs not. We should find more such coincidences for important things than for unimportant things.

Hey, maybe it's not our fault we live in the clown world. Maybe the clown world is a statistical inevitability.

Yet the universe runs on strikingly simple math (relativity, quantum mechanics); such elegance is exactly what an efficient simulation would use. Physics is unreasonably effective, reducing the computational cost of the simulation. This cuts against the last point.

This does not seem so consistent, and if the primary piece of evidence for me against such simulation arguments. I would imagine simulations targeting, eg, a particular purpose would have their physics tailored to that purpose much more than ours seems to (for any purpose, given the vast computational complexity of our physics, and the vast number of objects such a physics engine needs to keep track of). For example, I'd expect most simulations physics to look more like Greg Egan's Crystal Nights (incidentally this story is what first convinced me the simulation hypothesis was false).

One may argue its all there just to convince us we're not in a simulation. Perhaps, but two points:

1. Given the discourse on the simulation hypothesis, most seem to take our physics as evidence in favor of it, as you do here. So I don't think most think clearly enough about this for our civilizational decisions to be so dependent on this.

2. The simulators will have trade-offs and resource constraints too. Perhaps they simulate few highly detailed simulations, and many highly simplified simulations. If this is exponential, in the sense that as the detail decreases the number of simulations exponentially increases, we should expect to be in the least detailed world consistent with the existence of sentiences and for which its not blatantly obvious we're in a simulation.

Of course this argument would break given sufficiently different physics from ours, enabling perhaps our world to be simulated in as much depth as it is very cheaply. But then that seems intuitively at least very unlikely & complex a hypothesis.

This and other simulation arguments become more plausible if you assume that they require only a tiny fraction of the compute needed to simulate physical reality. Which I think is true. I don't think it takes nearly as much compute to run a useful simulation of humans as people usually assume.

I don't see a reason to simulate at nearly a physical level of detail. I suspect you can do it using a technique that's more similar to the simulations you describe, except for the brains involved, which need to be simulated in detail to make decisions like evolved organisms would. But that detail is on the order of computations, not molecules. Depending on which arguments you favor, a teraflop or a couple OOMs above might be enough to simulate a brain with adequate fidelity to capture its decision-making to within the large uncertainty of exactly what type of organisms might evolve.

"physical" reality can be simulated in very low fidelity relative to atoms, because it's not the important part for this and many proposed simulation purposes. It just has to be enough to fool the brains involved. And brains naturally fill in details as part of their fundamental computational operation.

For this purpose you'd also want to get the basic nature of computing right, because that's might well have a large effect on what type of AGI is created. But that doesn't mean you need to simulate the electrons doiing quantum tunneling for wafer transistors; it just means you need to constrain the simmulation so the compute behaves approximately like the quantum tunneling transistor is the base technology.

On this thesis, the compute needed is mostly that needed to run the brains involved.

This isn't a necessary twist, but one might even cut corners by not even simulating all humans in full fidelity. All of society does play in to the factors in the AGI race, but it's possible that that an AGI could run many times more simulations if they made only key decision-makers simulated in full fidelity and somehow scaled down others. However, I want to separate this even weirder possibility from both the main argument of the post and my main argument here: simulation for many purposes can probably be many, many OOMs smaller than the atomic level - possibly using very few resources if a lot of technology and energy is available for compute.

I'll make a separate comment on the actual thesis of this post. TLDR: I find this far more compelling than other variants of the simulation argument.

Another possibility is that the beings in the unsimulated universe are simulating us in order to do a Karma Test [LW · GW]: a test that reward agents who are kind and merciful to weaker agents.

By running Karma Tests, they can convince their more powerful adversaries to be kind and merciful to them, due to the small possibility that their own universe is also a Karma Test (by even higher beings faced with their own powerful adversaries).

Edit: see my reply [LW · GW] about suffering in simulations.