I’ve written a few pages of introduction at the beginning to get people into the general ballpark, but the rest is pretty much our shorthand notes of interesting features in the space. So don’t expect it to be a finished, comprehensible document; it’s just a bunch of vaguely clustered concepts. We were exploring a space, and our only goal was to see if the question is worth thinking about more. We think it is.
Can we find out if our universe is real?
Section 0: Introduction (are we completely nuts?)
The core idea here is that maybe we’re living in a simulated universe, and maybe we can detect that. If we could, it would be the single most important scientific discovery in the history of mankind.
It’s also possible that we’re in a simulated universe but we can’t possibly detect it. It’s a real possibility, and it’s the one people usually bring up as a reason why there’s no point in discussing it. But unless undetectably simulated universes are the only kind of simulated universe we can imagine, then we shouldn’t stop asking questions yet.
If there’s any possibility we’re both in a simulated universe and that we could detect that, it’s certainly worth spending some time on. How would we find out? What would the consequences be? Could there be bad consequences?
The positive possibilities include potentially being able to break the known laws of physics to help solve our problems, being able to negotiate with or steal parts of the “parent” universe and gaining knowledge for its own sake of the way things really are in a grand sense. The negative possibilities include crashing our universe (if it’s a badly written computer program and we experiment with edge cases of its behavior) or getting our universe shut down (if we do something that gets negative attention from whoever’s in charge.)
Are there plausible circumstances in which we might succeed? Well, here’s one example of how it might all play out:
Let’s hypothesize that our universe is a big brute force particle simulation running on a giant computer in our “parent” universe -- that is, there are a bunch of “gluons” represented as some sort of “bits” in the computer, and something resembling our mathematics is performed on them so that they interact according to what we perceive to be the laws of physics. That simulation then is the true underlying nature of everything in our universe.
And let’s further hypothesize that the math they use has some of the kinds of subtle errors that our own computations do -- that digits get chopped off past some number of decimal points, like getting past the edge of a calculator screen. These errors don’t interfere with the operation of our universe, but if you carefully measure, you can detect them as highly suspicious inconsistencies.
We run some tests in a particle accelerator, trying to create situations that magnify the subtle errors into something we can measure. And on the 574th setup, we discover that the usual laws of physics are a little off in this really specific case. That’s enough for a Nobel prize on its own, but in this case, if it turns out to strongly confirm our hypothesis, then we’ve just acquired the first hard evidence that there’s a whole universe behind what we can see.
It would open up a new branch of physics, one with absolutely limitless potential compared to the laws we’re used to dealing with, because this branch pulls back the curtain and gives us a peek at an underlying reality that we have never before glimpsed.
Section 0.5: Where we’re going with this
Section 1: The Problem Space, talks about the kinds of categories we can imagine simulated universes falling into.
Section 2: Goals covers the end goals we’re trying to reach.
Section 3: Crazy Side Ideas covers things that came up while we were thinking about this topic, that weren’t directly related. (But they’re a great side-effect of thinking about this topic!)
Section 4: Implementation Sketches talks about a few simulations we tried to imagine building, to give us hints into the kinds of inconsistencies that might accompany them.
Section 5: Suspicious Things lists things we see in our own universe that might be the result of being in a simulation -- phenomena that seem unlikely results of an undirected natural process.
Section 6: Testable Hypotheses is our first pass at imagining experiments whose results might conclusively prove that we’re in a simulation.
Section 1: The problem space
What are the ways that some intelligent agent might go about simulating our reality? We can imagine a lot of ways to do it, plus a whole class of things that we can’t possibly imagine. But it gives us some starting places to begin our search.
It seems like most scenarios would have some common elements:
- A “parent” universe with some sort of intelligent life that knows how to create reality simulations
- A “creator” -- an intelligent actor with some reason for creating a simulation
- A “simulation implementation” -- how did they do it? A big computer with a bunch of simulated particles? Something more efficient than that? A bunch of simulated brains?
Parent UniverseHere are some ideas for what the parent universe might look like:
- Just like our own in almost all respects
- Just like our own but with a lot more matter or energy
- Itself a simulation
- Similar laws in a higher dimensional space
- No similarity (we’re like a game-of-life experiment, and have no hope of understanding the parent universe)
Parent Universe timescaleHow does time in the parent universe relate to our notions of time?
- In lock step with our own (a “realtime” simulation)
- Time goes massively faster in our universe (they’re watching galaxies grow)
- Time goes massively slower in our universe (we look like statues because they think so fast)
- No correlation at all
- Time is a function of resources, so it sometimes runs slower or faster. (Does building supercomputers make our simulation run slower?)
Simulation ReasonsWhat kind of being would create our reality, and why?
Incomprehensible - we couldn’t possibly understand
Programmer - something like a person with a big computer
Dumb AI with lots of time - simple algorithm trying to optimize for something and building our universe as a very inefficient way of accomplishing its goal
Major Superintelligence - we can’t comprehend their brain
Accident - We’re an eddy in a cloud in some atmosphere; there’s a “parent” universe that created our big bang, but nobody did it on purpose.
End user - Somebody else wrote the simulation, and lots of other agents use it to create their own universes for their own purposes.
Simulation ReasonsWhy might they have created our reality?
- Business (someone is selling the art/code/music we make, or the chemical compounds and plants and animals that evolved on our world)
- Military (threat simulation)
- Attempt to create superintelligence (Are we smarter than our creators and in situations analogous to what they face?) -- maybe it’s easy for them to create big brute-force particle simulations and they’ve never bothered to figure out how to code up an AI directly, or they’re waiting for us to do it.
- Simple computations being mapped out as an attempt to do Solomonov induction on one’s own universe (though the line between being a sim and being the basement gets blurry if our universe is a simple computation, especially given Tegmark 4 type realities.)
- The experiences of the actors in our universe are somehow terminal-utility-ful to the simulators. (We’re living out their fantasies)
- Some other aspect of our universe is somehow terminal-utility-ful to the simulators. (Our universe is an abstract artwork of some sort hanging on a wall somewhere.)
- Psychological simulations of game-theoretically relevant situations
- Academic study of anthropology, sociology, psychology
- Historical curiosity -- see the primitive aliens!
- Education (class project)
- Blackmail Fodder (basement super villain is holding us hostage)
- Casual unintentional (it’s a screensaver)
- Causal intentional (it’s a game, or a child’s toy)
If we’re made of actual particles in a larger universe (eg., a bunch of hydrogen in a giant petri dish)- Food (tasty nebulas)
- Chemical process (making helium)
- Accidental byproduct (the hydrogen was fuel and we’re just exhaust)
- Biological process
Simulation ImplementationsWhat are the broad ways we can imagine trying to simulate a universe?
Basic building blocks- Brute-force gluon (particle) simulation
- Optimized particle simulation (fancy math that saves cycles or storage)
- High level brain simulation (just a brain living in a video game, like the Truman Show)
- Experience simulation (we might be “real” people in the parent universe whose realities are being totally subverted, like the Matrix)
- Actual particles (hydrogen in a petri dish)
- Their space has more dimension(s) than ours and our universe is just a bunch of particles held in a plane by some magnets or something.
- Technology not conceivable by us (catchall category)
- Very (large) intelligent mind on drugs/dreaming/day dreaming
- Unintentional butterfly effect in another universe.
It’s also possible that our universe is a simple computation that has some platonic reality of its own and is also being simulated by many other universes (which again is not something we could detect, except in cases where one of them has a lossy/buggy implementation or meddles.)
Simulation Scopes -- what’s being actually simulated, and what’s just a facade?- Our entire universe is a simulation
- Our Galaxy (and the rest of the universe is just fake “wallpaper”)
- Our Solar System (and the rest of the universe is just fake “wallpaper”)
- The planet Earth (and the rest of the universe is just fake “wallpaper”)
- Some group of people (you and your friends)
- Individual person (solipsism -- you’re the only one who goes on thinking when you’re alone)
- Individual experience (it’s only you, and you’ve only been around for a week)
Lazily evaluated mix of the above -- things don’t exist until we look at them.
Simulation Resource LimitationsRather than choosing explicitly what to simulate, maybe simulation parameters are a function of available resources -- perhaps we’ll terminate when:
- we run out of CPU cycles or RAM
- after time has elapsed in the parent universe
- after some goal is met (once intelligent life forms and reaches a certain level, the simulation freezes for evaluation)
Section 2: Our Goals- Create experiments to determine whether we’re in a simulation, and then
. Try to contact the creator(s) and
. make friends
. do recon and figure out friend/foe
. Try to root the machine and explore the parent universe without permission
. Try to break the rules
Section 3: Crazy Side IdeasCan any of these strategies work for us? (eg., could building a big particle simulator be easier than building an AI?)
Cosmic Background Radiation = Federation radio jammer to keep immature races from hearing signals from advanced races.
Section 4: Implementation SketchesIf we were going to simulate a universe, how would we do it?
Scenario 1: Brute-force universe-wide particle simulation on a digital computer
- Create the basic gluons classes and rules for interaction
- Store the gluons in some sort of data structure
- Run in time-slice epochs, with each gluon interacting with each other one, one at a time. Then repeat.
Scenario 2: Efficient particle simulation
- Like scenario 1, but a clever algorithm lets big clouds of gluons act as a single unit without having to worry about each individual one.
Scenario 3: Brain particle simulation -- scan some “real” brain, then simulate all the particles making up that brain and feed it fake stimulus.
Scenario 4: AI simulation of all humans -- a superintelligent being “pretends” to be all of us and our surroundings.
Section 5: Suspicious ThingsDoes our reality seem like the kind of thing that might be easy to simulate, or annoyingly difficult?
- Our universe appears to be discretized
- Humanity hasn’t stone-aged itself yet
- Drake’s equation: why do we appear to be alone?
- Small number of elementary particles -- there are only a few simple kinds of subatomic particles. That’s much easier to code than if, say, elements were indivisible (no such thing as atom smashing) with unique properties and there were thousands or millions of different ones.
- Aggregates of people/things behaving in ways more predictable than butterfly effects might imply. (eg. is someone always winning at the stock market?)
- Quantum entanglement: does that seem suspiciously easy to code up, or suspiciously hard? Are there parent universes that might make it really easy?
- Speed of light ensures that we have very limited interactions with distant objects.
- Relativity. More suspicion because of locality. Less suspicion because time is less linear and thus harder to keep track of than if it were universal.
- Time: we only go forward through time. State gets thrown away as time progresses (we do this all the time when we code, not bothering to save intermediate states)
- Dark Matter: Maybe someone just tweaked a parameter for how heavy our galaxy was so things wouldn’t fly away!
- Religion? (If we’re in an Anthropology department’s simulation or a video game, then were Gods real, and actually just bored grad students from the parent universe?)
- Emptiness: why bother simulating so much useless empty space?
- Cosmology -- universe seems consistently to have come from a small single big bang -- no really weird shit in the hubble deep field image. This hints that we’re not a tiny accidental chemical reaction on some weather pattern or plant in the parent universe.
Section 6: Testable Hypotheses
Detecting the wallpaper -- stars placed in a mathematically predictable pattern (eg., a simple pseudorandom number generator) would be a smoking gun.
Numerical errors -- if interactions between quanta were rounded down to the nearest integer, violating conservation in some extreme case
The Grand Unified Theory turns out to have its root in discrete math.
Pixels -- are positions discretized to a stationary grid in some reference frame? Are there jaggies? (Is true diagonal motion impossible?)
High energy experiments -- do things start behaving in suspicious ways at high energies (gluon position = NaN)
Rounding errors: eg. do a quantum experiment that is supposed to turn out a certain way 1/billion times and the rare way never happens.
Quantum entanglement stops working with sufficiently large clusters of particles.
Super computers or particle accelerators above a certain size always break.
Satellites sent out of the solar system start sending signals from slightly wrong origins.
- Satellites smash into “the sky” :)
Weird things happen if you actually try to compress too much matter into the same space on earth.
Bayes: Can we use human history as a training/verification set on hypotheses? That is, how would this document have looked in the 1800’s, and what would that have led us to believe when we successfully went to the moon, discovered gluons and left the solar system?- Victorian hypothesis: We’re sitting around in 1880, and guess that the earth is simulated, and the cosmos is all wallpaper. Test: go to the moon, planets, stars. Result: yep, real moon, real planets. (Crazy time -- moon landing really was fake... because there wasn’t a moon!)
- Victorian hypothesis: We guess we’re in a big particle simulation. Test: build big microscope, look at atoms. Result: very suspicious -- everything’s discretized!