This question has come in for a surprising amount of attention of late. For instance, take a look at this article from the Guardian. Or try this from the New Yorker. Or PBS, or the BBC, or Scientific American. It’s everywhere. Why so much attention to this question? Because Elon Musk believes it and therefore it is interesting to the world at large.
He is wrong. He is drawing attention to an incredibly important scientific issue. I fully endorse his enthusiasm for the topic, and I agree with him all the way about Mars. He has done wonders to raise awareness about electric vehicles, sustainability, and humanity’s role in the universe. But on this topic, he is wrong. So wrong.
Why am I confident that I know? Because nobody on this entire planet knows as much about simulating the fundamentals of the universe using algorithms as me. And it hurts.
How can I justify such a ludicrous remark? After all, aren’t luminaries such as Richard Terrile, Max Tegmark and Nick Bostrom all weighing in on this subject with incredibly weighty things to say? Do I know anything compared to them? Me, a lowly science fiction writer, part time software hack, and general-purpose babbler of random nonsense?
Yes. I know more. Horribly more.
By the end of this post, I aim to convince you of three things, firstly, that I am the global expert on simulated reality, secondly, that Elon Musk is wrong, and thirdly, that the money and attention associated with this subject should be properly invested not into navel gazing, but into warp drive research. Yes, warp drive. (As in Star Trek, as in faster than light. As in, that’s not real physics is it? I thought all that was impossible!)
But first, let me take a very small moment to scream. There, done it. Why am I screaming? Because this topic reflects on a subject that seized control of my mind and has been shaping my thoughts and actions for the last twenty years. And now that it finally has a public airing, it’s coming out all wrong.
Because there is not one question in here, but two.
It’s not whether we’re in a simulation, but 1) whether the universe is like a simulation. And 2) if it is like one, is anyone running it?
The answer to the first question is yes. And that’s why there’s attention in Silicon Valley right now.
Something has slowly been happening behind the scenes in the world of physics research. Confidence in string theory as the picture of reality is breaking down. The math holds up just fine, but it doggedly refuses to have anything to do with experimental results. Quietly, the options for supersymmetry, upon which string theory depends, are being ruled out by experiments at places like the LHC.
Nobody likes this, but there it is. String theory lives on as a handy theoretical framework for cosmologists, but not as something you’re supposed to take literally any more. It is dying a quiet, graceful death, while everyone in the field continues to deny that fact because their funding depends on it. What is left in the wake of that vision is confusion.
At the same time, folks in Silicon Valley and elsewhere have noticed that it’s getting easier and easier to simulate aspects of our physical reality. Spookily easy, in fact. And they have both the vision and the tools to be able to estimate when the day will come when we won’t be able to tell the difference.
If it’s that easy to simulate reality, they say, then how the hell do we know it’s not already happening? What if there’s some exterior universe out there with better kit than ours who’s just doing it?
Good question. I’m so glad that you all are finally starting to think about it.
For me, this question kicked on hard in 1997. I found myself reading the Emperor’s New Mind by Roger Penrose, in which he attempts to assert that human consciousness cannot be modeled on computers. He claimed that the processes that consciousness requires in physical reality, like quantum mechanics, weren’t computable. They were infinite. And so consciousness was too.
As an ex AI researcher, I found his position profoundly annoying. But as I read his book, I realized that he wasn’t just wrong. He was exactly wrong. Every supposedly infinite process he outlined was, in fact, something I could emulate on a machine. Furthermore, I could actually prove it.
And, like an idiot, I spent the next twenty years doing that. Not in a university department, though my crazy adventure has taken me to national research labs in Italy and the Institute for Advanced Study, but on my own time. As a hobby. TWENTY YEARS! What in god’s name was wrong with me? I have no idea.
Did I attempt to recreate reality? No. Was I looking for a unified theory of everything? Definitely not! That wasn’t the point. The point was to look at every core physical symmetry in turn, and show that it could be made computable.
Quantum mechanics and the superposition of states? Done it. Can run it on my Mac. I can watch the double slit experiment in real time.
Special relativity? Covered. Piece of cake. Want to check out my Youtube videos?
General relativity? Done. Spatial distortion, temporal distortion, you name it. I have it sorted.
The group structure of particle families? Yep. Nailed that one.
The Big Bang, and a background independent model of spacetime? Absolutely. Took a while, but was very satisfying to figure out.
Bell’s Inequality? Solved. Admittedly, didn’t finish the code because I have a kid, but I know how to reproduce it.
Now, let’s be clear about what I achieved here. Did I solve the mysteries of the universe? No. Are the tools I developed useful to physicists? I doubt it. Do I expect my theoretical solutions to reflect actual reality? Not for a minute. Was this a work of staggering genius? Hell no. It was probably a phenomenal waste of time.
So why do I even imagine that I achieved all this? How can I be so confident? Because I spent a ridiculous amount of time hanging out with physicists, in significant part by being married to one. Which is why I ended up as a complex systems theorist at Princeton before I started writing full time. Since becoming infected with this mental disease, I have chased down the implications at UC Santa Cruz, UC Berkeley, Princeton, Cambridge, and JPL, talking it over with anyone patient and tolerant enough to answer my stupid questions.
I mimicked all these symmetries of nature, not with math, but with code. And I’m confident that I got it working, because I duplicated the necessary experiments in simulations, and then collected the data.
My code is slow. It’s not elegant. It’s weird looking. I don’t expect anyone else to like it. But by Crikey, it has enabled me to do some fascinating things. Most of the stuff I’ve written has never made it into science papers because it’s a hobby and I have a life to lead, but it’s still there, calling to me, silently shrieking.
So yes, I’m all the expert in this one trivial, psychotic corner of science. On my tiny palm-tree island, I reign supreme, coconut in hand.
“But what has this to do with Elon Musk?” I hear you ask. “If you’ve actually proved that the universe can be simulated, doesn’t that speak in favor of the simulation hypothesis?”
No. No. And a thousand times no.
What I learned on my adventure is that Nature is computable. But what I also learned was that the class of algorithms you want to do that job are not nice, or cheap, or much like human software. They involve networks. Lots and lots of networks. And crazy iterative functions. And sets. So many sets! If there is someone simulating the universe, they are even crazier than I am.
An easy way to think about it is this: we live in a universe that appears to follow physical laws. So why shouldn’t it be simulated? Fine, but then you’re proposing a universe outside it that runs on physical laws too. Because if there weren’t consistent physical laws, they wouldn’t be able to make those simulations.
So then that universe is also something you can simulate. So maybe there’s someone sitting outside that. And so on. Forever. Assert that the universe is being simulated once, and you have to allow for a possible infinite number of levels.
Sure, you say. Why not? But here’s the thing. Every time you bump up a level of simulation, you need a faster, bigger, smarter universe in which computation is that bit easier. Yet, conversely, in each universe, you need your engineers to tinker with the code less.
Because that’s what simulation engineers do. They screw around with their code! They restart it with different parameters. They poke their fingers in. Every day that you aren’t bombarded with styrofoam elephants or teleported to a planet made of cheese is another day that a potentially infinite number of software engineers have decided not to play with the awesome thing that they built. Do I buy that? Not for a frickin’ instant. I have played God successfully, and the longest I can leave something alone is about eight hours.
But there’s another, deeper reason to be skeptical here. And that’s because exactly the same minimalist logic that makes discrete models of the universe scientifically relevant points away from the simulation hypothesis.
Discrete, algorithmic models of the universe kick ass because they produce results that align with experimental reality with the least amount of complexity. No infinities. No paradoxes. No dangling super-particle partners. No Godel universes. No mess. To then propose another layer of universe on top of that achieves exactly the opposite result. It adds complexity. Without evidence! Which if you’re doing real science, is generally frowned upon.
“But if the universe is like computer code,” you may ask, “who’s running it?”
NOBODY! DO YOU HEAR ME? NO ONE! Does the idea of a smooth, fluid universe require that it be run on a machine made of GIANT VALVES? Does the idea of universal equations imply a school of blind monks in hyperspace tracing out the symbols on a piece of magic felt while they keen out the song of infinity in their reedy voices? No. The universe is just there.
Discrete models of the universe are great because you can model all of Nature as a mathematical series rather like the Fibonacci numbers. If math exists, so does physics. Hardware is not required.
But there’s an added bonus to discrete physics. Once you’ve built a working simulation of a physical symmetry, you can play with it. You can tinker, just like the gods of hyperspace don’t. You can make those models do things that Nature never intended. You can, if you like, warp space.
Not assuming a flat Lorentzian manifold for spacetime and putting in the effort to build something that looks like it out of pointers instead opens the doors to all kinds of possible new physics. Suddenly, Nature becomes something you might be able to hack.
While your picture of reality is obscured by a set of differential equations that require perfectly serene behavior at all scales, you can’t see that. Which is why most classically trained physicists pooh-pooh the idea. However, not only are discrete models of the universe infinitely more likely to be correct, they also leave the door to awesomeness wide open.
Why don’t more people care? Why haven’t people looked into this more deeply? Your guess is as good as mine. It makes me want to tear my hair out. Because it’s a lot of work, I guess. Classical physical models take you further faster, even though they may hide all the cool stuff. And there’s no funding for discrete approaches because they’re risky. Who knows what’s the right implementation model to look for?
So here we go, Elon Musk. Here’s my plea. If we’re ready to take the time and the effort to contemplate the simulation hypothesis, let’s go one better. Let’s give humanity the stars instead. Yes, Elon, please get people interested. But don’t try breaking out of the Matrix, bend it to your will instead like Neo. Do physics-kung-fu, and I will do everything in my power to help you.
(My first novel Roboteer has its US debut this month on October 18th. If you know someone of the American persuasion who loves intelligent, action-packed SF, let them know!)
Oh Alex, I miss working with you! Yes, I read the entire blog and followed it . . . well, mostly. I hope Elon Musk listens. Congrats on your book’s U.S. debut!
Cool post, Alex! Just to be contrary:
“Yet, conversely, in each universe, you need your engineers to tinker with the code less.
Because that’s what simulation engineers do. They screw around with their code! They restart it with different parameters. They poke their fingers in.”
The time scale may be quite a bit off!
Maybe any billion of our years now someone will wake up from their overnight simulation and finally be excited by their results.
Or maybe all this has taken just a few meta-minutes and someone’s staring at a visualization and hasn’t noticed anything interesting happening over in our local supercluster and humanity (or some intelligent life somewhere) had better figure out a way to arrange galaxy clusters into a smiley face before the meta-coder hits ctrl-c, or to (along the breaking-out lines) overflow a buffer and start taking over the computer.
On other matters, I do quite like your turtles all the way down critique!
One consideration I would throw in there is that if the ultra-programmers do in fact fiddle with the code, change rules and initial conditions, even (carefully) tweak things along the timeline, we as entities within the system are not in general going to perceive such changes subjectively. We are part of the (modified) system, so our personal experience and memory will be consistent with the world as modified. Unlike in many unacceptable Hollywood takes on time travel, if a timeline is changed, no one in the universe in question sees people fade out, buildings change architecture etc.
The only way a ‘fiddle’ would be perceived directly by in-system entities would be if it explicitly abrogated the normal laws of the system, in a place, time, and at a scale that is observable by the in-system entities in question.
Why would the Ur-scientists running this simulation do such a thing? It would seem to reduce or eliminate the scientific usefulness of the simulation being run.
One could make the argument that the Makers In Charge would want to go to great lengths to avoid providing inhabitants of the simulation definitive proof of the facts of the situation, unless and until they decide to “break the 4th wall” and explicitly show their hand to their creations as some sort of final chapter to the exercise.
Given the 13+ billion years of subjective existence we seem to have experienced in this supposed simulation, the probability of experiencing such a one-time event within a human lifespan seems small.