The Real Matrix: Physicist Says Our Universe Is Likely a Neural Network

We don’t encounter writings that seek to change reality very often.

Vitaly Vanchurin, a physics professor at the University of Minnesota Duluth, seeks to reframe reality in a particularly eye-opening way in a preprint sent to arXiv this summer. He suggests that humans are part of a vast brain network that controls everything in our environment. In other words, it’s “possible that the entire cosmos on its most fundamental level is a neural network,” he stated in the report.

Physicists have been working to make quantum physics and general relativity compatible for a long time. The former contends that time is relative and connected to the structure of space-time, whereas the latter asserts that time is universal and absolute.

In his paper, Vanchurin argues that artificial neural networks can “exhibit approximate behaviours” of both universal theories. Since quantum mechanics “is a remarkably successful paradigm for modeling physical phenomena on a wide range of scales,” he writes, “it is widely believed that on the most fundamental level the entire universe is governed by the rules of quantum mechanics and even gravity should somehow emerge from it.”

“We are not just saying that the artificial neural networks can be useful for analyzing physical systems or for discovering physical laws, we are saying that this is how the world around us actually works,” reads the paper’s discussion. “With this respect it could be considered as a proposal for the theory of everything, and as such it should be easy to prove it wrong.”

Most of the physicists and machine learning specialists we contacted declined to comment on the topic on the record due to their skepticism about the paper’s findings. However, Vanchurin dug into the debate and expanded on his proposal in a Q&A with Futurism.

Futurism: According to your paper, the cosmos could be basically a neural network. How would you justify your conclusions to someone who was unfamiliar with physics or neural networks?

Theodore Vanchurin: Your question might be answered in two different ways.

The first method is to begin with an accurate model of a neural network before examining how the network behaves when there are many more neurons present. What I’ve demonstrated is that quantum mechanical equations very well capture the behavior of a system close to equilibrium, whereas classical mechanical equations accurately capture the behavior of a system farther from equilibrium. Coincidence? Perhaps, but as far as we are aware, the physical world operates according to quantum and classical mechanics.

Starting with physics is the second approach. We are aware that general relativity operates rather effectively at vast sizes and that quantum mechanics operates fairly well at tiny scales, but we have not yet been able to integrate the two theories. This is referred to as the quantum gravity issue. We obviously have a significant gap in our knowledge, and to make matters worse, we have no idea how to deal with observers. In terms of quantum mechanics and cosmology, this is referred to as the measurement issue, respectively.

Then, one may contend that quantum physics, general relativity, and observers are the three phenomena that require unification, not just two. Most scientists (99%) would agree that quantum mechanics is the fundamental theory and that all else should somehow flow from it, but no one is certain how it can be accomplished. In this study, I explore a different hypothesis: that everything, including quantum physics, general relativity, and macroscopic observers, arises from a tiny neural network that serves as the basic structure. Things seem fairly decent thus far.

Who or what initially thought of this?

I first produced a paper titled “Towards a theory of machine learning” in order to simply understand deep learning better. The initial plan was to analyze neural network activity using statistical mechanics techniques, but it turned out that, within certain bounds, neural network learning (or training) dynamics are extremely comparable to the quantum dynamics we observe in physics. I chose to investigate the notion that the physical universe is essentially a neural network at the time since I was (and still am) on sabbatical leave. The notion is undoubtedly absurd, but is it truly absurd enough to be true? That is still up in the air.

In the paper you wrote that to prove the theory was wrong, “all that is needed is to find a physical phenomenon which cannot be described by neural networks.” What do you mean by that? Why is such a thing “easier said than done?”

There are a lot of “theories of everything,” but the majority of them must be false. According to my hypothesis, everything you see around you is a neural network; thus, all that is required to disprove it is the discovery of a phenomena that cannot be explained by a neural network. However, if you stop to think about it, it is a really challenging undertaking, especially given how little we truly understand about how neural networks operate and how machine learning functions. That’s why I initially sought to construct a theory of machine learning.

The notion is undoubtedly absurd, but is it truly absurd enough to be true? That is still up in the air.

Does your study take the observer effect into account, and how does it relate to quantum mechanics?

Could you please clarify the relationship between this concept and natural selection? The evolution of intricate biological cells and structures is influenced by natural selection.

I’m saying something really simple. The microscopic neural network is composed of both more and less stable components (or subnetworks). Evolution would favor the more robust structures, causing the less robust ones to disappear. I anticipate that natural selection will produce extremely simple structures at the lowest sizes, such as chains of neurons, but increasingly complicated structures at larger scales. The claim is that everything we see around us (particles, atoms, cells, observers, etc.) is the outcome of natural selection since I see no reason why this process should be restricted to a certain length scale.