The Universe Might Be a Self-Learning Computer. Here's What That Means

It’s a wild solution to an intractable problem.

The Universe Might Be a Self-Learning Computer. Here's What That Means
A theory of everything would have to combine the choppy quantum world with the smooth curves of relativity... or would it? Michael Dziedzic / Unsplash

One problem has proven impossible for the brightest minds in physics.

Albert Einstein and Stephen Hawking – the most famous physicists of the twentieth century — both spent decades trying to find a single law that could explain how the world works on the scale of the atom and on the scale of galaxies. In short, the Standard Model describes the physics of the very small. General relativity describes the physics of the very large. The problem? The two theories tell different stories about the fundamental nature of reality. Einstein described the problem nearly a century ago in his 1923 Nobel lecture, telling the audience that a physicist who searches for, “an integrated theory cannot rest content with the assumption that there exist two distinct fields totally independent of each other by their nature.” Even while on his deathbed, Einstein worked on a way to unite all the laws of physics under one unifying theory.

Hawking eventually gave up.

Now, Stephon Alexander, a professor of physics at Brown University, is taking a stab at the challenge. In a preprint that hasn’t yet been peer-reviewed, Alexander and several collaborators, including technologist Jaron Lanier and physicist Lee Smolin, put a slightly different spin on the problem. Instead of focusing on what the laws of physics are, they’re asking why the physical world is governed by certain laws and not by others. They write that while physicists haven’t “finished that task” of discovering the laws of physics, “we seem to know enough to take a few steps towards answering a deeper question.”

The standard model and general relativity actually have a lot in common

What physicists do know is that two theories — the Standard Model and Einstein’s General Theory of Relativity — offer powerful, empirically sound explanations of physics at the scales they are designed to explain. It turns out that the theories share some fundamental mathematical qualities, too.


According to Alexander, both are based on gauge theories and symmetry principles, which use math to describe how objects can move and interact. Researchers working in the tradition of string theory have relied on these similarities in their efforts to unify the two theories by reimagining some particles as one-dimensional objects called strings. But there's a catch. That approach results in “a vast richness of laws" that are mathematically possible. That "multiverse of theories,” as Alexander calls it, includes the standard model and general relativity — and a lot more theories that apparently could describe our physical world, though they exist on a scale too small to be tested.

According to string theory, physical features, like the mass or electric charge of a product, are the result of how the string vibrates in many dimensions, including many hypothetical dimensions that exist on a very small scale, Brain Greene wrote in Smithsonian. Determining the shape of those dimensions would be key to understanding how it is that those strings constitute physical reality, but the math doesn’t offer a clear answer. While early string theorists identified a handful of possible shapes of those tiny extra dimensions, the list of mathematically possible shapes grew into the millions, then into the billions, and eventually into “numbers so large that they’ve never been named,” according to Greene.


“[S]tring theory doesn’t answer the why question,” Alexander told New Scientist in September. “It lacks a mechanism to select which of the slot machine of 10500 possible universes is our universe.”

His idea is that the physical world as we know it today isn’t the one true reality, it’s the result of many iterations of the universe trying an arrangement of laws that didn’t work. In Alexander’s telling, our universe probably has taken on the characteristics of many possible universes. It eventually “found itself in a configuration… that was stable” and allowed it to “build-out in a consistent way.”

He compares this process of trying, failing, and trying again to playing an arcade game with a very big bag of quarters. “If you have infinite [lives], you play, you die, you play, you continue playing, you die, but you get to continue playing, right? I think that's kind of like the idea,” he said.


The universe is always able to “continue trying.” In this respect, the universe is 'learning' what works and doesn't as it evolves. Since the universe has no teacher but is learning its lessons as it goes, the researchers call it “autodidactic.”

The answer could be a primordial “meta-law”

How could this be possible? Alexander’s answer is elegant: a “meta-law” that existed long before the laws of physics that we know as the Standard Model and general relativity. It’s this meta-law that contains the ability to try things and learn. Confusingly, the meta-law is the universe itself, at least in some respects.

“The thing that's weird here is that the hardware is the software and the software is the hardware,” Alexander explains. General relativity and the Standard Model emerged later, once the universe found the stability it was looking for. He compares the idea to Darwin’s theory of evolution.


“In biology, there used to be a ‘Why these species?’ problem: explain why dogs and cats exist while unicorns and werewolves do not,” the researchers wrote in the pre-print. They explain that Darwin introduced a handful of principles governing life in general that makes it possible to understand how any specific species came to be. For instance, the idea that species emerge because individuals that are well-suited to their environment have a better chance of passing valuable traits to their offspring. Finding that first glimpse of those underlying principles was a gargantuan achievement, but it wasn’t the end of the story. Researchers have spent 160 years filling in details, and they still aren’t finished.

Alexander and his colleagues aren’t claiming to have discovered the physics equivalent of evolution. They describe their contribution as “tiny, baby steps” toward a complete theory. For Alexander, it was important to propose a formal version of the theory in order to see if it holds up to the scrutiny of other theoretical physicists and the empirical work of experimentalists.


“We have to commit ourselves to something so that we can play and we can try to do some calculations and explore the idea," he said.

How would an autodidactic universe work?

The idea that the universe has evolved according to some other rules isn’t totally new. The philosopher Charles Sanders Pierce applied the principles of natural selection to cosmology in 1893, less than four decades after On the Origin of Species was published. Alexander and his colleagues were inspired by advances in theoretical physics, computer science, and philosophy of science to make the much bolder claim that the universe is learning its laws, not merely evolving to achieve greater fitness in its environment.

The argument rests on three fundamental elements: matrix models, quantum gauge theories, and learning machines.

The team began with the insight that theories of physics could be understood as matrix models. That is, the theories could be written as equations or they could be substantiated as tables of numbers with perhaps billions of columns and rows. Such a vast matrix could contain all possible laws that could govern the universe. Insights from computer science light the way from there.


“The mathematics of matrix theory seems to have some of the ingredients of a particular type of neural network,” Alexander said in a New Scientist interview.

"Maybe there's some input and some output, and the universe is adjusting the weights such that [it] ends up learning the standard model… and gravity,” he told Interesting Engineering. “That's the basic idea.” Here, “weights” refers to the specific mathematical relationships that determine how the inputs that go into a neural network are transformed to generate its outputs.

If Alexander is right, we humans almost definitely aren’t in a position to see everything the universe has learned. “There may be other corners of the universe where other interesting things happened that are completely radical from our perspective,” he said, pointing out that there’s no reason to think the universe had a preference to learn laws that have allowed for life and consciousness to emerge.

“As far as I'm concerned, it was a part of the evolution,” he said. “We're the ones that claim that that's great.”

Confirmation of the autodidactic theory of the universe wouldn’t necessarily rule out the possibility of a theory of everything as Einstein or Hawking might have imagined it, but it would definitely underscore Alexander’s belief that physicists should be exploring ideas that go well beyond the confines of traditional boundaries. The answers to these questions may not lie in what we currently think of as “physics.”

Greene says that directly observing strings would require a particle accelerator “the size of the galaxy.” Alexander’s hope, which he details in his new book, Fear of a Black Universe: An outsider’s guide to the future of physics, is more modest — but far from guaranteed. “Experts should engage in exploration with people in other fields.”

Follow Us on


Stay ahead with the latest science, technology and innovation news, for free:

By subscribing, you agree to our Terms of Use and Privacy Policy. You may unsubscribe at any time.