Are We Living in a Cosmic Neural Network? A Physicist’s Bold Claim
The Real Matrix: Physicists Say Our Universe Is Likely a Neural Network
In a striking paper sent to the arXiv this summer, Vitaly Vanculin, a physics professor at the University of Minnesota Duluth, explores the possibility that the entire universe at its most fundamental level is a neural network.
For years, physicists have tried to reconcile quantum mechanics and general relativity. Quantum mechanics holds that time is universal and absolute, while general relativity maintains that time is relative and connected to the structure of space-time.
In his paper, Vanculin argues that artificial neural networks can exhibit approximate behaviors of both universal theories. He writes, “Quantum mechanics has been an extremely successful paradigm for modeling physical phenomena at various scales. It is widely believed that at the most fundamental level, the entire universe is governed by the rules of quantum mechanics and that even gravity must somehow arise from quantum mechanics.”
However, Vanculin clarifies, “We are not saying that artificial neural networks can help us analyze physical systems or discover the laws of physics. It can be seen as a proposal for a theory of everything, and as such, it should be easy to prove it wrong.”
Due to the boldness of the concept, many physicists and machine learning experts declined to comment on the record, citing skepticism about the paper’s conclusions. In a Q&A with Futurism, however, Vanculin elaborated on his thoughts.
Futurism: Your paper argues that the universe may be a neural network. How would you explain your reasoning to someone who knows little about neural networks or physics?
Vitaly Vanculin: There are two ways to answer this question. One is to start with an accurate model of a neural network and study its behavior when the number of neurons is large. I have shown that the equations of quantum mechanics describe well the behavior of systems close to equilibrium, while the equations of classical mechanics describe well the behavior of systems away from equilibrium. Is it a coincidence? Maybe, but quantum mechanics and classical mechanics are the very mechanisms of the physical world.
The second way is to start with physics. We know that quantum mechanics works well on small scales and general relativity works well on large scales. This is known as the problem of quantum gravity. But even worse, we do not know how to deal with observers. This is known as the measurement problem in quantum mechanics and the metrology problem in cosmology.
One might then argue that there are not two phenomena to be unified, but three: quantum mechanics, general relativity, and observers. This paper discusses the possibility that everything—quantum mechanics, general relativity, and macroscopic observers—emerges from the basic structure of a microscopic neural network. So far, it looks quite promising.
How did you come up with this idea?
Initially, I wanted to understand more about how deep learning works, so I wrote a paper called “Towards a Theory of Machine Learning.” The idea was to apply statistical mechanics methods to study neural network behavior. It turned out that, in some limits, the learning dynamics of neural networks were very similar to the quantum dynamics found in physics. While on sabbatical, I decided to explore the idea that the physical world is actually a neural network. This idea is certainly crazy, but is it really crazy? That remains to be seen.
In your paper, you write that to prove the theory wrong, we need to “find a physical phenomenon that cannot be described by a neural network.” What do you mean by that? Why is such a thing “easier said than done”?
There are many “theories of everything,” most of which must be wrong. My theory is that everything around us is a neural network, and to prove it wrong, we need to find phenomena that cannot be modeled by a neural network. However, we know very little about how neural networks behave and how machine learning works. That is why I set out to develop a theory of machine learning.
How does your research relate to quantum mechanics?
There are two main schools of quantum mechanics: the Everett (or many-worlds) interpretation and the Bohm (or hidden variables) interpretation. I am not going to say anything new about the many-worlds interpretation, but I think I can contribute to the theory of hidden variables. The hidden variables in emergent quantum mechanics that I have been studying are the states of individual neurons, and the trainable variables (such as bias vectors and weight matrices) are quantum variables. Note that hidden variables are very nonlocal and may violate Bell’s inequality. Approximate spatio-temporal locality is assumed, but the system need not be local because every neuron can be connected to every other neuron.
Can you explain how this idea connects to natural selection?
What I am saying is simple. There are more stable structures (or subnetworks) in microscopic neural networks, and there are less stable structures. The more stable structures survive evolution; the less stable structures perish. At the smallest scale, I predict that natural selection will produce structures of very low complexity, such as chains of neurons, but at larger scales, structures will become more sophisticated. Thus, it is my contention that everything around us (particles, atoms, cells, observers, etc.) is the result of natural selection.
The idea is certainly crazy, but is it really crazy? That remains to be seen.
Source: Are We Living in a Cosmic Neural Network? A Physicist’s Bold Claim
Strange Motion of Neutrons Proves Nature Is Fundamentally Bizarre
Strange Motion of Neutrons Proves Nature Is Fundamentally Bizarre