AI in the Middle of a Big Paradigm Shift in Physics

paradigm shift in physics

paradigm shift in physics

Ladies and gentlemen, this is big.

But, let’ start from the beginning.

What is physics? A natural science.

How does physics operate? Well, there are observations first. Then, we/you/somebody writes down a mathematical formula that is supposed to align with the observed results. Finally, somebody else tests it.

Is there a final mathematical formula that ultimately fits with the observations? We don’t know. At first: we don’t know how accurate our observed results are, and at second: we cannot guarantee that with more precise observation, our formulas will still.

The most relevant to talk about this are astrophysicists and quantum mechanics.

Example of the extreme reasoning from the first ones:

  • Aha, galaxies can’t exist with the currently measured mass! Ok, so we just discovered a dark matter.
  • Aha, the universe is expanding at a higher rate than calculated by us? Ok then, so we just “discovered’ the dark energy.

That was easy. Talking about quantum mechanics, don’t even try to speak with those guys about Schroedinger cat, spooky action at a distance, and the tunneling effect. They will freak out simply because even they don’t understand all of these things. They are notoriously trying to formulate physical laws in terms of mathematical formulas, with calculations that fit into observed measurements. Then, they are notoriously and unsuccessfully trying to explain to the ordinary people what do their formulas tell. Nothing more.

Now, let’s talk about Neural Networks!.

A wide-known Universal Approximation Theorem states that every function can be interpolated with any wanted precision. All we need are sufficient neurons in a hidden layer.

How this relates to physics?

How physics laws begin: with a given set of observations.

What is in the middle: a set of mathematical formulas that calculates outputs on a given input.

At the end of the pipeline: predictions, i.e., calculations from the mathematical formulas provided by some specific physic law.

Now, how about replacing the middle “math” layer with an AI model?

As said above, there is for sure (i.e., mathematically proven) a model that can “model” the reality expressed with the measurements.

The cherry on top of this is the fact that building such models is much less time demanding than building old-fashion mathematical models.

I remember one summer day several years ago on a Data Science meeting in Skopje, Macedonia, where one of the presenters said he had worked on software that executes physical experiments for quantum physics. How: using ML models. Why: because the software runs 1% of the time needed for an actual physical experiment and provides an accuracy of 99% of the “real” experiments’ accuracy, which is quite good for their purposes.

These days I found the following article:

AI Humiliates Spacetime Empire. Cracks In The Opaque Glasshouse Of… | by Marcus van der Erve | SoCyc | Feb, 2021 | Medium

Worthless to paraphrase or quote the author, the article is quite good enough to read it all.

Another cool article on the topic:

New machine learning theory raises questions about nature of science (

“A novel computer algorithm, or set of rules, that accurately predicts the orbits of planets in the solar system could be adapted to better predict and control the behavior of the plasma that fuels fusion facilities designed to harvest on Earth the fusion energy that powers the sun and stars.”

And guess what: the computer algorithm mentioned is yet-another-ml-model.

Welcome to 2021. Enjoy your weekend.

Deep Learning

AIdata sciencephysics

Leave a Reply

Your email address will not be published. Required fields are marked *

15 − 5 =