Artificial Intelligence Aids in Uncovering the Mysteries of Physics

It could one day help to reformulate quantum mechanics. Swiss physicist Rennato Renner and his team have proposed an algorithm that can distil vast amounts of information into simple rules, thereby mimicking the ways in which physicists arrive at new equations.

This involves two neural subnetworks that have undergone a "lobotomy" (they are connected to each other through only a few connections). The first behaves like a classical neural network, but due to the reduced number of connections, it is forced to transmit the learned information in a highly condensed form.

One of the initial tests was conducted on simulated data of the movements of Mars and the Sun in the sky. The algorithm produced equations reminiscent of those proposed by Nicolaus Copernicus at the beginning of the 16th century. In other words, it concluded that the Earth orbits the Sun! 🙂

Could physicists collaborate with artificial intelligence in their research in a few years' time?

https://www.nature.com/articles/d41586-019-03332-7