A group of scientists using machine learning “discovered” the law of universal gravity.
To do this, they trained the “graphic neural network” to simulate the dynamics of the sun, planets and large solar moons from 30 years of observation. Then they used symbolic regression to discover analytical expression for the law of strength, by default learned by the neural network.
The result, as expected, was the law of Newton's universal gravity.
Initially, the system had no idea about the masses of the sun, planets and physical permanent. They came during research.
According to scientists, this study is a key step towards realizing the potential of machine learning in order to accelerate scientific discovery.
More details can be found in https://arxiv.org/pdf/2202.02306.pdf
