Will the next Einstein be a machine? A number of recent studies show this scenario may not be that implausible after all (thread) 
In "Discovering physical concepts with neural networks” @eth researchers presented a neural network architecture that can discover physical concepts from experimental data without being provided with additional prior knowledge
Their network recognised the number of deg. of freedom describing a simple quantum mechanical system. Not impressed? Given a time series of the positions of the Sun and Mars as observed from Earth, it discovered the heliocentric model of the solar system arxiv.org/abs/1807.10300
In "Toward an AI Physicist for Unsupervised Learning” @tegmark and Tailin Wu implemented common physics strategies (divide-and-conquer, Occam’s razor, unification and lifelong learning) in a “learning agent”
Their “AI Physicist” correctly identified the laws of motion of different environments from unsupervised observation of “worlds" involving combinations of gravity, electromagnetism, harmonic motion and elastic bounces arxiv.org/abs/1810.10525
In case your reaction is “Well, can they discover Einstein’s theory of general relativity?”
More seriously, these pioneering studies demonstrate the enormous potential of AI in science. Will it ever surpass replace human scientists? Nobody knows, but we’d better embrace these methods
At the very least they can help us become better physicists. And, who knows, maybe even better human beings
If you are interested in the interplay between Physics and Machine Learning, check out @dark_machines and darkmachines.org
