There is strong evidence that deep learning methods can quantify properties of complex, flowing, and/or chaotic systems, including turbulence and gas-surafce interactions, better than the state-of-the-art. This opens new possibilities in which machine learning supports, or even unlocks, scientific discovery.