I’m pretty sure these guys are not exaggerating when they claim this is a whole new way to do science. Just last year mathematicians figured out how to calculate the tipping point of complex systems. For example, everyone knows what temperature water boils at and water beginning to boil is its tipping point, where it turns into steam. Complex systems include hurricanes and other natural phenomena for which nobody knows the tipping points. Combining new insights into Chaos theory with this kind of automated approach that can search for things like tipping points means chaos theory, fractal geometry, and even quantum mechanics will never be the same again.
Machine learning has already been applied to quantum mechanics with impressive results, and physicists are closing in on describing all of matter as we know it. For example, five years ago they created the equivalent of an elemental chart for physics that describes 500 states of matter according to quantum mechanics. They are realizing the dream of the Alchemists and this type of work will speed that up considerably. It is also related to fundamental theories, with Quantum Chaos theory (not classical) and Quantum Darwinism being two of the leading theories today that have already received their first experimental confirmations. In condensed matter physics, long range forces have proven to be every bit as important as short range ones for phase transitions, implying we are close to describing the supersymmetry of the universe.