Slightly outside of my area of expertise?

Sometimes a Quanta Magazine article makes me feel like I understand something very complex and it is only when I try to explain it to someone else, or to myself, that I realize I didn't understand much at all. Here is an attempt at skipping the understanding part and jumping straight into my plan for the future of string theory.

To be clear, I can't even begin to comprehend the mathematical apparatus underlying string theory -a theory that assumes that vibrating strands of energy could explain all the elementary particles of our world, including the ones that haven't been found yet. Except it has not done that in the more than 60 years that it has been around. No testable prediction. Not even a plausible description of the existing physical world.

The theory requires the existence of tiny, complex 6-dimensional manifolds, Calabi-Yau manifolds. But there are so many such possible manifolds and each one leads to a different set of elementary particles, none of which has so far been consistent with the world we know. The mapping from the shape of a Calabi-Yau manifold to its corresponding particles and forces is an enormous computational challenge that has been partly addressed, some of it with new techniques and machine learning. But there are still more possible manifolds than atoms in the universe: even though the mapping can be computed quickly, it only helps a little. So instead of computing the mapping of individual manifolds, others are trying to "narrow down" the consideration set by determining properties shared by all consistent manifolds. The current state of the narrowing down still leaves more possibilities than atoms in the universe.

So here my super naive proposal to revolutionize string theory: treat it as an inverse problem! The objective of the inverse problem is to start from the endpoint (our universe, with its 12 particles) and work backwards from there using symbolic machine learning. I am sure no one has ever thought of that. Just kidding.