Modeling is a new form of pseudo-scientific speculation. ML is closer to reality as long as the training sets are not a dogmatically interpreted noise, like it usually is in finance and other pseudo-sciences.

You could read about a fundamental difference between a properly controlled, replicable scientific experiment and computer simulation according to some abstract/unverified model and why results of such simulations cannot be substituted for experimental results or any form of evidence in my older comments.