IMHO, this article makes grand claims but doesn't substantiate them. In what way is ML-based biology any different from the myriad statistics-based mechanistic

IMHO, this article makes grand claims but doesn't substantiate them. In what way is ML-based biology any different from the myriad statistics-based mechanistic models that systems or computational biology has employed for 50 years to model biological mechanisms and processes? Does the author claim that theory-less parameterless ML models like those in deep NNs are superior because theory-based explicitly parameterized models are doomed to fail? If so, then some specific examples / illustrations would go a long way toward making your case.