You forgot to mention pure inductive learning by definition cannot do causal reasoning (abductive learning). Second, the is-ought distinction suggest another huge area where inductive ML--automated or not--will never succeed. I discuss these issues here in case you're interested. https://medium.com/swlh/algorithmic-injustice-and-the-fact-value-distinction-in-philosophy-e35355ca5a23?source=friends_link&sk=b7a467196039c4ba9c4d3e66e943f2d3