I think what you're describing might just be impossible! David Hume said that when we observe a billiard ball hitting another ball and then the second ball starts moving, we didn't really witness a billiard ball causing another to move, all we saw was the first ball stopped and then the second started moving. We don't observe causation, we infer or invent it. We observe what Hume called 'constant conjunctions', i.e. putting one's hand in the flame is always associated with one's hand burning. So what science does, is they come up with a theory that posits (invents) causation and makes predictions, and if those predictions come true (molecules moving very fast would create heat and that would cause one's hand to burn etc.), then we say the theory is true.
Now maybe algos could do that too, but it would all be based on statistical patterns anyway! "All cases of putting one's hand in the flame so far have led to burning, therefore my statistical apparatus says fire 'causes' burning". The algo will necessarily generalize a small part of the world to the whole of the world. And as Bertrand Russell said, "The man who has fed the chicken every day throughout its life at last wrings its neck instead, showing that more refined views as to the uniformity of nature would have been useful to the chicken."
There is a subtle nuance here/ DL/ML techniques don't make the pattern explicit. They identify the statistic but don't identify the causation.
Without the causation part each topic down here https://www.tylervigen.com/spurious-correlations would be worthy of a nobel-prize :3
This somewhat relevant article from Guardian found its way to my inbox this morning.
That seems interesting I will bookmark it and give it a read later :3