Then "learning approaches" camp (which BTW is also an academic camps and somehow able to do science) cannot prove that knowledge embedded in practices is actual

Then "learning approaches" camp (which BTW is also an academic camps and somehow able to do science) cannot prove that knowledge embedded in practices is actually being learned at all by their machines, if one rejects legibility. It's not even non-legible: it's self contradictory, i.e., unintelligible (which means a different thing than merely legible).