The one argument I rarely see against automated analysis is that any analysis has a false positive rate. If you hire 1,000 humans to do that analysis there's simply a limit to how many errors they can make in total.

If you introduce automation doing the same thing at a scale corresponding to 10,000× the human capacity, you will have 10,000× as many errors. This will violate Blackstone's ratio and erode trust in the system.