Problem is that is too indirect. Imagine trying to do that intricate work on sight alone. Like trying to do stuff with numb hands. Without tactile feedback, we are useless in a lot of these tasks. But the tactile feedback is not instrumented. To the extent that machine learning works, it needs to have all the inputs and outputs instrumented, and feeling is left out.

We can see something done and have an explanation and internalize it and then 'get a feel for it' later. But AI doesn't currently feature this ability to really translate that sort of instruction to an entirely different set of sensation.