Or someone who understands that it's not magic, and there's still humans choosing the dataset, labelling the data and picking the loss function. This is exactly the kind of gaslighting GP was talking about.
Just because there are more edge cases where the human has no idea what their algorithm does, doesn't absolve them of responsibility. You don't get to go free after running a bunch of pedestrians over by claiming you were too drunk to know where the road was or which direction your car would turn when you moved the wheel. If I put a running metal lathe in a kindergarten, I don't get to throw my hands up and say 'you clearly don't understand machining' when some children get dismembered.
Just because there are more edge cases where the human has no idea what their algorithm does, doesn't absolve them of responsibility. You don't get to go free after running a bunch of pedestrians over by claiming you were too drunk to know where the road was or which direction your car would turn when you moved the wheel. If I put a running metal lathe in a kindergarten, I don't get to throw my hands up and say 'you clearly don't understand machining' when some children get dismembered.