I am not arguing against them having responsibility. They do. I am arguing that they might not have realized what the weights are doing. Because, yes, indeed, large neural networks ARE black boxes.