Comment by D-Machine

What theory says this? Surely not a proven or applicable one, as this post makes it very clear a great deal of understanding is in fact possible, even with highly complex neural networks.