Because artificial neural networks work nothing like human neural networks, despite the same names.
Its a false equivalence. Its no more valid a comparison than asking why "CLOS Networks" can't play Mario by themselves, despite also being a network. (A CLOS Network is one kind of network topology for switches)
The human brain does NOT take the derivative of the error function and glide down it using gradient descent. There are no matrix multiplication circuits in the brain. Things work in completely different ways that biologists and psychologists barely understand today. It has to do with chemicals, neurotransmitters and other such, which are completely alien to Comp. Sci.