A Falsificationist Account of Artificial Neural Networks