What's unreasonable about neural networks (in general, not just recurrent ones) is that we don't really have any theoretical understanding of why they work. In

What's unreasonable about neural networks (in general, not just recurrent ones) is that we don't really have any theoretical understanding of why they work. In fact, we don't even really understand what sorts of functions neural networks compute.