Unless we start thinking about fundamentally different ways of solving AI, we will always be 10 years away. It's puzzling that no one wants to think beyond backpropagation