[landon32]

Right now we create feed forward neural networks that have a very specific architecture so that we can easily train them. If we want a really "general purpose" sufficiently complex neural network we would need it to use way, way, way more computation power and take up many 100s of times as much data. It would be insane. So instead of having an amorphous blob of artificial neurons, it's much easier to design clever components and assemble them in a general purpose way.