It's true that those two approaches are "a bit dumb", but the statement > The more sensible view it seems to me is to compare neural computation to what we can

It's true that those two approaches are "a bit dumb", but the statement > The more sensible view it seems to me is to compare neural computation to what we can do with silicon and then project on a Moore's law type assumption as to when it would be likely is no less dumb imo. What makes you say this? What's the evidence? What neural network system has achieved something even close to being on a path to AI?