[jfoster]

There is a limit to the current method. The generalised AI you described likely is quite a while away. It's also economically questionable, now that many of the specialised AIs can be done very affordably.

For example, if you want to teach a robot to flip burgers, you could invest a few trillion dollars into generalised AI and wait many years for an uncertain outcome, or you could create training sets for a neutral network and be done in a few months for a few million dollars. The reason we know we have made advancement is that until a few years ago, the problem of training that robot did not seem to be within reach. Today it mightn't be completely easy, but most people would agree that it's quite achievable.