The general mechanisms for neural networks go beyond being mere replicants of training data ("copy machines").

It has been proven mathematically that neural networks can approximate any computable function to an arbitrary degree of closeness.

If you choose to believe that Intent and conscience are computable functions (as I do---they have to be because they run on finite hardware (i.e, the brain)), then there is no reason that we cannot develop a neural network that behaves with intent and conscience.

The exact mechanism to achieve this will be developed sooner or later, but we need to be ready for it.