Originally Posted by AdultKing

The work won't be done by creating a 3D model and then animating it.

The work will be done by filming a real person, then using deep learning to handle the animation, speech and interaction. Relatively speaking there are simple applications of this now in mainstream, such as an AI driven virtual newsreader.

As massively multicore multiprocessor computing power becomes more affordable these things will be easier to do at scale and at lower cost.