The field is evolving quickly and it's pretty to see. Another deeper reason to me is that it employs some neural network techniques -which have differentiabili

The field is evolving quickly and it's pretty to see. Another deeper reason to me is that it employs some neural network techniques -which have differentiability-. This means that you can backpropagate errors from a domain to another. In the future, we'll have videos as embeddings that can understand written signs, and interpret the voices and what they refer to in the scene. This is not AGI, but will be much closer to the sensor fusion us humans do.