Artificial General Intelligence (AGI) should be able to see. In particular, it should be able to recognize objects and to learn to recognize new classes of objects from as few examples as possible.
This means that it should generalize. How would that work?
