A representation of dog, for example, can be realised in a variety of different (though similar) node-level activation patterns.
Thus, no single node-level description can fully describe the dog representation.
Moreover, node-level discrepancies make it possible for the dog representation to produce a variety of different behaviours.