Because of generalization. Updating weights allows the model to internalize patterns and apply what it learned to novel situations rather than just retrieving and reusing stored information