On Scientific Representations