Artificial Intelligence (AI) is a science which does not quite fit in with other categories of sciences. Although it can be compared with mathematics, physical sciences, and cognitive or behavioural psychology, each of these comparisons leads to different difficulties. There is therefore a case for considering AI as a protoscience in search of a general framework of description which will allow subsequent abstractions or general theories. This paper argues that two of the abstractions, which should be present in any future description, are heuristics and models.