Excerpt: So called “Second-Order” cybernetics developed the notions of self-organization in a manner more radical than Wiener, attempting to both give a strong sense to the prefix “auto” and, at the same time, to explain reflexive phenomena, memory, and the relation between the organism and its history, in terms of the emergence of a global effect out of local interactions and the networking of retroactive processes. But the proposed theories didn’t always work. Today’s connectionist systems are analogous in some ways to the “non-trivial machines” of von Foerster. They lay claim to a common filiate (the formal networks of neurons of McCulloch and Pitts) and, more recently, they attack the problem of the recognition and the representation of recursive, nested structures. This new step is reminiscent of the one which took us from Ashby’s homeostat to the application by von Foerster of the concept of recursivity to the problem of non-trivial or memory-equipped machines. We will attempt to show that the analogy is a real one and that von Foerster anticipated many of the current problems. But we will also see that with the development of working models, other more redoubtable difficulties have appeared precisely where Second-Order cybernetics failed to foresee them.