This article was initially started as a manifesto for the Tunes project down in early 1995. For years, only the main arguments in the first part were laid down and explained. I still haven't finished it.
Even simple purely mathematical objects like the digits of number pi or the Mandelbrot set have behavior beyond comprehension, not to talk about Chaitin's ``number of wisdom'' Omega [3]. Physical systems are even harder to comprehend, and the most precise computer simulations of even the simplest non-linear dynamic systems quickly diverge into chaos. Finally, intrinsicly ununderstandable is the chaos that results from competitive interaction with a large number of sentient, as witnesses the stock market, for instance, for any perceived regularity self-defeatingly gathers against it the behavior of those who perceive it.

All in all, there are many sources of chaos, from mathematical complexity, physical randomness, competition with other systems and interaction with human beings, that could give birth to ``intelligent'' behavior in machines. The whole problem is for AI researchers to develop tools to identify and harness this chaos, so as to take advantage of it.
I've received reproaches about my definition including in an OS all the ``interactive'' parts. Firstly, my definition of an OS being formal, I wouldn't like a fuzzy concept like informal ``interactivity'' to be used in it. If that's to mean anything that the user can directly see on screen, then there are lots of OSes based on dynamic languages, like LISP, FORTH or RPL, where just everything is thus interactive, so a definition for an OS definitely should include such interactive things. Now, if ``interactive'' is to mean ``purely interactive'', or ``not program-accessible'', that is, ``anything that no program written over the OS can access/modify/simulate'' (which would amount to nothing in a ``Good'' OS, by the way), then an OS also should include interactive things, to account for all the expectations one has about the system behaving in such a way when such thing is done (typing such thing on the console, clicking with a mouse, etc). Such behavior is rightfully described in books teaching how to use such OS. So I see no reason why to exclude these from my definition of an OS. Surely, I reckon that the concepts of being user-accessible or program-accessible, are indeed interesting ones. But they are orthogonal concepts to me to what is to be expected from a random computer extracted from a considered set. Surely the conjunction of these interesting concepts might also be interesting, but these concepts are more expressive (hence more useful) when kept orthogonal. Else, how would you name the purely interactive part of what I call an OS?
Please make me wrong and tell me how to convert my idea of being rich into actually being rich.