>>11there's no guarantee it will continuously enhance itself in any particular direction.
Algorithmic enhancement can come only from the mind-designers and the executive programmers. The
http://ai.neocities.org/MindBoot.html enhancement will serve only to give the AGI more knowledge -- of words, and of ideas conveyed with words -- and also serve to teach the AGI community how to write code for many different kinds of AGI enhancement.
Since early adopters may fear to run self-modifying AGI code in their unprotected computers, early versions may include the self-replicating and self-enhancing code in a "commented-out" form, so that end users may study the code before letting it spawn one enhanced AGI after another. Such an "AI Go FOOM" scenario warrants more discussion in the
http://old.reddit.com/r/ControlProblem forum.