Thanks Ben, Russel et al for being so patient with me ;-) To
summarize: AIXItl's inefficiencies are so large and the additional
benefit it provides is so small that it will likely never be a logical
choice over other more efficient, less optimal algorithms.
Stefan
The additional benefit it *would* provide would be large, but, the
inefficiencies are so large that the theoretical benefits or otherwise
are irrelevant..
Basically, what AIXItl does is, before each action it takes, search
the set of all possible computer programs of length l and runtime
t, and figure out which of these programs should be allowed to
control the action (based on prior history). But there are vey
vey many computer programs of length l and runtime t to search
through, so this is a totally infeasible way to ever do things in
practice.
The reason AIXItl can do anything that any other program can do, is
that it searches through all other programs (subject to the length and
runtime requirements). But the reason it is so slow is that it is
continually searching through a humongous space of possible programs.
Juergen Schmidhuber has tried to partially work around these problems
in his OOPS AI program, but the attempt has not been very
successful...
Ben
-
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/[EMAIL PROTECTED]
