Intelligence defined as prediction

From a thread on "The value of ideas in language" on the grammatical-incompleteness Google group (http://groups.google.com/group/grammatical-incompleteness/browse_thread/...):

You may want to look at the AI work of Schmidhuber and co. which
defines intelligence in terms of prediction. This enables them to
avoid the pitfalls of logic, at least. They are hoping to build a new,
random, AI on this basis:

http://www.idsia.ch/~juergen/newai/newai.html

I believe Marcus Hutter extended this definition of intelligence to
show it is equivalent to some kind of "usefulness". See his book:

http://www.hutter1.net/

E.g. Hutter - "Most, if not all known facets of intelligence can be
formulated as goal driven or, more precisely, as maximizing some
utility function. It is, therefore, sufficient to study goal driven
AI. ... The problem is that, except for special cases, we know neither
the utility function, nor the environment in which the system will
operate, in advance. The mathematical theory, coined AIXI, is supposed
to solve these problems."

As for myself, while I believe any complete description of the world
must be statistically random, I don't think statistical randomness
itself is the way to approach it.

I prefer Chaitin's algorithmic randomness.