--- "John G. Rose" <[EMAIL PROTECTED]> wrote: > > From: Matt Mahoney [mailto:[EMAIL PROTECTED] > > > > > > I think all of these boil down to a simple equation with just a few > > > variables. Anyone have it? It'd be nice if it included some sort of > > > computational complexity energy expression in it. > > > > Yes. Intelligence is the expected reward in an environment simulated by > > a > > random Turing machine. http://www.vetta.org/documents/ui_benelearn.pdf > > > > Right, that paper. I am trying to absorb the definition fully, reading up on > the background. > > It's a general definition but particular agents can be put into the > equation. But not computable right? Only estimable? I was kind of thinking > of something more granular that actually doesn't include K or any other > complexity function like Kt, etc.... but that's just speculation...
That is correct. Universal intelligence is not computable. But is is useful because we can prove things about it. For example, if the environment has a known upper bound on complexity, then a maximally intelligent agent must have at least the same complexity. It follows that recursive self improvement is a probabilistic algorithm, if by "improvement" we mean increased universal intelligence. We could not prove such things using computable measures like the Turing test or text compression. -- Matt Mahoney, [EMAIL PROTECTED] ----- 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/?member_id=8660244&id_secret=62611804-a17c0b
