Shane Legg wrote:
On 3/8/07, *Ben Goertzel* <[EMAIL PROTECTED] <mailto:[EMAIL PROTECTED]>>
wrote:
using AIXI-type ideas. The problem is that there is nothing,
conceptually, in the whole army of ideas surrounding AIXI,
that tells you about how to deal with the challenges of finite
computational resources. (And my view is that dealing with
these challenges is actually the crux of the AGI problem.)
Yes, indeed I have asked Marcus Hutter about this and his feeling
was that real AGI may well turn out to be large and complex, even
if the theory behind it isn't too bad. For example, conceptually a
database is pretty simple, but actually making an efficient reliable
database that can scale to huge data volumes is very complex and
takes many many years of work to get right.
And, I think that the theory underlying a real AGI is going to be more
complex than the theory of AIXI **or** the theory underlying
relational databases....
I spent some time late last year articulating a set of 17 mathematical/
theoretical propositions, the proof of which would go a long way toward
mathematically justifying the Novamente AGI design. (I didn't make
the propositions fully rigorous, but got halfway there and made them
semi-rigorous, then got distracted by more practical stuff.)
The propositions came out not having much to do with AIXI type
theory, and more to do with (for example) the actual statistical
properties of realistic-scale program spaces induced by the biases
of particular search algorithms (using "search" very generally)
interacting with particular sorts of environments.
Rigorously formulating and proving these propositions would be a lot of fun
for me, but there are more pressing tasks in the Novamente project
at the moment...
But my point is: I don't think that "AGI theory" is intrinsically
hopeless ... but I think that the sort of theory you need to grapple with
realistic-resources AGI is a bit different from the sort you need to grapple
with near-infinite-resources AGI. (Though there are certainly
commonalities,
e.g. the language of probabilities, theoretical computing machines and
program spaces...)
-- Ben G
-----
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/?list_id=11983
