On 9/12/06, Russell Wallace <[EMAIL PROTECTED]> wrote:
On 9/11/06, Stefan Pernar <[EMAIL PROTECTED]> wrote:
> Assuming that AIXItl be effectively better at achieving its goal than
> any other agent with the
> same space and time resource limitations (t,l) would still make it the
> algorithm of choice no matter how computationally intense, as any
> other system given equal resources would come to a less optimal
> solution by definition.
If you believe that, then go ahead and implement AIXItl and see what
happens; it's a straightforward enough algorithm.
I do not claim that I understand the math behind AIXItl and doubt that
I either have the diligence or skill to implement it. My approach to
providing feedback to Ben's and Shane's writings was to highlight
logic inconcistancies (at least from my limited understanding) and
point towards similar works in the same direction.
My assumption that for all x f1(x) > f2(x) with f1 being the AIXItl
algorythm and f2 being all other algorithms, x available computational
resources and the output being 'intelligence', let to my statements
above. Ben and Shane now confirmed that f1(x) > f2(x) is only true for
very large x. AIXItl still is the most intelligent algorythm possible
with the drawback of being very computationaly expensive. Looks to me
now that other approaches - such as Ben's Novamente design - while
yielding sub optimal intelligent results only require a fraction of
the computational resources.
An interesting question for me now is when the comparative
inefficencies of the AIXItl algorythms will become irrelevant due to
ever decreasing cost and availability of computational resources.
How much more intelligent is the AIXItl algorithm compared with
others? Marginally or substantial?
How much more computationally intense? By a factor of 10^3...10^9?
With that knowledge, one could plot the cost benefit functions of
available algorythms and predict at what point one should switch to
the optimal AIXItl.
Just for kicks - let's assume that AIXItl yields 1% more intelligent
results when provided 10^6 times the computational resources when
compared to another algorythm X. Let's further assume that today the
cost asscociated with X for reaching a benefit of 1 will be 1 compared
to a cost of 10^6 for a benefit of 1.01 when using the AIXItl. To
simplify I will further assume that cost of computational resources
will continue to half every 12 month. In this scenario it will be
computationally cheaper to apply the AIXItl in less than 20 years.
I ordered both Ben's and Hutter's books and will read more on both -
looking forward really ;-)
Cheers,
Stefan
--
Stefan Pernar
App. 1-6-I, Piao Home
No. 19 Jiang Tai Xi Lu
100016 Beijing
China
Mobil: +86 1391 009 1931
Skype: Stefan.Pernar
-----
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]
