Steve Richfield wrote:
> Ben, Mike, et al:

> MIke: Anything that can be put into words (and MUCH more) can be
> expressed mathematically. Presuming that your claims are true, then
> people are fundamentally unable to express AGI operation. If so, there
> can be no collaboration, and I think we can all agree that no one person
> is going to build an AGI from scratch. Hence, you are in effect saying
> that it is impossible to build an AGI. Right? If so, then just say so
> (which I think you already have) and let the smart ones heed your
> warnings not to waste their lives, and move on to something ELSE.

I think what the real confusion here is that people fail to realize that
there are three, almost perfectly orthogonal mathematical domains
related to AGI. Understanding the nature of and distinctions between
these domains might actually go a long way towards breaking the ice-dam
here.


These domains are as follows:

1. The domain proper.
2. the domain of implementation.
3. the domain of description.


First, the domain proper is what we are really trying to deal with. It
is an enormous, and messy world. To the extent it obeys precise laws, it
does so on a scale literally 23 orders of magnitude smaller than what we
deal with on a daily basis. Some of those laws appear to be "select a
perfectly random number according to such and such a probability
distribution". Or even "find an equilibrium point in this 4 dimensional
system ignoring all the stupid monkeys who think the 4th dimension is
strictly one-way".

There are many critical things to understand about the first domain at
the practically accessible scale. The first of which is that it is
utterly asymbolic. Any symbol used to describe it comes from within the
observer, not the system. Everything said about it that isn't about some
quantum value is fuzzy and imprecise. All models are approximate and
applicable only in a few situations under certain conditions. Any
general agent must be able to construct and apply models and ontologies
on the fly.


2. The domain of implementation. We hope that this domain is both
computable and tractable. This domain is largely orthogonal to the first
domain because, for the sake of generality, it contains as little
information as possible about the domain but rather has the capabilities
it requires to solve the problems that are necessary to solve. It is
organized around the structure of the *AGENT*. Some non-trivial agent
architecture is a fundamental requirement of any useful AGI system.
Beyond that, it is the slate part of the blank slate.

Finally, we have the domain of descritption. This is the domain which
attempts to measure and describe the solution implemented by the second
domain, and by extension features of the first domain. It is entirely
disposable and unnecessary for a successful AGI. It's primary purpose is
to assess the success of a given AGI design. Far too many people confuse
progress in this domain with progress in the second domain. =( The only
contribution domain three can offer is that it lets you see what you're
doing in the second domain, to some extent. For this reason, this domain
is completely orthogonal to the other two domains. I'm really sick to
death about hearing about results from research in this domain, I just
don't care. =|



Okay, now that is cleared up, go out and build some 'bots! =P

-- 
E T F
N H E
D E D

Powers are not rights.





-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-c97d2393
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-2484a968
Powered by Listbox: http://www.listbox.com

Reply via email to