On Wed, May 1, 2013 at 11:57 AM, Mike Archbold <[email protected]> wrote:

> Regarding #1,.... true if you are trying to sell something in the near
> future.  What I learned from programming in the last 25 years or so is
> that there is no point in writing any code unless you fully understand
> the problem at hand.


So you totally disagree with Paul Graham, Alan Kay and others that advise
it is sometimes more important to take a more "painterly" approach (as in
oil painting) of trying things and exploration of the problem space using a
set of programming tools?  It has not meen my experience in 35 years of
programming (trumped you <g>) that we generally fully understand the
problem at the beginning of creating a software system.  We understand
early formulations of goals and have some ideas about approaches and know
from experience and research what sorts of constraints and characteristics
those generally entail.  But very seldom would I say there was a full
understanding.

Once back in early Structured Programming days a manager wanted full SP
techniques applied to a system although we didn't have any CASE tools
(which weren't very good anyway) to help with that.  So four of us labored
for about 5 months.  We generated an impressive large document full of nice
diagrams, data dictionaries, cross-references.. We felt good about it
because we were quite sure it would be trivial to write the code that did
exactly that after we had thought through all the nooks and crannies in
such detail.  But upper management looked at the costs to date and the size
and complexity of such a document and canned the entire thing.  :P



> If you start too soon somebody will introduce
> some requirement that seems innocent enough, but throws the design off
> so badly you have to start over.  The problem is that people often
> rush to write code to have something to show for it without gaining
> adequate understanding of a problem domain.
>

Well, that is the thing.  You don't really know fully what you want and
your users even more often do not know until you build something and put it
in front of the concerned parties.  It is often extremely useful to put a
first guesstimate in people's hands to play with to refine what is actually
desired.  Humans are not very good at thinking through all ramifications
and refining what they want to the smallest most powerful statement of
need.  They are notoriously bad at whole system thinking.   And even worse
at running a computer inside their mind.


>
> Now, in the context of AGI, what we are saying is that we want a

program that does, what?  Everything!  So that throws a bit of a tire
> iron in the works, creating a seeming deadlock, not easily resolved.
>
> I'm not saying don't write any code.  I guess I am saying that I
> wouldn't bank heavily on emergence or some magic happening later to
> make it work.
>


Now there I agree with you somewhat.  I think without fundamental insights
into General Intelligence that we are very unlikely to get anywhere with
AGI.  But there are lots of plausible looking notions about GI out there.
 And many of them have not been rendered in software to the satisfaction of
various AGI researchers.   Without that rendering it is difficult to sort
out which of these notions are more plausibly useful.

>From another perspective, I don't think we will ever have a full theory of
how General Intelligence works.  I think that something more like what
happened via evolution to make us generally intelligent will play a part.
 I think we will make systems that have enough of some set of critical
capabilities and enough degrees of freedom around the use and combination
of those capabilities that General Intelligence will emerge.  At the least
I think we human hackers are incapable of hacking together more than a
"seed" of a true AGI that needs to go through some
evolution/self-improvement cycle to become truly AGI.

My 2 cents.

- samantha



-------------------------------------------
AGI
Archives: https://www.listbox.com/member/archive/303/=now
RSS Feed: https://www.listbox.com/member/archive/rss/303/21088071-f452e424
Modify Your Subscription: 
https://www.listbox.com/member/?member_id=21088071&id_secret=21088071-58d57657
Powered by Listbox: http://www.listbox.com

Reply via email to