Re: [isml] Making HAL Your Pal (fwd)

From: Samantha Atkins (samantha@objectent.com)
Date: Thu Apr 19 2001 - 17:28:46 MDT


Personally, I think this is where the decision to not Open Source the
project or at least aspects of it hurts a lot. If a Flare SourceForge
project was started it wouldn't take much more than a face-to-face or
exchange of emails with interested individuals after a very short few
pages (or less) for the initial project description. If some of these
parts are done with OS then you don't need to wait to hire a paid
full-time lead.

Some of us who would very much like to donate time/energy will not write
any significant code unless we are assured that code is Open Source.

On the face of it, basing a language on XML makes very little sense to
me. Lisp is far more flexible. XML is made for distributing data that
is partially self documenting. It is highly redundant and the encoding
is elementary and inefficient. You could easily get the parts you want
from it, as I understand it, using frames and slots or property lists
and so on in a language like Lisp and have all much of the machinery
already in place to do development, debugging and so on. To write a
program language in XML with tags is to go down the same dumb rat-hole
that many Web Servers have. It is a gross and inefficient way to define
programs. You need tools that speed thought and development not ones
that bore real programmers to death.

Writing a language for AI is not a proprietary or one team undertaking.

Why do some of the sparse papers out there talk of implementinf Flare in
C++? There is no need from an efficiency point of view. Lisp can be as
optimized just as much and is much more flexible.

The idea of "domdules" that contain semantic and usage information about
modules is a good one. The tricky part is having good semantic language
conventions that are computable. That itself is not a small research
project.

- samantha

Adrian Tymes wrote:
>
> "Eliezer S. Yudkowsky" wrote:
> > Guys, I can burn myself out trying to get everything to everyone, and all
> > it means is that *nothing* will ever get finished.
>
> You missed my point. You will remain as one full-time guy until you
> change your actions to allow others to help. You don't have to retype
> your notes for each and every person; just type up what you have in
> some place everyone can see, point questioners there, and use that
> place as your main working repository.
>
> That doesn't mean white papers. That means *running code*, or at least
> the framework of such (at least pseudocode with no "undefined magic
> happens here" bits), which you then won't have to rewrite for various
> architectures (code mongers can port your stuff to their platforms:
> porting is a common operation).
>
> > I'm sorry if I managed
> > to give you the impression that it was just a couple of weeks, way back
> > when, but we can't start implementing Flare until we have at least one
> > full-time person to run the project, and I've been too busy working on
> > "Friendly AI" - am *still* too busy - to write anything up on it.
>
> You're in such a rush to complete your project, you don't have time to
> complete your project. It's a common malady/paradox among modern
> projects, though it's more visible among commercial projects (since
> they ship their buggy, incomplete, rushed project anyway, on a schedule
> close to what their developers originally estimated but with an attempt
> to do it in half the time, and the other half trying to patch up that
> mistake).
>
> That is to say, I was volunteering to *be* that "full-time person" to
> run Flare, but I couldn't since you failed to get me the info; other
> people look at what you have, see a total lack of specs one could
> develop from, and don't bother even to ask, assuming that those specs
> don't and won't exist. My situation has changed in light of this and
> other things, and I may no longer be in a position to write Flare - but
> there are others.
>
> Suggestion: write up and post what you have, then mostly forget about
> it until someone (in a similar situation as I was) starts writing it
> without your oversight. Get it ready for handoff, so it can take off
> with minimal, maybe zero, further input of your time afterwards.
>
> Question to the rest of the list: I'm sure someone else can see the
> point I'm trying to make. Can anyone else state it in a clearer
> fashion, such that Eliezer can see the path out of being a perpetual
> one-man job? (A certain set of actions will be necessary; funding is
> not the only path, nor even the one most likely to get results if
> funding and a biz plan are not actively pursued.)



This archive was generated by hypermail 2.1.5 : Sat Nov 02 2002 - 08:07:06 MST