RE: 'Singularity Realism' - A few thoughts

From: Mark Waser (
Date: Tue Mar 16 2004 - 18:57:01 MST

> But all this is just using XML as a *data language*, not using it to
> store procedures.

 But your procedures ARE data. Why not store/transfer them like your other
data? Isn't this exactly the unhelpful dichotomy that always seems to get
in the way of recursive self-improvement?

> Still, this is using XML for data storage. While Atoms defining
> procedures may be stored in this way, they're being manipulated on a
> very coarse level (i.e. being categorized according to general qualities
> like topic area, period of time in which they were used, etc.). They're
> not being manipulated "as procedures", used to generate new procedures,
> etc.
> XML is, in my view, a pain in the butt for this purpose.

I would agree that you don't want to manipulate procedures when they're in
their XML form - - you'd clearly load them into their data structures first.
I'm not sure that I understood your comment about > not being manipulated
"as procedures", used to generate new procedures, though.

> I don't share Samantha's LISP evangelism -- I like LISP in principle,
> but personally I find it a bit annoying. I prefer Haskell if I'm going
> to use a functional language. And I love Refal in principle, though I
> haven't done much coding in it. For Novamente, we chose C++ because we
> were paranoid about scalability to deal effectively with massive amounts
> of memory....

 LISP is interesting and good to know. EVERYONE who wants to argue
should be FORCED to learn Prolog (the only language where I've had the
experience of programming awkwardly in it until I actually FELT something
twist in my head and then suddenly programming it felt and was natural . . .

 A curious question - - > Are you planning to build Sasha on top of Haskell
or in C++?

Refal has a ton of cool features but also looks unnecessarily verbose at
times. I don't know whether it is to support the optimization for
supercompilation or to support some of it's other neat features but it seems
as if some of the verbosity could be eliminated in favor of
behind-the-scenes expansion (i.e. could you support all of Refals features
with the programmer using Haskell-type syntax). I'm probably missing a lot
here though.


This archive was generated by hypermail 2.1.5 : Wed Jul 17 2013 - 04:00:46 MDT