RE01 Rice Brian T. EM2
Fri, 4 Dec 1998 21:49:15 +0800
> >>: Paul Dufresne
> >: Brian Rice
> Anyway, our problem isn't lack of low-level infrastructure;
> there are plenty of low-level runtimes (including Linux) waiting for
> some reflective system to sit on top of them.
> What we lack is well-defined guides for implementing high-level things.
> Or else, we'll end like VSTa.
> > this entire
> > development is centered around building a population of data-structures
> > from which can be read logical meaning. in other words, these
> > data-structures would form a reasoning system
> I don't believe in "one size fits all" data-structures.
> Well, we'll still need some for bootstrap or communication purposes.
> But the datastructures are strongly dependent on the algorithms to
> be used upon them, and are to be co-optimized with the programs
> that manipulate them. Now, we're not here to discover brand new algorithms
> (though it'd be great if we did find one or a few), but rather to
> put together a reflective system out of existing techniques.
> So before we start thinking about new data structures, we'd better
> get the existing ones right, and see how people with a long experience
> of using them do.
sure, but to deal with a universe of 'subject-object' pairs (programs vs.
data-structures), the homo-iconic paradigm seems to be the best one for a
reflective system. and in order to come up with a new homo-iconic ontology
that really works well, i personally see the need to form a reasoning
also, i should just re-emphasize that context-modification questions are
right around the corner, so that the 'one size fits all' idea is only a
temporary hypothesis waiting for a better-developed replacement. (just so
you know that i'm not crazy).
> In the particular field of logical reasoning, there is a LOT of literature
> by "AI" people about inference engines, and deduction-based systems. I
> readily admit I know very little about what AI people did, except that
> they achieved a lot of great things, but are (sadly) isolated from
> mainstream computing, and particularly from formal method people and
> completely from operating systems people.
yes, and this division is something with a deep sociological impact, due to
the insofar weak attempts at creating software that 'takes care of itself',
so to speak. of course, Tunes is designed to address this, and i believe
that it is a simple extension that Tunes should enable machines to do more
> Agreed. Still, for communication purposes, there need be well-known
> standard commonly agreed-upon default starting contexts of evaluation!!!
> So the above just means that the system is reflective,
> but doesn't give us a starting point.
well, we could make up tunes as a sea of ideas with 'standards' floating