If we're talking language for AGI _content_ (as opposed to framework
for which Ben Goertzel has made a fair case for even C++), then more
like removal of features. Because for AGI content, it's not what you
can do in principle, it's what you can be _casual_ with.
Correct, this is an important distinction.
One thing that's nice about LISP --- at first glance -- is that it looks
like it can be a language for both AGI content and AGI framework.
But I believe this is somewhat deceptive. In principle LISP could be OK
for AGI framework (though I'm not convinced it's there yet ... though
Allegro LISP arguably comes close...), but I don't think it's right for
AGI content.
On the other hand, you could build an AGI-content language by
**extending** LISP ... whereas if your framework language is C++ you
need to make a content language totally separately.
In fact our content language, Combo, looks a bit like LISP, but with
other features like
-- explicit higher-order typing [not yet implemented, but needed soon]
-- a particular kind of uncertain truth values
-- probabilistic tools for dealing with statements based on their truth
values
Currently this content language is used almost only for internal
AI-generated programs, and has an awkward textual syntax, but we intend
to improve the syntax so that in some cases we can supply the system
with human-programmed modules to use as a starting-point for learning.
Anyway, the plan is that initial NM self-modification will take the form
of NM modifying its **cognitive control scripts** that are written in
the internal Combo language ... modification of the underlying C++ code
is going to be a later-phase thing. [This also enforces some basic,
non-absolute "AGI safety" in that the C++ layer provides certain
constraints.]
Ben
-----
This list is sponsored by AGIRI: http://www.agiri.org/email
To unsubscribe or change your options, please go to:
http://v2.listbox.com/member/?list_id=303
