combo, embodiment, linear logic

15 views
Skip to first unread message

Linas Vepstas

unread,
Jul 3, 2012, 6:59:02 PM7/3/12
to opencog
Some boring questions, concluding with linear logic.

-- The embodiment code uses combo... will this continue, or is this eventually being replace by scheme (as hinted in earlier emails)?

-- The embodiment code uses a combo evaluator that performs "variable unification"  However, in the combo code base, this code is marked obsolete, and is slated for removal ... so again, is this code eventually going away? Being re-written?

-- There are 4 embodiment test cases failing. As far as I can tell, its because someone moved/removed a script to autogenerate the scheme bindings for the various embodiemnt-specific atoms.  Anyone working on fixing this, or is this code going obsolescent?

-- Combo seems to have all sorts of extensions for actions and perceptions in it, for use by the embodiment code.  Given the above questions, do these continue to live on, or are these somehow obsoleted?

-- If they do live on, is there a particular theory of logic that applies to them?  by "theory of logic", I'm thinking of a system such as some form of "constraint logic programming", or some kind of planning logic (e.g. "can't walk through doorway until door has been opened").

-- Does PLN have some kind of constraint-programming-like subsystem defined in it?  i.e. for solving planning problems?

-- I've recently been reading about "linear logic", a kind of crisp logic, generalizing the usual propositional & predicate logic and "symmetrizing" sequent calculus.  It has a natural embedding of constraint-type declarations, e.g. "with one dollar you can buy one apple or one candy bar but not both"  and have been wondering if it might provide a better foundation for learning any kind of action/perception stuff within combo/moses.  Maybe. or maybe not.  I dunno.

Linas

Ben Goertzel

unread,
Jul 3, 2012, 9:22:31 PM7/3/12
to ope...@googlegroups.com
I'll let someone else answer about the Combo/embodiment code...

But a couple brief comments...

Right now, we are focusing on using PLN and Fishgram to learn actions
in the game world, not MOSES.... Although we have a plan to, later
this year, integrate MOSES for another purpose (to supplement Fishgram
for recognizing patterns among game-world percepts)

PLN's planner now does something simple similar to STRIPS planning.
There is a partially-complete version that does something more similar
to a partial-order planner, which will be done soon (for real, this is
part of our near-term plan...)

-- Ben
> --
> You received this message because you are subscribed to the Google Groups
> "opencog" group.
> To post to this group, send email to ope...@googlegroups.com.
> To unsubscribe from this group, send email to
> opencog+u...@googlegroups.com.
> For more options, visit this group at
> http://groups.google.com/group/opencog?hl=en.



--
Ben Goertzel, PhD
http://goertzel.org

"My humanity is a constant self-overcoming" -- Friedrich Nietzsche
Reply all
Reply to author
Forward
0 new messages