inference trails

[Ben Goertzel]

Some discussion on whether inference trails are needed or not occurred
on the Open-NARS email list .. pasting one of my mails here as it
seems relevant...


***
Although -- we still do have inference trails in PLN, and we also
sometimes go beyond that and use an auxiliary Atomspace to store the
whole inference digraph that gave rise to a given truth-value
update... (in which case revision can make a better stab at using
inference history to account for dependencies)

So unfortunately I think the answer is that sometimes trails (or more
complex inference history structures) are what you want, whereas other
times you can do without them and let the circular multiple-counting
of evidence kinda come out in the wash...

Crudely, I feel that large-scale low-accuracy inference ---- like in
perception processing, or "unconscious" episodic memory recollection,
etc. --- can get by with the "comes out in the wash" approach ...
whereas focused precise deliberative reasoning has got to use trails
or something more sophisticated

I would conjecture that the brain's trail-like mechanisms exist via
the cortex-hippocampus interface, and thus are only invoked for
inferences where working memory plays a major role ... not for the
vast mass of "long-term memory only" background "unconscious"
inferences..

-- Ben
***

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

“I tell my students, when you go to these meetings, see what direction
everyone is headed, so you can go in the opposite direction. Don’t
polish the brass on the bandwagon.” – V. S. Ramachandran

[Linas Vepstas]

Kind-of off-topic, and I've read only about 5 pages so far, but:

Girard introduces the concept of "Loci" (kind-of-like memory pointers)
in Locus Solum whose subtitle is "from the rules of logic to the logic
of rules".

Insofar was PLN is about rules that capture logic, and the URE is
something that applies rules, its worth understanding what kind of
logic applies to the URE itself. I think Girard has a partial answer
to this, unfortunately, its not written at an easy-to-grasp level.

--linas

> --
> You received this message because you are subscribed to the Google Groups "opencog" group.
> To unsubscribe from this group and stop receiving emails from it, send an email to opencog+u...@googlegroups.com.
> To post to this group, send email to ope...@googlegroups.com.
> Visit this group at https://groups.google.com/group/opencog.
> To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CACYTDBc2PzT_WA0J%2Bbr8dQ4zcsTpbKv%3Dy3%3DVY56XUi29bBKZ2w%40mail.gmail.com.
> For more options, visit https://groups.google.com/d/optout.

[Linas Vepstas]

Oops I forgot the URL

http://iml.univ-mrs.fr/~girard/0.pdf its from 2001

I suspect that there are later, more approachable/readable intros to
the topic from other authors. The wikipedia coverage sucks, though.

--linas

[Ben Goertzel]

thx, it's on my reading list!

> To view this discussion on the web visit https://groups.google.com/d/msgid/opencog/CAHrUA35L0ZNPLPVhjXVC99GsRyuS3P%3D87MLhLhgabXw0PJOXmQ%40mail.gmail.com.

> For more options, visit https://groups.google.com/d/optout.