I will read your message carefully tonight or sometime in the next few
days when I'm not in a hurry...

But I will respond to this one point first, hoping to spur you on to
provide more detail on your thinking...

On Sat, Sep 3, 2016 at 1:24 PM, Linas Vepstas <[email protected]> wrote:
> Backward and forward chaining are very crude, very primitive tools. Far
> superior algorithms have been invented. I'm quite sure that we can do a lot
> lot better than merely backward/forward chaining in PLN.  But we can't get
> there until we start talking at the correct level of abstraction.

I do agree on that point.

The previous implementation of the backward chainer (by William Ma)
fell apart because of some "plumbing" related to lambdas and
variables....   Nil's re-implementation will get that plumbing right.
But even with correct variable-management plumbing, yeah, forward and
backward chaining are extremely crude mechanisms that are not
sufficient for AGI in themselves...

I wonder if you would like to try to spell out in slightly more detail
one of the alternatives you are thinking of, though....  This would be
potentially relevant to Nil's work during the next couple months...

You have mentioned Viterbi and forward-backward algorithms before, and
I understand how those work for standard belief propagation in Bayes
nets and Markov chains etc., but I don't yet see how to apply that
idea in a general logic context tractably....  Of course one can
alternate btw forward and backward steps, or do them concurrently, but
that's not really what you're getting at...

Probably the point in the above paragraph ties in with the
probabilistic-programming stuff I wrote about and linked to before,
but I haven't figured out exactly how yet...

Anyway I gotta leave the computer and take my f**king broken car to
the shop now, but I will read this stuff and reflect on it with more
care sometime in the next few days when the time emerges...

ben

-- 
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 [email protected].
To post to this group, send email to [email protected].
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/CACYTDBdRe%2BAbB4PFcZGBNTUE1XK%3DEvyXNCQ%3DeXWPSWK-8HxmRw%40mail.gmail.com.
For more options, visit https://groups.google.com/d/optout.

Reply via email to