Re: Prolog and AION

On May 25, 7:38 am, Peter Van Weert <Peter.VanWe...@xxxxxxxxxxxxxx>
wrote:
Backtracking is an orthogonal to backward or forward chaining. All
forward chaining systems I can think of are committed choice, but this
does not mean backtracking with forward chaining would not make sense.
So no, that's not it, your previous post is closer to what I was
thinking of.

Cheers,
Peter

Thanks for that, that clarifies things for me. Also we seem to be
getting down to the likely differences between AION and Prolog quite
well, despite never even seeing AION...

Committed choice seems fine in a purely forward chaining system. That
is, if a implies b and we know a to be true then committing to b seems
quite sound. In prolog we try to prove b by hypothesising that a is
true and working from there, and backtrack in the event that we find a
not to be proveable after all. So, I wonder, these systems that
combine forward and backward chaining, do they provide an undo
operation that you can call explicitly? or do you have to tread very
carefully to avoid creating unsound rule sets?

I notice that in that article he talks about having subgoal rules and
action rules. Seems to me that so long as the actions come after the
subgoals, then things will be fine. So a rule of the form:

if a then (first do x,y,z) add b to working memory.

is like the following in Prolog:

b :- a,x,y,z

Having a forward rule that puts actions before subgoals would seem to
be unsound. So a rule of the form:

if a then add b to working memory (then do x,y,z).

is kind of equivalent to:

b :- a,x,y,z

but the Prolog fails cleanly.

Maybe these languages don't let you put actions before subgoals? I've
no idea. Or do you write:

if a then add b to working memory (then prove x,y,z onfailure undo b)?

Rupert

.

Relevant Pages