[alicebot-archcomm] Loose ends

[Kim Sullivan]

> Don't take those numbers absolut, there are cats doubled in French or
> just test cats. But they show a trend.

The question is (since most of these have been written by Rich himself) if
they aren't just a matter of personal preference, or 'coding style' - Rich
is rather fond (obsessed, if you pardon me, Rich) of extremely simple
categories. What if someone wrote the same amount of categories, but used at
least three random replies in every category, setting at least one of X
state variables and had a condition to process them? Would that make a
better bot?

My super simple bot has only about 170 categories, but already more than 25
conditions and about 70 randoms - and almost each time  I go through them, I
try to add a random to categories that don't have it yet (or a condition,
again with a random for every choice).

> Or to say it in another way: Is there really a need for tags, which
> mutate AIML into a real programming language? Are there other usage
> statistics? I have the impression - mainly by looking at the style list
> - that many things, even shuffle, array pop etc. - can be done in pure
> AIML, aka "programmed" ;-)

The question is, what's a 'real' PL - this probably is the main task, to
differentiate between things that are necessary and things that can already
be achieved in a reasonable way with what's there. The question wether to
keep AIML at the level of Brainf*ck (or to advance it to a really high level
language like C (all of them are real programming languages - you can write
the same program with bf's 8 instructions as you can with C). I prefer the
first approach (as stated in 1.1 goal 2 of the spec) - all other
programmer's candy should be left to individual implementations and not
taken into the standard (unless a feature becomes common among several
implementations and botmasters use it regularly).  The function of AIML is
IMHO not to actually solve complex mathematical problems, but to understand
the NL input and pass it to the solver in a format that is more 'machine
readable'.

> And - just to play advocatus diaboli - should a general AIML not be more
> extensive? What about neural networks, knowledge webs, reasoning etc.
> defined by AIML tags? All that stuff normally subsumed under AI.

Yes, but only by individual implementations and not by standard (we don't
need to have everything in the spec, just enough for people to be able to
extend the basics). At least not at first. As I said - if there is enough
demand, someone will make it. We should use the opensource model for our
benefit, and not have it work against us.

As for the remaining points (patternside disjunction...), I'll wait for
additional comments/input, because I think I've already voiced my opinion
several times, and there's no point in repeating it (yet :-)

Regards,
Kim