|
This post is in response to [Name Omitted]'s request for info re: a model for
Modeling, or a model for decision strategy elicitation (or knowledge
acquisition). Mark Stanarevic kindly gave me a glowing reference with
respect to my Expert Systems background before getting into NLP. I'll
start by describing my present state with respect to my knowledge
acquisition &/or modeling skills. Then, I quote the original request and
answer any points not addressed by my general answer. I hope this isn't
too long for you, and that it provides some value for you at this early
stage in my own work.
Throughout my exploration of NLP and other disciplines, if there were one
area in which I've been repeatedly hearing kudos, I think I'd have to point
at learning strategies and skills modeling. I may not be an ideal model
for any number of things, yet I do keep hearing people saying they want to
model some of the things I seem to do, like, learn quickly.
So obviously I pay attention to topics like this, because I believe I'm
still formulating my model for modeling, if that's what you'd call it. And
I haven't yet found one pattern (or set) that keeps repeating. I do
believe its there, its just that since I started keeping much better track
of WHAT I'M DOING as I'm doing it, I keep finding myself doing things
differently depending on the specific circumstances.
So far, then, I'd say my effectiveness -- wherever I'm effective -- seems
to be my behavioral flexibility. I break many of my own patterns wherever
I find them (unless they're patterns I *REALLY* enjoy... ;)
I do write my specific strategies up, as I find them, in order to identify
my own models for modeling, and for potential publishing, but I -feel- that
they're not yet mature enough by my own standards for public evaluation.
So I don't want to post them to the list for review -until- I -feel- better
about : knowing more about : where the model applies, AND how I
modify/torque/apply it in flexible ways.
I wish to avoid any "and you will see this material in the upcoming book
which if you buy it early BLA BLA BLA BLA BLA " stigma -- and so, when I
feel more ready, I will gladly post some of my concrete modeling ideas to
the list for review. I will need to have some forum as a test for
developing models, so I believe it could be beneficial to those here just
as I believe that I can learn from all of you. And how wonderful would all
that be?
[Omitting Quote of Request for a model of modelling experts]
Expert Systems (or Rule-Based Reasoning) is an area of artificial
intelligence wherein people develop software to behave just like Human
Experts do when presented with a new case requiring some process or
diagnosis. This domain of work assumes that there is a static body of
knowledge, or, a set of rigid rules (called a "knowledge base"), that are
valid for... right now. I think this has had merit, given some
limitations. And despite being static, a knowledge base can be quite
sophisticated in terms of the varying order of how rules apply, the level
of rules -- i.e. how specific do they get...
You can have rules which assume something and make any number of inferences
from that point, such that if the original assumption becomes incorrect
everything relative to that original assumption disappears...
You can have rules which depend on (1) the presence of specific
information, on (2) the absence of information, and on (3) millions of
combinations of both (1) & (2).
You can have rules which have priorities that change over the course of a
certain process.
Using these ideas and MANY more, the opportunity for designing or building
complex relationships of rules are endless.
Now, as a knowledge engineer (KE), I found that one can reach several
limits as to what you can successfully build in any financially viable
timeframe without running into degradation on several fronts. One of the
biggies is the fact that the model of rule-based-reasoning alone makes it
difficult (not impossible) to design a Knowledge-Base whose 'architecture'
or intrinsic flow structure will survive multiple major adjustments/
augmentations to the knowledge base itself!
You see, Rule-Based-Reasoning was not developed as a successful (i.e.
viable) learning model, though some have developed systems that were
reasonably good at developing certain kinds of new rules on-the-fly without
a KE to intervene & write the code.
For that, the software industry turned to fuzzy logic, neural networks &
genetic algorithms, to name some of the more popular approaches.
Nowadays the best work in modeling human intelligence -- with software --
is in the creative combination of these and other approaches (this is known
as Hybrid Systems).
So. I DO still use KE techniques for eliciting knowledge, but more often
than not, since many of those often do NOT account for the successful
elicitation of KINESTHETIC information, I delve directly into the META
model, and back out of it whenever my INTUITION tells me to pursue a
certain path. So I am working, these days, on taking my INTUITION and
codifying it somehow within the framework of how I move from one model of
elicitation to another... ;)
[Omitting Request for any info on a programmed form of the META Model.
Knowing a bit about how inflexible even a reasonably solid attempt at this
would be, I have some anticipation of a SUCCESSFUL effort toward this being
a rather daunting task (size/time-wise). I'd like to be part of that
effort, of course, time-permitting...
Now I can teach what I do by demonstration. But if people want the goods
on HOW I do what I do, either, I suggest waiting patiently for me to try
and codify it, or, come down to Tampa, and I can do some private work
modeling you -- and you could model my modeling behavior.
Anyway, as part of all this, I'll be teaching a couple of weekend Strategy
Elicitation Seminars sometime soon. Feel free to check my website in the
next few days or so for a new calendar, if it interests you. As of this
writing, I've only got thru October up on the site & visible!
Regards,
- Jonathan Altfeld
| 
