Ouch! My email filter said, and I quote "Re: [SPAM] - JESS: Date: Thu, 25 May 2006 11:43:56 -0500 - Bayesian Filter detected spam" Probably is, but I'll answer anyway.... :-)
Okey, dokey - I don't agree with everyone either. On the other hand, I didn't say anything about taking "years" to master the Jess syntax and dialog. It doesn't take that long. Nor does it take that long to master C, C++, Java, Oracle, nor any other of a dozen languages syntax before using them. On the other hand, I have seen more than one major project fail because the "KE" was really nothing more than a Java programmer with two or three months experience with a rulebased system, knowing nothing of rule architecture as opposed to Java architecture, nothing of the "fine art" of harvesting rules, etc., etc.
And, yes, I would expect them to gain various levels of proficiencies at C, C++, Java, Oracle or Jess BEFORE attempting to architect an enterprise-size project in any of those languages. That's why even Sun has several levels of examinations, beginning with programmer, developer, architect, etc. Would that we had something like this for the BRMS industry.
This free-base "kind of interaction" between "experts" and BRMS (rulebase) is what killed several programs in private industry and many more with Homeland Security, Fannie Mae and Freddie Mac. BTW, Fuzzy Jess is a nice addition to the binary form of Jess making it quite extensible without having to go to Neural Nets. And, finally, I have YET to see any worthwhile-size project pull out of ILOG JRules after completion and switch to Jess or vice versa. Nor to any other sophisticated BRMS of any size. If you have, PLEASE let me know since I would like to discuss with them how they accomplished such a modern day accomplishment.
Note: I am NOT saying that there are NONE - just that I haven't seen anything beyond the demo phase of switching from rulebase to another. And even that was flawed because one system does not necessarily have the power and potential of the other. To me, it would be like switching an enterprise C++ project to a Java project system. Both are OO, both have converter programs that will let you switch from one to the other. But every time (to my personal knowledge) it has been an unmitigated disaster.
Now, since Ernest doesn't like to clutter up the board with this kind of diatribe and meaningless debates, may I suggest that you and I (along with any others who wish to continue this thread) do it off-line and just between us and not the entire board. Anyone should feel free to email me about it and I promise to "try" to answer any questions or debate it further. Just, off-line. :-)
[Finally, my deepest and most humble apologies if my initial email sounded like arrogance and pride - it wasn't meant in that vein. And I don't have black belt in rulebased systems. The guys who wrote Jess (i.e., Ernest) and the original ILOG Rules, Neuron Data Expert and/or Neuron Data Advisor, ART, OPS5, Rete, Rete 2, (i.e., Forgy, Newell, Norvig, Nillson, Durkin, Turban, Lenat, Waterman, both of the Hays-Roth team, Simon - a Nobel Prize winner - and all of those guys - (including Paul Haley) some of whom are in the first book on "Pattern Directed Inference Systems" - those guys are black belts. I have, maybe, a purple or first brown; if that. Another 17 years or so and maybe I'll have a Sho Dan in BRMS and maybe double up in AI. Maybe... right now I'm just a really, really good hacker. Besides, the rumor is that if you do this for 30 years you die a lonely death in the bottom of a dungeon of a corporate IT department somewhere in outer Slobovia. :-) ]
SDG jco "This above all: to thine own self be true,And it must follow, as the night the day, Thou canst not then be false to any man." Hamlet, Act 1, Scene III
On May 25, 2006, at 2:46 PM, Mike L. VanGeertruy wrote: I couldn't disagree more. Currently, the vast majority of software end-users are not knowledgable about expert-systems, but you expect them to gain various levels of proficiencies before they take advantage of Jess. In my opinion, the assumption that our end-users will understand software engineering to the point that they can program is Jess is an invalid one. That assumption creates massive obstacles to broad adoption of this kind of technology. Rules engines' largest area of potential growth can only be realized by leveraging Rules' engines ability to act as an interface between true experts and thier machines. If we require subject matter experts in all areas to also become software engineering experts in Jess, there wont' be that many people adopting the technology. Imagine you are working with an old-sage of an analyst who has a very highly developed specialty in a critical area, but no computer proficiency? How would you capture thier expert knowledge? Would you require they spend 6 months learning a specific type of rules-engine syntax? Wouldn't that be a bit of a waste of thier time? Would you assign an AI specialist to take notes from this end-user's stories and have the AI specialist write the rules-engine code? Wouldn't that be a waste of two individual's time? Why not create an interface that would allow them to select from a list f "Facts", "Predicates", "Actions", and "Logical connectors" in order to create thier own highly evolved logic? Imagine a mechanical-looking interface that would allow such interaction, even in a 3-D environment; they are not too far off. We (industry) is already seeing requirements evolving in the medical and military industries to leverage these technologies to thier fullest. RulesML and JSR-94 intermixed with Web-services, the J2EE and a number of mature implementation scenarios make this kind of interaction more of an option today than ever before. However, these standardizations go even one more step further. What if you wanted to use fuzzy logic instead of Jess' binary logic? What if you wanted to use a logic engine based on ontologic data-mining? JSR-94 and RulesML combined have the potential to make the changing of logic-engines as simple as changing J2EE plugins. Will you require that everyone then become masters of every logic engine they want to plug-into their architecture? So then, why do I disagree with you? Well, the idea that a person must spend years mastering one logic engine before learning another only increases the obstacles to wide accessability of expert systems. If expert system's true potential is to be realized, it can only be done by removing obstacles. If someone wants to get an understanding of a number of expert-systems and thier logical algorithms, allow it. If they ask seemingly stupid questions, answe them. Be proud of your accomplishments as a "black-belt" of numerous expert systems; but be wary when your pride becomes arrogance and vanity. v/r, Mike Van, PMP
From: [EMAIL PROTECTED] on behalf of James C. Owen
Sent: Thu 5/25/2006 1:56 PM
To: James Owen
Cc: [email protected]
Subject: [SPAM] - JESS: Date: Thu, 25 May 2006 11:43:56 -0500 - Bayesian Filter detected spam
All:
Frequently I see folks wanting (needing?) to know about some books that they can read to learn more about AI and Expert Systems. One of the principles of the martial arts is to fully learn ONE system (meaning get a black belt in one style) before trying to learn another. Too many times students jump from one style to another trying to find an easy way to do something. If you start with one expert system, such as Jess, the for PETE"S SAKE learn that system first! Learn all about it. Get your black belt in Jess. THEN move on to another one, such as CLIPS, drools, Mandarex, ILOG JRules, Blaze Advisor, PegaRules, Haley, ART, MindBox, Aion or one of the others. You'll find most of the principles are about the same from one to another even though the language may change a bit. BU,T get a good foundation before you jump around.
One of the best all-round books is a rather old one (1979) and is a collection of white papers from the God Fathers and God Mother (only one in the batch in this particular book) of rulebased systems, including one rather good article on page 177 on conflict resolution. The name of the book is "Pattern Directed Inference Systems" and is edited by D.A. Waterman and F. Hayes-Roth. The list of authors is like a Who's Who of the early days of AI at Stanford and Carnegie-Mellon as well as a few others. You have to go to the used book list to get this one though. It costs from US $8 to US $45 depending on condition. None of the articles actually tell you HOW to program a rulebased system - very few real examples. BUT, it does give you a really, really good view from an academic viewpoint. The same book is probably in your university library or can be found at Amazon
Other than the Jess book (this I also have) there are a couple of other books that you can still find on the internet. One, directly applicable to CLIPS, the grand daddy of rulebased systems, is by Giarratano and Riley, "Expert Systems: Principles and and Programming" The books runs about US $125 for the new 4th edition, less for the 3rd in used condition and the Amazon link is
|
