Sunday, March 15, 2009
Ladder Logic Relays, Computers and MYCIN
If you have a sufficient mass of grey hair (or should that be "gray hair") you might remember "Ladder Logic" that was used for relays (those electro-mechanical monstrosities long, long ago) that were used in order to control machines in the dark ages before solid-state, ladder-logic that came to be called a PLC (Programmable Logic Controller) as popularized by both Square D and Allen-Bradley companies. A good view of this kind of logic is available on http://en.wikipedia.org/wiki/File:Ladder_diagram.png if you would like to see it.
Today's spreadsheets and decision tables are about as archaic as that particular concept and serve basically the same function.
IF this is true AND this is true AND this is true
THEN do that
How simple. Just like Ladder Logic. Yet, when many of these IF-THEN rules are strung together, how complex they swiftly seem to become. (Not really - but the human brain cannot usually do more than four or five things at once.) So, we bunch them up and put them in groups to make it simpler for the simple minds of humans. And we are left with simple solutions for simple minds.
But, think about it... What is happening is not very complex. To move just ONE STEP toward complexity consider what the group of MYCIN did; they implemented the concept of "probability of belief and disbelief" into the rule as well as some other statistical concepts and nobody has EVER gone back to that concept since it is too complicated for the simple minds of most "computer scientists" - those minions to whom we trust our programming today. If you can obtain the book by Buchannan and Shorliffe on the MYCIN project, GET IT, READ IT, UNDERSTAND IT !!
Then, after you read it and understand it, let's move beyond MYCIN. What if we integrate rulebased systems and neural nets? What if we integrate digital computers and analog computers? (Typical response to that one is: "Tried it. Failed. Not possible.") Cyc (Doug Lenat's huge, monstrous, immensely complicated 'real' KnowledgeBased System to be completed in 2025) may or may not be the answer, but surely we can move toward something in between the brain-dead, incredibly simplistic BRMS that are out there and immensely complex Cyc Project. Any ideas?