rules: model of huamn cognition
	short term memoy
	long term memory
	experts and experts because tof the "pwoer" patterns they have
		laboruiously leanred and added to their long term
		memory (100,000+ patterns)
	novices are novices because they clog short term memory (which
		is really tiny, say 7+- 2) with irrelevant goals
		(and experts dodge that since thei LTM patters tell them
		what is relevant using their magic feature extractors)
	feature extrators
		biases telescopes on the world
		only see what is important
		e.g. chess boards
	
decades of work in mapping this cognitive sphycoholoy insght into AI programs
	utilmately, other methods worked better on computers
		e.g. stochastic search
	but very useful tool for structuring a domain
	
maintenace adavnatneces (subbospedly)
engineering advantages (uniformity)
	rules are rules
	feature extractors are rules that run before the other rules
	conflict resolution operators are rules that select which
		rules to fire

scale up

computational problems
		match-select-act: 80% of the time in match
		global search through all possible assertions : slow

maintenance probelms
		XXX what was XCON
		XXX why was XCON important 
			- the AI bubble
		R1: KE scucess of the early 1980s became the maintenance
			nightmare of the mid-1980s.
		What works for 400 rules does not work for 8000
		At its peak, DEC was claiming it saved the, $20M/year 
			but if any of XCON's 20 deveveloper's left,
			ouch!
		
solving the computational problem
	throw away rules
		go to simpler represntations
			e.g. state machines (used in gaming, next lecture)
	the rete netowrk (see OPS5, ART, JESS)
		XXX
		http://upload.wikimedia.org/wikipedia/commons/thumb/9/92/Rete.JPG/800px-Rete.JPG
	divide and conqueor
		divide global rule space into rule groups
			which actually leads into the maintaince problem
			
solving the maintenace problems			
	divide
		how1: using domain knowledge
			e.g. frames for the big picture
				frame slots have rules
			Q: what frames to use
			A: domain knowledge
		how2: using background knowledge of problem
			solving types (Clancey, then KADS)
		how3: using background knowledge of RBS
				TAQL
	patch in context
		never re-organize, patch in context of last error
		people make it up as they go along, why not machines?
		
who won?
	problem solving types? not so much
		that all went away
	patch in context