@BEGIN(COMMENT) Talk I gave to the KA seminar, Winter Quarter, 1982
O. Overhead
  A. Informal talk
	1. Please interrupt 
	2. My ideas not clear, do help to hone them
  B. Handout deficient in scenario
	1. DOVER & SAIL down => gave penultimate version
	2. Not happy with it
  C. Short (overhead) talk - then comments
	Task itself (diagram)
	Why am I doing this
	How different
	This EMACS task
	Issues
	Future directions -- related and other
	Questions/discussion

-+-+-+-
KA task - see diagram
-+-+-+-

I. Why am I doing this?
  A. Why analogy?
	1. People use it all the time
	2. Powerful, and largely unexplored arena
	3. Personal interest -- indicates how people think
  B. Why learning/teaching?
	1. (Not Exploration, nor Rep'n)
	2. Ideal scenario - demonstrates process
		needed for other aspects.
  C. Why in Knowledge Acquisition?
	1. KA is bottleneck (current theme of HPP today)
	2. Validation - program will either run or not

II. Difference from others?
	(ie others have tried, why will I succeed?)
  A. Defn of Analogy - different from standard
  B. "Expert system" approach
	Domain = Heuristics, permitting different analogies

III. What is my approach?
  A. Work thru a few examples -- scenario which will be implemented
  B. Compiling thoughts on analogy - based on readings, ...

IV. What is my task?
  A. Task: Expand existing kb by adding new cases
  B. Useful front end -
	Other Examples:
	1. Add new disease to Mycin system, or new type of disease (urinary tract)
	2. Add new class of chemicals to Dendral
	3. Add new class of non-restriction enzymes to Molgen
	4. Add new programs to a (LISP) system
  C. Choose EMACS
	1. Only because I'm fluent in it
	2. Seemed to address relevant issues

V. Issues
  A. Student's initial knowledge
	1. If too much, no need for analogy, just inheritance
	  (Knowing the TECO code, why bother with analogy)
	2. If too little - incomprehensible
	  (Knowing only C-F, how to learn when to use M-F)
  B. Language for discussing analogy
	"A is like B, except ___"  - what goes there?
alter "known" facts
  C. Representation (related to A)
	1. If pos'n in form <word, char>, M-F trivial
	2. Otherwise may want to twiddle how facts are stored,
		to facilitate positioning
  D. Evaluation
	1. How to give a task, which doesn't already use answer
		[Consider "Move forward 3 words"]
	2. Possibly translation from E to EMACS?

VI. Future, directly related, work
  A. Finish a few scenarios - with EMACS
  B. Knowing EMACS, learn E
  C. Figure how to facilitate learning EMACS -
	1. What details had to be known
	2. What general types of things had to be known

VII. Other uses of Analogy
  A. Prediction -- perhaps AM-like Exploration
  B. Designing a new expert system -- Urinary tract from Mycin
  C. Tieresias like work - to figure what should be included, ...
	1. Weak, "empirical" use of semantics
	2. Stage II, in JSB's notation
  D. Using a different rep'n scheme (based on Analogy)

Comments/questions?


-------
-------

	Other things
VI. Other uses for analogy
* A. Designing a new Expert System - see JSB's Roget
* B. Using it in reasoning - to store the stuff (see TGD)
  C. For deciding how to represent the data - use of efficient
	hardware (eg employ {N, >} when dealing with instances of
	a linear ordering

VII. Properties of analogy

VIII. Discuss other cases of analogy
  A. Basically, different tact: analogy is a mapping between parts
  B. AI: Evans (A:B :: C: ?)
	 Kling (from groups to rings, ...)
  C. Program Synthesis: Moll/Urlich, ... [same idea]
  D. Philosophy - in terms of models (to understand a gestalt - see TGD)

Given *complete* Core Knowledge (eg first principles) of some domain
[eg facts about geology, and perhaps of the machines used for gathering data]
and facts about the specifics of an applications 
[here, data about the area of Saudi Arabia],
one can, in theory, produce an expert system for the task, given some
background in KEing.

However, that complete knowledge is a myth -- not only isn't it in any computer,
it also doesn't exist in the world itself.
Instead all we have are approximations -- in the form of 1/2-order theories,
or simplifications, or convenient (computationally efficient) models.
Hence we need something else for generating that expert system -- in the
form of ....

Now for the second such system -- here we can use the information painstakenly
produced during that first run -- by analogy.

Clearly the more of these "first principles" we have, the better we can
do when generating this second expert system.  In the absense of such
"deep(er)" knowledge, we can approximate it using that other systems --
essentially by copying the syntactic form (ala copy-and-edit scheme),
and maybe even a little bit more -- some of the content - c.f. Teriesias.

Note we are making a pretty strong meta-physical statement here, on the
similarity of the world: that things which share some features will share
others -- ie will satisfy the same "abstraction" or belong to the same
category.  (Ex: the same equations can be used for water flow or
electrons; or most arms are similar in function, independent of animal.)

@END(COMMENT)