-- Messages from file: [SUMEX-AIM]<LENAT>MESSAGE.TXT;1
		 -- Monday, June 18, 1979 13:05:13 --

Mail from RAND-UNIX rcvd at 18-Jun-79 1057-PDT
From: Rick at Rand-Unix
Date: 18 Jun 1979 at 1054-PDT
To: Don at Rand-Unix, lenat at Aim
cc: Rick at Rand-Unix, lenat at Sail
Subject:WORKSHOP OVERVIEW

Guys--
	Here's my statement of what the workshop on expert systems should
be.  I think it's pretty interesting, and hope you agree.  If so, why don't
the two of you take a turn, and convert it into a real letter with a
set of invitees.  When I return July 1, I'll edit that and send it to NSF
and to the invitees simultaneously, if we'll still in general agreement.
I realize that the program I've sketched is somewhat broad and ambitious,
but I could get genuinely interested in it for the kinds of real questions
it raises.

	If you need to talk to me, I'll be calling Jackie daily this week.

			I hope you like it,

				Rick

==============================================================================

EXPERT SYSTEMS WORKSHOP
Tentative Plan (June  18, 1979)
Author: Hayes-Roth



	Date: June 1980 (gets us away from the fog in l.a.)
	Place: The big island or some nice spot in northern california prior to
		stanford dcs's big show; i'm ambivalent
	Purposes:

		0.  Define and justify "expert"

		1.  Serve the needs of ai researchers who are trying to
			build and apply expert systems

		2.  Identify what kinds of research ought to be done

		3.  Identify potential applications or spin-offs

		4.  Develop a set of applications rules, such as:
			use this language, machine, and programming tool,
			follow this recipe.

Format:

	One set of papers/panels on each of the 5 general areas above


0.  Define and justify "expert"

	Issues here include:

		What's a human expert?
		What's a machine expert?
			What are all the machine experts?
		What's the difference?
		To what extent is knowledge inspectability crucial, as
	opposed to knowledge compilation or speed?


1.  Serve the needs of ai researchers who are trying to
			build and apply expert systems

	Issues here include:

		What's hard about building expert systems?
			Knowledge representation, acquisition,compilation,
			integration, refinement, testing.
		Is this an AI problem, or is it basically a different problem
			in every domain?  If so, why treat it as an AI
			problem per se?
		Why can't all expert systems be recast as problem solvers and
			then use general purpose systems?
		What are the current technical bottlenecks and impending
			big opportunities?


2.  Identify what kinds of research ought to be done

	Issues here include:

		Some well defined problems that play key roles in
		expert system building:
			e.g. operationalization (Mostow&Hayes-Roth)
			     knowledge base maintenance (stefik,davis)
			     explanation
			     focus & resource allocation
		Some particular performance analysis problems:
			apportionment of credit in heuristic systems
			meta-planning or meta-problem solving
		Comparing heuristic methods to analytical ones:
			What's the advantages of each?
			Why can't they be coupled?


3.  Identify potential applications or spin-offs

	Examples of promising areas:
		Mathematics--eurisko,macsyma
		medicine
		chemistry
		military
		education


4.  Develop a set of applications rules, such as:
	use this language, machine, and programming tool,
	follow this recipe.

	Languages:  Which language is numero uno, for whom and why?
		Why interlisp?
		Why Rosie?
		Why small talk vs. director vs. new APM?
		Why KRL vs. Units vs. ...?

	Problem solving systems:  which & why?
		EMYCIN vs. GPS vs. opportunistic planning vs. HearsayIII vs. ...

	Machines:
		One big hairy address space vs. distribution?
		How to exploit cheap new machines?

	Recipe:
		What, really, is the expert system building recipe?