\init{
\input cmpdes
\input ammacs
\def\draft{T}
\titlepage
\newpage
}
\partbegin{Part Two}
\parttitle{TEIRESIAS: APPLICATIONS OF META-LEVEL KNOWLEDGE}
\rjustline{{\:A TEIRESIAS:}}
\rjustline{{\:  A Applications of}}
\rjustline{{\:A   Meta-Level Knowledge}}
\runningrighthead{INTRODUCTION}
\vskip 8pc

\tenpoint

\noindent
The  creation and  management of large  knowledge bases  has become a
central  problem  of  artificial  intelligence  research.   This  has
occurred largely as a result of two recent trends: an emphasis on the
use of  large stores of domain specific knowledge  as a base for high
performance programs, and a concentration on problems taken from real
world settings.   Both of these mean an  emphasis on the accumulation
and  management of large  collections of knowledge.   In many systems
embodying these  trends much  time  has been  spent on  building  and
maintaining  such  knowledge   bases.   Yet  there  has  been  little
discussion or analysis of the concomitant problems.

This section  of  the book  attempts to  define  some of  the  issues
involved,  and explores  steps taken toward  solving a  number of the
problems encountered.  It describes the organization, implementation,
and  operation  of  a program  called  TEIRESIAS,  designed  to  make
possible the interactive transfer of expertise from a human expert to
the  knowledge base  of  a  high  performance program,  in  a  dialog
conducted in a restricted subset of natural language.

The two major goals set were (i) to make it possible for an expert in
the  domain  of  application to  "educate"  the  performance  program
directly,  and (ii)  to ease the  task of  assembling and maintaining
large amounts of knowledge.

The central theme of this work  is the exploration and use of what we
have labelled  meta level  knowledge.  This  takes several  different
forms as  its use  is explored,  but can  be summed  up generally  as
"knowing what  you know".  It makes possible a  system which has both
the capacity  to  use  its knowledge  directly,  and the  ability  to
examine it, abstract it, and direct its application.

We report  here  on the  full extent  of  the capabilities  it  makes
possible,  and  document  cases   where  its  lack  has  resulted  in
significant difficulties.  We describe efforts to enable a program to
explain  its actions, by giving  it a model of  its control structure
and an understanding of  its representations.  We document the use of
abstracted  models   of  knowledge  (rule  models)   as  a  guide  to
acquisition  and demonstrate the  utility of describing  to a program
the structure  of its representations.  Finally,  we describe the use
of strategies  in the  form of  meta rules,  which contain  knowledge
about the use of knowledge.

\worldend