perm filename RENEW.83[S83,JMC] blob
sn#710678 filedate 1983-05-09 generic text, type C, neo UTF8
COMMENT ⊗ VALID 00004 PAGES
C REC PAGE DESCRIPTION
C00001 00001
C00002 00002 .<<renew.pro[s83,jmc] 1983 NSF renewal>>
C00004 00003 .every heading()
C00006 00004 .cb Progress report and plans
C00018 ENDMK
C⊗;
.<<renew.pro[s83,jmc] 1983 NSF renewal>>
.device XGP;
.turn on "α%";
.font 1 "baxl30"; font 2 "baxm30"; font 3 "baxb30"; font 4 "baxs30";
.font 5 "metl";
. << output a ligature & restore fonts >>
.recursive macro lig(A); ⊂start f_f←thisfont; ("%*%4αA%*%"&f_f) end⊃;
.
.AT "ff" ⊂ IF THISFONT=1 THEN lig(@) ELSE if thisfont=2 then lig(P)
. else if thisfont=3 then lig("`") else "fαf" ⊃;
.AT "ffi" ⊂ IF THISFONT=1 THEN lig(A) ELSE if thisfont=2 then lig(Q)
. else if thisfont=3 then lig(a) else "fαfαi" ⊃;
.AT "ffl" ⊂ IF THISFONT=1 THEN lig(B) ELSE if thisfont=2 then lig(R)
. else if thisfont=3 then lig("b") else "fαfαl" ⊃;
.<<
.AT "fi" ⊂ IF THISFONT=1 THEN lig(C) ELSE if thisfont=2 then lig(S)
. else if thisfont=3 then lig(c) else "fαi" ⊃;
.>>
.AT "fl" ⊂ IF THISFONT=1 THEN lig(D) ELSE if thisfont=2 then lig(T)
. else if thisfont=3 then lig(d) else "fαl" ⊃;
.AT "--" ⊂ IF THISFONT=1 THEN lig(E) ELSE if thisfont=2 then lig(U)
. else if thisfont=3 then lig(e) else "α-α-" ⊃;
.sides←1;
.require "twocol.pub[sub,sys]" source_file;
.macro sskip ⊂ skip to column 1⊃; << section skip control >>
.every heading();
.onecol;
.begin "cover"
.center;
%5Computer Science Department
Stanford University
%51983 May 5
%1Proposal to
%5National Science Foundation
%1for continuation of
%5Basic Research in Artificial Intelligence
.skip 2
John McCarthy
Professor of Computer Science
Principal Investigator
.skip 10;
%3Abstract
.skip
.begin fill;
%1This is a request for continuation of NSF Grant
MCS-8104877 (Basic Research in Artificial Intelligence).
The renewal will
continue for the third year the activities proposed for three years in
our proposal of August 1980. The amount requested is α$111,324.
.end
.end "cover"
.every heading(,"%3Basic Research in Artificial Intelligence",{page});
.count page;
.if xcribl then twocol;
.cb Progress report and plans
.fill adjust; select 1;
Since the last report McCarthy has made progress as follows:
1. The work on the "Kowalski doctrine" led to a concept of
introspection in Prolog and related languages. There turned out
to be at least two kinds of introspection - syntactic and semantic.
The "postponement heuristic" is an example of syntactic introspection.
We imagine the system looking at its program - or the program
looking at itself - and deciding that the program will do less searching
if it is re-ordered. The "Kempe heuristic" involves semantic
introspection. The program is running, a goal fails, and the program
decides to explore certain kinds of changes of the values of the variables.
Writing this in Prolog required that the interpreter be able to
go into an "introspective mode" in which the state of the search
and proof trees are data available to the program. While McCarthy
was visiting Imperial College London, Peter Szeredi of the Hungarian
Academy of Sciences, also visiting at that time, wrote a Prolog
interpreter embodying some of the proposed features.
It isn't clear that using an interpreter with introspective
features will actually be useful. While the Kempe algorithm for the
four color problem admits an elegant formulation as an introspective
problem, the straightforward program seems just as easy to write.
There is a correspondence between McCarthy's ideas on introspective
Prolog and Brian Smith's work on "reflective LISP." Smith has been
planning to look at Prolog as providing a more natural scope for
his reflective reasoning. In some minor respects, McCarthy and
Szeredi got there first.
2. McCarthy has been applying circumscription to formalizing
various AI problems. This has led to an extended version of circumscription
which allows circumscriptions to be assigned priorities relative
to one another. This seems to be required to express human common
sense knowledge.
Non-monotonic reasoning, for which circumscription is one
of the few still living candidates, turns out to have many aspects.
Different formulators of non-monotonic reasoning systems have been
looking at different parts of an elephant.
One straightforward use is as a communication convention.
For example, if I ask you to make a bird cage and you make it without
a top, I have a right to complain if I fear that my bird may fly out.
On the other hand, if you make it with a top and my bird is a penguin,
I do not have a right to complain that you have wasted material unless
I have given some indication that my bird can't fly. We can regard
it as a communication convention, and one nicely expressed using
the circumscription formalism, that if a bird under discussion
can fly, this need not be mentioned. Whereas, if the bird cannot
fly, and it is relevant, the information must be provided.
This convention arises because most birds can fly, but it
seems more straightforward to regard it as a convention.
Data bases can have similar default conventions about what
inexplicit information may be inferred unless there is specific
information to the contrary. Expressing such rules formally in a "preamble"
to the data base is more flexible and powerful than simply building
a closed world assumption into the program.
Jon Doyle's default policy conventions (The meetings will
be held on Wednesday unless there is reason to do otherwise.) can
be treated similarly.
All this is rather different from the examples that
motivated McCarthy's original introduction of circumscription.
It is to be conjectured that there are no relevant material
objects or other circumstances other than those whose existence
follows from the known facts. The point is that conventions
and rules of conjecture are distinct kinds of entities.
3. McCarthy has made further progress on his inventory
of common sense concepts. Yoram Moses, a research assistant,
has succeeded in formalizing a simple case of concurrent events.
A new paper on circumscription will be submitted for publication
shortly.