perm filename RAND.ABS[TLK,DBL] blob
sn#197259 filedate 1976-01-19 generic text, type C, neo UTF8
COMMENT ⊗ VALID 00003 PAGES
C REC PAGE DESCRIPTION
C00001 00001
C00002 00002 .COMMENT !XGPCOMMANDS←"/PMAR=2200"
C00004 00003 We often face the difficult task of formulating new research problems
C00009 ENDMK
C⊗;
.COMMENT !XGPCOMMANDS←"/PMAR=2200";
.DEVICE XGP
.FONT 1 "BASL30"
.FONT 2 "BASB30"
.FONT 4 "BASI30"
.FONT 5 "BDR40"
.FONT 6 "NGR25"
.FONT 7 "NGR20"
.FONT 8 "SUP"
.FONT 9 "SUB"
.TURN ON "↑α↓_π[]{"
.TURN ON "⊗" FOR "%"
.TURN ON "@" FOR "%"
.PAGE FRAME 54 HIGH 83 WIDE
.AREA TEXT LINES 1 TO 53
.COUNT PAGE PRINTING "1"
.TABBREAK
.ODDLEFTBORDER←EVENLEFTBORDER←1000
.AT "ffi" ⊂ IF THISFONT ≤ 4 THEN "≠" ELSE "fαfαi" ⊃;
.AT "ffl" ⊂ IF THISFONT ≤ 4 THEN "α∞" ELSE "fαfαl" ⊃;
.AT "ff" ⊂ IF THISFONT ≤ 4 THEN "≥" ELSE "fαf" ⊃;
.AT "fi" ⊂ IF THISFONT ≤ 4 THEN "α≡" ELSE "fαi" ⊃;
.AT "fl" ⊂ IF THISFONT ≤ 4 THEN "∨" ELSE "fαl" ⊃;
.SELECT 1
.MACRO B ⊂ BEGIN VERBATIM GROUP ⊃
.MACRO E ⊂ APART END ⊃
.MACRO FAD ⊂ FILL ADJUST DOUBLE SPACE PREFACE 2 ⊃
.MACRO FAS ⊂ FILL ADJUST SINGLE SPACE PREFACE 1 ⊃
.FAS
.COMPACT
.SELECT 1
.PORTION THESIS
.PAGE←0
.NEXT PAGE
.INDENT 0
.TURN OFF "{∞→}"
.BEGIN CENTER
⊗2SEMINAR:
Friday, February 13, 1976
10:00 a.m.⊗*
⊗5↓_Automating the Discovery of Mathematical Concepts_↓⊗*
⊗2Douglas B. Lenat⊗*
Artificial Intelligence Lab
Stanford University
.END
We often face the difficult task of formulating new research problems
which must be both soluble and nontrivial. It's usually easier to
tackle a specific given problem than to propose interesting yet
managable new questions to investigate. For example, contrast
⊗4playing⊗* Monopoly with the more difficult task of ⊗4inventing⊗*
new games of the same quality. Can such "originality" be mechanized?
My thesis -- and this seminar -- describe one approach to partially
automating the development of new mathematical concepts. First, we
shall consider how to ⊗4explain⊗* a discovery D, how to
systematically analyze it until it seems obvious. This is
accomplished by constructing a chain of minor discoveries, stretching
from D back to previously known concepts. By inverting this
analytical procedure, we obtain a simple scheme for ⊗4generating⊗*
new discoveries, for synthesizing new theories. To combat the
combinatorially explosive nature of this process, heuristic rules of
thumb are used to prune away unpromising lines of investigation.
An experimental interactive LISP program called ⊗2AM⊗* has been
developed, which attempts to do such simple concept-growing. That is,
AM carries out some of the activities involved in simple mathematical
research: noticing obvious relationships in empirical data,
formulating new definitions out of existing ones, proposing some
plausible conjectures, and estimating the potential worth of each new
concept. It was necessary to devise a rudimentary calculus of
"interestingness" which enabled AM to choose which activity to work
on at each moment. AM also relies upon a human in the role of
co-researcher.
AM was initially given some scanty information about very elementary
pre-numerical notions (sets, relations, composition, set-equality,
and a hundred others), plus a few hundred guiding heuristics of
varying generality. Using these, AM developed several well-known
concepts (including: numbers, arithmetic, prime numbers, and unique
factorization). By interacting heavily with a human, AM was able to
motivate one original piece of math research (maximally-divisible
numbers).
After explaining the workings of AM, we can discuss such issues as
.BEGIN PREFACE 0 INDENT 4,8,0
(i) Choice of task domain: Why mathematics? Suitability of various
other sciences.
(ii) Experiments one can perform on AM: What do we hope to learn?
(iii) The role of the human user: spectator ⊗4vs⊗* co-researcher.
(iv) How can one judge the performance of a concept-proposer which
has no fixed goal?
(v) What kinds of discoveries are most difficult to mechanize?
(vi) Assimilating new information into data bases: global updating
⊗4vs⊗* living with contradiction.
.END
.SKIP TO COLUMN 1