/LMAR=0/XLINE=3/FONT#0=BASL30/FONT#1=BASB30/FONT#3=BASI30/FONT#4=NGR40/FONT#5=NGR25/FONT#6=NGR20/FONT#7=GRFX35/FONT#8=FIX20/TMAR=30/PMAR=2150/BMAR=20







␈↓ ↓¬␈↓∧␈↓ ↓e␈↓ ∧∞DUPLICATION OF HUMAN ACTIONS
␈↓ ↓¬␈↓∧␈↓ ↓e␈↓ α"BY AN INTERACTING COMMUNITY OF KNOWLEDGE MODULES



␈↓ ↓¬␈↓↓␈↓ ↓e␈↓ ¬JDouglas B. Lenat

␈↓ ↓¬␈↓↓␈↓ ↓e␈↓ ∧↑Arti≡cial Intelligence Laboratory
␈↓ ↓¬␈↓↓␈↓ ↓e␈↓ ¬6Stanford University





␈↓"β␈↓ ↓¬␈↓␈↓ αU1. Theory .......................................................................................................................................␈↓ 
 1
␈↓"β␈↓ ↓¬␈↓␈↓ αU␈↓ε               Origins of the Constraints␈↓ ...............................................................................................␈↓ 
 1
␈↓"β␈↓ ↓¬␈↓␈↓ αU␈↓ε               How to Build Such a System␈↓ ............................................................................................␈↓ 
 3
␈↓"β␈↓ ↓¬␈↓␈↓ αU2. Application to Automatic Programming .................................................................␈↓ 
 3
␈↓"β␈↓ ↓¬␈↓␈↓ αU␈↓ε               Analogy: Cooperating Human Experts␈↓ ..............................................................................␈↓ 
 3
␈↓"β␈↓ ↓¬␈↓␈↓ αU␈↓ε               Details of the Experimental System␈↓ .................................................................................␈↓ 
 5
␈↓"β␈↓ ↓¬␈↓␈↓ αU␈↓ε               Results␈↓ .............................................................................................................................␈↓ 
 6
␈↓"β␈↓ ↓¬␈↓␈↓ αU␈↓ε               Conclusions␈↓ ......................................................................................................................␈↓ 
 8
␈↓"β␈↓ ↓¬␈↓␈↓ αU3. Application to Mathematics Research ......................................................................␈↓ 
 9
␈↓"β␈↓ ↓¬␈↓␈↓ αU␈↓ε               Correcting difficulties of the Automatic Programming system␈↓ ........................................␈↓ 
 9
␈↓"β␈↓ ↓¬␈↓␈↓ αU␈↓ε               Characteristics of a Proposed System␈↓ ..........................................................................␈↓ 	⎇ 10
␈↓"β␈↓ ↓¬␈↓␈↓ αU␈↓ε               Projected Results␈↓ ..........................................................................................................␈↓ 	⎇ 12
␈↓"β␈↓ ↓¬␈↓␈↓ αU␈↓ε               Synthesis␈↓ .......................................................................................................................␈↓ 	⎇ 12
␈↓"β␈↓ ↓¬␈↓␈↓ αU References
␈↓"β␈↓ ↓¬␈↓␈↓ αU␈↓ε               Acknowledgements␈↓







␈↓"β␈↓ ↓¬␈↓ε␈↓ αU␈↓ ¬[Mr. Douglas B. Lenat
␈↓"β␈↓ ↓¬␈↓ε␈↓ αU␈↓ ¬&Artificial Intelligence Laboratory
␈↓"β␈↓ ↓¬␈↓ε␈↓ αU␈↓ ¬.Computer Science Department
␈↓"β␈↓ ↓¬␈↓ε␈↓ αU␈↓ ¬`Stanford University
␈↓"β␈↓ ↓¬␈↓ε␈↓ αU␈↓ ¬≥Stanford, California 94305, U.S.A.













␈↓"β␈↓ ↓¬␈↓ε␈↓ αU␈↓ ¬uApril 15, 1975
␈↓ ↓¬␈↓␈↓εDouglas B. Lenat␈↓ ¬;Interacting Knowledge Modules␈↓ 
∩April 15, 1975      page 1␈↓


␈↓ ↓¬␈↓␈↓ ¬w␈↓∧␈↓&1. Theory␈↓)αβ␈↓


␈↓ ↓¬␈↓␈↓ ↓e␈↓↓␈↓&1.1. Origins of the Constraints␈↓)αβ␈↓


␈↓ ↓¬␈↓Despite␈α⊂the␈α⊂awesome␈α⊂physical␈α⊂forces␈α⊂a≥ecting␈α⊂Earth␈α⊂(weather,␈α⊂gravitation,...␈α⊂),␈α⊂it␈α⊂is␈α⊂␈↓βbiological␈↓␈α⊂entities␈α⊂who

␈↓ ↓¬␈↓dominate␈α∩our␈α⊃planet.␈α∩ We␈α⊃shall␈α∩assume␈α⊃that␈α∩life␈α⊃"works",␈α∩extract␈α⊃some␈α∩obvious␈α⊃characteristics␈α∩of␈α⊃living

␈↓ ↓¬␈↓organisms (especially of Man), and then design a program which incorporates them.


␈↓ ↓¬␈↓Perhaps␈αthe␈αmost␈αobvious␈αdistinction␈αbetween,␈αe.g.,␈αWeather␈αand␈αAnimals,␈αis␈αthat␈αof␈α␈↓βdiscreteness␈↓.␈αYet␈αbesides

␈↓ ↓¬␈↓having␈αdistinguishable␈α
boundaries,␈αeach␈αspecies␈α
has␈αa␈αspeci≡c␈α
␈↓βanatomy␈↓.␈αThat␈α
is,␈αeach␈αsuch␈α
creature␈αis␈αalike␈α
in

␈↓ ↓¬␈↓many␈α⊂important␈α⊂functional,␈α⊂physiological,␈α⊂structural,␈α∂and␈α⊂psychological␈α⊂respects.␈α⊂ More␈α⊂than␈α⊂uniformity␈α∂is

␈↓ ↓¬␈↓required:␈α␈↓βIndividual␈αabilities␈↓␈αmust␈αbe␈α
recognized.␈αDespite␈αtheir␈αsimilarity,␈αeach␈α
member␈αof␈αa␈αspecies␈αis␈α
slightly

␈↓ ↓¬␈↓di≥erent␈α
in␈α∞his␈α
perception␈α∞of␈α
features␈α
of␈α∞his␈α
environment,␈α∞in␈α
his␈α
responses␈α∞to␈α
those␈α∞features,␈α
in␈α∞his␈α
drives,

␈↓ ↓¬␈↓and␈α
in␈α
his␈α
capabilites.␈α
In␈αmany␈α
species,␈α
including␈α
humans,␈α
this␈α
has␈αbeen␈α
carried␈α
to␈α
the␈α
extreme␈αof␈α
␈↓βprofessions.␈↓

␈↓ ↓¬␈↓A collection of diverse experts is more viable than a similar collection of generalists.


␈↓ ↓¬␈↓So␈αwe␈αlimit␈αour␈α
attention␈αto␈αa␈αcommunity␈α
of␈αdiscrete␈αentities,␈αsensitive␈αto␈α
the␈αworld␈αaround␈αthem,␈α
each␈αwith

␈↓ ↓¬␈↓his␈α∩own␈α⊃distinct␈α∩expertise,␈α∩yet␈α⊃all␈α∩anatomically␈α⊃and␈α∩psychologically␈α∩similar.␈α⊃ How␈α∩do␈α⊃they␈α∩␈↓βinteract␈↓␈α∩in␈α⊃a

␈↓ ↓¬␈↓meaningful,␈α∞productive␈α∞manner?␈α∞Biology␈α∞would␈α∞lead␈α∞us␈α∞to␈α∞hypothesize␈α∞␈↓βcompetition␈↓,␈α∞but␈α∞perhaps␈α∞the␈α∂key␈α∞is

␈↓ ↓¬␈↓␈↓βcooperation.␈↓␈αSpecialization␈α
is␈αof␈αadvantage␈α
only␈αif␈α
each␈αspecialist␈αis␈α
willing␈αto␈α
use␈αhis␈αtalent␈α
to␈αserve␈αthe␈α
entire

␈↓ ↓¬␈↓community.␈αIn␈αpart␈αthis␈αmay␈αbe␈αinstinctive,␈αbut␈αmuch␈αof␈αMan's␈αsupremacy␈αstems␈αfrom␈αlearning␈αto␈αcooperate.

␈↓ ↓¬␈↓The␈α∃institutions␈α∃of␈α∃the␈α∃law,␈α∃education,␈α∃and␈α∃government␈α∃universally␈α∃re∨ect␈α∃(distortions␈α∃of)␈α∃this␈α∃ideal.

␈↓ ↓¬␈↓Advancement ␈↓βwithin␈↓ a society may be by competition, but advancement ␈↓βof␈↓ a society necessitiates synergy.


␈↓ ↓¬␈↓But␈α∞how␈α∞do␈α∞people␈α∞help␈α
each␈α∞other;␈α∞what␈α∞is␈α∞the␈α∞"mechanism"?␈α
Perhaps␈α∞it␈α∞lies␈α∞in␈α∞␈↓βcommunication␈↓.␈α∞This␈α
has

␈↓ ↓¬␈↓three␈α∩aspects:␈α∩(i)␈α∩broadcasting␈α⊃your␈α∩own␈α∩needs␈α∩to␈α⊃the␈α∩specialists␈α∩who␈α∩might␈α⊃help␈α∩you,␈α∩or␈α∩to␈α∩the␈α⊃entire

␈↓ ↓¬␈↓community;␈α∞(ii)␈α∞recognizing␈α∞when␈α∂you␈α∞are␈α∞relevant␈α∞to␈α∞satisfying␈α∂someone␈α∞else's␈α∞speci≡c␈α∞need;␈α∂(iii)␈α∞satisfying

␈↓ ↓¬␈↓that␈αneed,␈αoften␈αby␈αposing␈αnew␈αsub-requests␈αto␈αthe␈αcommunity.␈α Besides␈αcommunication,␈αanother␈αconstructive

␈↓ ↓¬␈↓process␈α
of␈α
living␈α∞entities␈α
is␈α
␈↓βreproduction.␈↓␈α
Heredity␈α∞and␈α
teaching␈α
ensure␈α
that␈α∞the␈α
o≥spring␈α
will␈α∞be␈α
somewhat

␈↓ ↓¬␈↓similar to his parents, yet not identical.


␈↓ ↓¬␈↓These␈αcharacteristics␈αare␈αnow␈α␈↓βpostulated␈↓␈αas␈αconstraints␈αon␈αthe␈αcomputer␈αsimulation␈αwe␈αwish␈αto␈αconstruct.␈α
The
␈↓ ↓¬␈↓␈↓εDouglas B. Lenat␈↓ ¬;Interacting Knowledge Modules␈↓ 
∩April 15, 1975      page 2␈↓


␈↓ ↓¬␈↓program␈α∂will␈α⊂be␈α∂made␈α⊂up␈α∂of␈α∂discrete␈α⊂modules,␈α∂which␈α⊂we␈α∂shall␈α∂call␈α⊂BEINGs,␈α∂each␈α⊂possessing␈α∂a␈α⊂clump␈α∂of

␈↓ ↓¬␈↓specialized␈αknowledge.␈α Each␈αBEING␈α
shall␈αhave␈αthe␈αsame␈αinternal␈α
structure,␈αcan␈αrespond␈αto␈αthe␈α
same␈αkinds

␈↓ ↓¬␈↓of␈α⊃questions␈α⊃(though␈α⊃with␈α⊃di≥erent␈α⊃answers␈α⊃in␈α⊃each␈α⊂case).␈α⊃The␈α⊃only␈α⊃things␈α⊃that␈α⊃a␈α⊃BEING␈α⊃can␈α⊃do␈α⊂are:

␈↓ ↓¬␈↓broadcast␈α
a␈α
plea␈α
for␈α
assistance,␈α
recognize␈α
his␈α
own␈α
ability␈α
to␈α
help␈α
in␈α
a␈α
certain␈α
situation,␈α
provide␈α
the␈αneeded

␈↓ ↓¬␈↓answer␈α⊂in␈α⊂such␈α⊂cases,␈α⊂combine␈α⊂with␈α⊂0,1,␈α⊂or␈α⊂2␈α⊂other␈α⊂BEINGs␈α⊂to␈α⊂produce␈α⊂a␈α⊂new␈α⊂BEING,␈α⊂and␈α⊃teach␈α⊂that

␈↓ ↓¬␈↓o≥spring some of what it knows.


␈↓ ↓¬␈↓The␈αreader␈αmay␈αprotest␈αthat␈αmuch␈αhas␈αbeen␈αignored:␈αprimitive␈αdrives,␈αinstincts,␈αemotions,␈αaesthetics,␈αethics,...

␈↓ ↓¬␈↓Some␈α
of␈α
these␈α
will␈α
be␈α
dealt␈α
with␈α
in␈α
Section␈α
3,␈α
when␈α
we␈α
discuss␈α
design␈α
issues␈α
for␈α
a␈α
system␈α
which␈α∞can␈α
make

␈↓ ↓¬␈↓simple␈α∞mathematical␈α∞discoveries.␈α
 Such␈α∞a␈α∞system␈α∞must␈α
have␈α∞a␈α∞sense␈α
of␈α∞aesthetic␈α∞beauty,␈α∞harmony,␈α
potential

␈↓ ↓¬␈↓utility,␈αinterestingness,␈αa␈αsense␈αof␈αintuition,␈αa␈αpriority␈αqueue␈αof␈αinstincts.␈α It␈αis␈αnot␈αfelt␈αthat␈αthe␈αcomplexity␈αof

␈↓ ↓¬␈↓human␈αemotions␈α
is␈αprerequisite␈α
to␈αintelligence;␈α
BEINGs␈αneed␈α
not␈αsimulate␈α
emotions.␈α No␈α
case␈αshall␈αbe␈α
argued

␈↓ ↓¬␈↓in␈αfavor␈αof␈α
these␈αprejudices;␈αrather,␈α
we␈αshall␈αtest␈αthem,␈α
observing␈αwhat␈αhappens␈α
in␈αa␈αsystem␈α
which␈αassumes

␈↓ ↓¬␈↓them. This is the subject of Section 2. Before that discussion, let us formally de≡ne some of our terms.


␈↓ ↓¬␈↓A␈α∩␈↓βsystem␈↓␈α⊃of␈α∩BEINGs,␈α⊃also␈α∩called␈α⊃a␈α∩␈↓βpool␈↓,␈α∩is␈α⊃a␈α∩collection␈α⊃of␈α∩knowledge␈α⊃modules,␈α∩plus␈α∩auxilliary␈α⊃support

␈↓ ↓¬␈↓functions␈α(like␈α"Pattern-match",␈α"Sort").␈αEach␈α␈↓βBEING␈↓␈αis␈αnothing␈αmore␈αthan␈αa␈αcollection␈αof␈αabout␈αthirty␈αparts.

␈↓ ↓¬␈↓Each␈α␈↓βpart␈↓␈αof␈αBEING␈αB␈αis␈αan␈α
ordered␈αpair␈α␈↓↓(q,a)␈↓,␈αwhere␈α␈↓↓␈↓&q␈↓)αβ␈↓␈αis␈αan␈α
abbreviated␈αname␈αof␈αa␈αquestion,␈αand␈α␈↓↓␈↓&a␈↓)αβ␈↓␈α
is␈αa

␈↓ ↓¬␈↓program␈αwhich␈α
can␈αbe␈αrun␈α
to␈αprovide␈αB's␈α
answer␈αto␈α␈↓↓␈↓&q␈↓)αβ␈↓.␈α
In␈αthe␈αcourse␈α
of␈αexecuting␈α␈↓↓␈↓&a␈↓)αβ␈↓,␈α
new␈αquestions␈α
may␈αbe

␈↓ ↓¬␈↓raised,␈α∞new␈α∞BEINGs␈α∞may␈α∞be␈α∞created,␈α∞parts␈α
of␈α∞existing␈α∞BEINGs␈α∞may␈α∞be␈α∞changed,␈α∞and␈α∞messages␈α
(hopefully

␈↓ ↓¬␈↓including the desired answer) may be passed.


␈↓ ↓¬␈↓Once␈α
again:␈α
the␈α
concept␈α
of␈α
a␈α
pool␈α
of␈α
BEINGs␈α
is␈α
that␈α
many␈α
specialized␈α
entities␈α
coexist,␈α
each␈α
having␈α
a␈α
complex

␈↓ ↓¬␈↓structure,␈α∂but␈α∂that␈α∞structure␈α∂does␈α∂not␈α∞vary␈α∂from␈α∂entity␈α∂to␈α∞entity.␈α∂ This␈α∂idea␈α∞has␈α∂analogues␈α∂in␈α∂many␈α∞≡elds:

␈↓ ↓¬␈↓transactional analysis in psychology, anatomy in medicine, modular design in architechture.


␈↓ ↓¬␈↓It␈αseems␈α
probable␈αthat␈αthe␈α
set␈α␈↓↓Q␈α=␈α
{q}␈↓␈αmight␈αneed␈α
to␈αbe␈α
immense,␈αto␈αcover␈α
all␈αthe␈αkinds␈α
of␈αqueries␈αany␈α
expert

␈↓ ↓¬␈↓might␈αever␈αwant␈αto␈αput␈αto␈αanother␈αexpert.␈αIn␈αfact,␈αin␈αorder␈αto␈αmake␈α||␈↓↓Q␈↓||␈αmanageably␈αsmall,␈αit␈αis␈αnecessary␈αto

␈↓ ↓¬␈↓restrict␈α
the␈α
task␈α
domain␈α
of␈α
the␈α
BEINGs␈α
system.␈α
For␈αexample,␈α
a␈α
set␈α
Q␈↓#vc␈↓#␈↓#vf␈↓#␈α
of␈α
29␈α
parts␈α
can␈α
be␈α
found␈αwhich␈α
su≠ce

␈↓ ↓¬␈↓to␈αdiscuss␈α
concept␈αformation,␈α
and␈αa␈αset␈α
Q␈↓#vmath␈↓#␈αof␈α
25␈αwhich␈αsu≠ce␈α
to␈αdiscuss␈α
mathematical␈αresearch,␈αbut␈α
those

␈↓ ↓¬␈↓two sets are almost disjoint.
␈↓ ↓¬␈↓␈↓εDouglas B. Lenat␈↓ ¬;Interacting Knowledge Modules␈↓ 
∩April 15, 1975      page 3␈↓


␈↓ ↓¬␈↓␈↓ ↓e␈↓↓␈↓&1.2. How to Build Such a System␈↓)αβ␈↓


␈↓ ↓¬␈↓How␈αcan␈α
we␈αtest␈α
out␈αthe␈α
feasability␈αof␈αthis␈α
design␈αscheme␈α
for␈αrepresenting␈α
knowledge?␈α We␈α
must␈αbuild␈αa␈α
pool

␈↓ ↓¬␈↓of␈α
BEINGs,␈α
a␈α
modular␈α
program␈α
which␈αwill␈α
interact␈α
with␈α
a␈α
human␈α
user␈αand␈α
do␈α
some␈α
speci≡c␈α
task␈α
(like␈αwrite␈α
a

␈↓ ↓¬␈↓certain␈α∞computer␈α∞program,␈α∞or␈α∞propose␈α∞some␈α∞speci≡c␈α∞mathematical␈α∞conjecture␈α∞and␈α∞then␈α∞≡nd␈α∞a␈α∞proof␈α∞for␈α∞it).

␈↓ ↓¬␈↓Recasting the idea into operational terms, we arrive at this procedure for constructing a pool of BEINGs:


␈↓ ↓¬␈↓(1)␈αStudy␈αthe␈αtask␈αwhich␈αthe␈αpool␈αis␈αto␈α
do.␈αSee␈αwhat␈αkinds␈αof␈αquestions␈αare␈αasked␈αby␈α(simulated)␈α
experts,␈αas
␈↓ ↓¬␈↓␈↓ ↓Ethey cooperate to carry out that task.

␈↓ ↓¬␈↓(2)␈αDistill␈αthis␈αinto␈αa␈αcore␈αof␈αsimple␈αquestions,␈α␈↓↓Q␈↓,␈α
in␈αsuch␈αa␈αway␈αthat␈αeach␈αinter-expert␈αquestion␈αor␈αtransfer␈α
of
␈↓ ↓¬␈↓␈↓ ↓Econtrol␈αcan␈α
be␈αrephrased␈α
in␈αterms␈αof␈α
some␈α␈↓↓␈↓&q␈↓)αβ␈α
ε␈αQ␈↓.␈α The␈α
size␈αof␈α
␈↓↓Q␈↓␈αis␈αvery␈α
important.␈α If␈α
Q␈αis␈α
too␈αlarge,
␈↓ ↓¬␈↓␈↓ ↓Eaddition␈α
of␈α
new␈α
BEINGs␈αwill␈α
demand␈α
either␈α
great␈αe≥ort␈α
or␈α
great␈α
intelligence␈α(an␈α
example␈α
of␈α
a␈αsystem
␈↓ ↓¬␈↓␈↓ ↓Elike␈αthis␈αis␈αACTORS␈↓#
1␈↓#).␈αIf␈αQ␈αis␈αtoo␈αsmall,␈αall␈αthe␈αnon-uniformity␈αis␈αsimply␈αpushed␈αdown␈αinto␈αthe␈αvalues
␈↓ ↓¬␈↓␈↓ ↓Eof one or two general catchall questions (all ≡rst-order logical languages do this).

␈↓ ↓¬␈↓(3)␈αList␈αall␈αthe␈αBEINGs␈αwho␈αwill␈αbe␈αpresent␈αin␈αthe␈αpool,␈αand␈α≡ll␈αin␈αtheir␈αparts.␈αThe␈αtime␈αand␈αe≥ort␈αrequired
␈↓ ↓¬␈↓␈↓ ↓Eto␈α≡ll␈αin␈αa␈αnew␈αBEING␈αis␈α␈↓βindependent␈↓␈αof␈αthe␈αnumber␈αof␈αBEINGs␈αalready␈αin␈αthe␈αpool,␈αbecause␈αBEINGs
␈↓ ↓¬␈↓␈↓ ↓Ecan␈αcommunicate␈α
via␈αnondeterministic␈α
goal-directed␈αmechanisms,␈α
and␈αdo␈α
not␈αhave␈α
to␈αknow␈α
the␈αnames␈α
of
␈↓ ↓¬␈↓␈↓ ↓Ethe BEINGs who will answer their queries.

␈↓ ↓¬␈↓(4) The human user interacts with the completed BEING community, until the desired task is complete.


␈↓ ↓¬␈↓In␈αthe␈αnext␈αsection,␈α
we␈αapply␈αthis␈αrecipe␈αto␈α
the␈αtask␈α␈↓β"Write␈αa␈αconcept␈α
formation␈αprogram␈αin␈αLISP,␈α
similar␈αto

␈↓ ↓¬␈↓βthe␈αone␈αdescribed␈αby␈αWinston␈↓#
2␈↓#".␈↓␈αThe␈αproblems␈αencountered␈αwill␈αguide␈αus␈αto␈α≡nd␈αa␈αnew␈αdomain␈α(in␈αSection␈α3)

␈↓ ↓¬␈↓and change the formal requirement that each BEING's set of questions be identical.



␈↓ ↓¬␈↓␈↓ βo␈↓∧␈↓&2. Application to Automatic Programming␈↓)αβ␈↓


␈↓ ↓¬␈↓The␈αsystem's␈αtask␈αdomain␈αis␈αchosen␈αto␈αbe␈αAutomatic␈αProgramming;␈αin␈αparticular,␈αwe␈αwant␈αa␈αBEINGs␈αsystem

␈↓ ↓¬␈↓which␈α∀can␈α∪interact␈α∀with␈α∪a␈α∀human␈α∪user␈α∀and␈α∪synthesize␈α∀a␈α∪concept␈α∀formation␈↓#
3␈↓#␈α∪program.␈α∀ By␈α∪␈↓βAutomatic

␈↓ ↓¬␈↓βProgramming,␈↓␈α
we␈α
mean␈α
this␈α
interactive,␈α
nonformal␈αprocess␈α
whereby␈α
a␈α
human␈α
␈↓βuser␈↓␈α
speci≡es␈α
a␈αdesired␈α
program

␈↓ ↓¬␈↓to␈αa␈α␈↓βsystem␈↓,␈αin␈αa␈αtiny␈αsubset␈αof␈αsome␈αnatural␈αlanguage,␈αafter␈αwhich␈αthe␈αsystem␈αand␈αuser␈αdiscuss␈αdetails␈αabout

␈↓ ↓¬␈↓the␈α
task,␈α
until␈α
the␈α
desired␈α
program,␈α
called␈α
the␈α
␈↓βTarget␈α
Program␈↓,␈α
is␈α
≡nally␈α
synthesized.␈↓#
4␈↓#␈α
Any␈α
BEINGs␈α
which

␈↓ ↓¬␈↓contain␈α⊃knowledge␈α⊃about␈α⊃concept␈α⊃formation␈α⊃will␈α⊃be␈α⊃called␈α⊃␈↓βtask-speci≡c␈↓;␈α⊃those␈α⊃possessing␈α⊂program-writing

␈↓ ↓¬␈↓knowledge are called ␈↓βdomain-speci≡c␈↓; the most general BEINGs are ␈↓βdomain-independent␈↓.


␈↓ ↓¬␈↓␈↓ ↓e␈↓↓␈↓&2.1. Analogy: Cooperating Human Experts␈↓)αβ␈↓
␈↓ ↓¬␈↓␈↓εDouglas B. Lenat␈↓ ¬;Interacting Knowledge Modules␈↓ 
∩April 15, 1975      page 4␈↓


␈↓ ↓¬␈↓We␈α∂now␈α∂carry␈α∂out␈α∂the␈α∂four-step␈α∂procedure␈α∂outlined␈α∂near␈α∂the␈α∂end␈α∂of␈α∂the␈α∂last␈α∂section.␈α∂ The␈α∂≡rst␈α∂step␈α∂is␈α∞to

␈↓ ↓¬␈↓examine␈α∂a␈α∂community␈α∂of␈α∂expert␈α∂organisms␈α∂performing␈α∞the␈α∂program-writing␈α∂task.␈α∂ Let␈α∂us␈α∂imagine␈α∂a␈α∞room

␈↓ ↓¬␈↓≡lled␈α∪with␈α∪humans␈α∪from␈α∩a␈α∪variety␈α∪of␈α∪professions:␈α∩scienti≡c␈α∪programmers,␈α∪psychologists,␈α∪system␈α∩hackers,

␈↓ ↓¬␈↓engineers,␈α∪troubleshooters,␈α∩logical␈α∪theorists,␈α∩managers,␈α∪etc.␈α∩There␈α∪is␈α∩one␈α∪distinguished␈α∩member:␈α∪we␈α∩may

␈↓ ↓¬␈↓consider␈α∂him␈α∂to␈α∂be␈α∂the␈α∂director,␈α∂sponsor,␈α∂or␈α∂ultimate␈α∂User␈α∂of␈α∂the␈α∂≡nal␈α∂product.␈α∂ the␈α∂user␈α∂is␈α∂the␈α∂one␈α∂who

␈↓ ↓¬␈↓initially␈α
speci≡es␈α
what␈αis␈α
wanted.␈α
He␈α
can␈αdecide␈α
matters␈α
of␈αpolicy,␈α
but␈α
the␈α
experts␈αare␈α
loathe␈α
to␈α
bother␈αhim

␈↓ ↓¬␈↓unnecessarily.␈α The␈αexperts␈αcooperate␈α
by␈αasking␈αeach␈αother␈αquestions,␈α
helping␈αanswer␈αwhatever␈αthey␈αcan,␈α
and

␈↓ ↓¬␈↓noticing␈α⊃when␈α∩they␈α⊃might␈α∩be␈α⊃able␈α∩to␈α⊃assist.␈α∩Some␈α⊃of␈α⊃them␈α∩can␈α⊃write␈α∩or␈α⊃modify␈α∩parts␈α⊃of␈α∩the␈α⊃partially-

␈↓ ↓¬␈↓synthesized target program; ␈↓βall␈↓ of them can direct programmers to do speci≡c encodings.


␈↓ ↓¬␈↓A␈α∃hypothetical␈α∀dialogue␈α∃was␈α∃simulated,␈α∀in␈α∃which␈α∀the␈α∃concept␈α∃formation␈α∀program␈α∃was␈α∃discussed␈α∀and

␈↓ ↓¬␈↓synthesized.␈α
This␈α∞eventually␈α
called␈α∞for␈α
80␈α∞di≥erent␈α
experts,␈α
asking␈α∞about␈α
10,000␈α∞questions␈α
altogether,␈α∞all␈α
of

␈↓ ↓¬␈↓which␈αfell␈α
nicely␈αinto␈α29␈α
categories␈αof␈α
questions.␈αWe␈αconstructed␈α
a␈αBEING␈αto␈α
model␈αeach␈α
particular␈αhuman

␈↓ ↓¬␈↓expert.


␈↓ ↓¬␈↓So␈α∞our␈α
experimental␈α∞system,␈α
called␈α∞PUP6,␈α
must␈α∞consist␈α
of␈α∞a␈α
community␈α∞of␈α
80␈α∞BEINGs␈α
(and␈α∞several␈α
minor

␈↓ ↓¬␈↓utility␈αfunctions).␈αEach␈αBEING␈αhas␈α29␈αparts.␈α For␈αeach␈αpart␈αof␈αeach␈αBEING,␈αits␈α␈↓↓␈↓&q␈↓)αβ␈↓␈αsubpart␈αis␈αsimply␈αa␈αLISP

␈↓ ↓¬␈↓identi≡er, an atom; the ␈↓↓␈↓&a␈↓)αβ␈↓ subpart is a LISP program, ranging from 1 to 60 lines in length.


␈↓ ↓¬␈↓The␈αconcept␈αof␈αanatomy␈αis␈αrealized␈α
by␈αconstraining␈αthat␈αeach␈αBEING␈αin␈α
the␈αPUP6␈αsystem␈αhave␈αthe␈αsame␈α
set

␈↓ ↓¬␈↓␈↓↓Q␈α∞=␈α∞{q␈↓#v1␈↓#,...,q␈↓#v2␈↓#␈↓#v9␈↓#}␈↓␈α
of␈α∞questions␈α∞which␈α∞it␈α
can␈α∞answer.␈α∞ Of␈α
course␈α∞the␈α∞values␈α∞␈↓↓␈↓&a␈↓)αβ␈↓␈α
are␈α∞di≥erent␈α∞from␈α∞individual␈α
to

␈↓ ↓¬␈↓individual.␈↓#
*␈↓#␈↓#
*␈↓#


␈↓ ↓¬␈↓Analyzing␈αthe␈αsimulated␈αexperts'␈αdiscussion,␈α
there␈αseems␈αto␈αbe␈αa␈α
single␈αdominant␈αscenario␈αfor␈αcreation␈αof␈α
new

␈↓ ↓¬␈↓code.␈α∂They␈α∂work␈α∂together␈α∂to␈α∂simulate␈α∂the␈α∂target␈α∂concept␈α∂formation␈α∂program␈α∂as␈α∂it␈α∂should␈α∂run.␈α∂ When␈α∞the

␈↓ ↓¬␈↓concept␈α
formation␈α
program␈αhas␈α
to␈α
carry␈αout␈α
some␈α
task␈α
X,␈αsome␈α
expert␈α
will␈αrecognize␈α
his␈α
relevance.␈α
He␈αwill

␈↓ ↓¬␈↓then␈α⊃carry␈α⊃out␈α⊃X␈α⊃himself,␈α∩introspect,␈α⊃extract␈α⊃the␈α⊃particular␈α⊃knowledge␈α⊃he␈α∩is␈α⊃using␈α⊃to␈α⊃do␈α⊃X,␈α⊃and␈α∩tell␈α⊃a

␈↓ ↓¬␈↓programmer␈α⊂exactly␈α⊂how␈α⊃he␈α⊂is␈α⊂doing␈α⊃X.␈↓#
5␈↓#␈α⊂The␈α⊂programmer␈α⊃then␈α⊂oversees␈α⊂the␈α⊃encoding␈α⊂of␈α⊂this␈α⊃and␈α⊂the

␈↓ ↓¬␈↓merging into the already-synthesized pieces of target code.

␈↓ ↓¬␈↓****************************************************************************************
␈↓ ↓¬␈↓ε␈↓#
*␈↓#␈↓#
*␈↓#␈α	For␈α	example,␈α	each␈α
BEING␈α	must␈α	have␈α	an␈α	EFFECTS␈α
part,␈α	that␈α	is,␈α	a␈α
part␈α	whose␈α	question␈α	is␈α	abbreviated␈α
"␈↓&EFFECTS␈↓)αβ".␈α	This␈α	part␈α	really␈α
answers␈α	"What
␈↓ ↓¬␈↓ε␈↓ α%final␈α
effects␈α
can␈αyou␈α
produce;␈α
can␈αyou␈α
acheive␈α
this␈αparticular␈α
goal...?"␈α
The␈α
value␈αfilled␈α
in␈α
for␈αthe␈α
EFFECTS␈α
part␈αof␈α
a␈α
BEING␈αB␈α
is
␈↓ ↓¬␈↓ε␈↓ α%simply␈αa␈αset␈αof␈α␈↓&<situation>/<action>␈↓)αβ␈αproductions:␈αif␈αthe␈αdesired␈αeffect␈αmatches␈α␈↓&<situation>␈↓)αβ,␈αthen␈αit␈αcan␈αbe␈αachieved␈αby␈αB,␈αby
␈↓ ↓¬␈↓ε␈↓ α%executing the program ␈↓&<action>␈↓)αβ.
␈↓ ↓¬␈↓␈↓εDouglas B. Lenat␈↓ ¬;Interacting Knowledge Modules␈↓ 
∩April 15, 1975      page 5␈↓


␈↓ ↓¬␈↓In␈α
PUP6,␈α
new␈α
BEINGs␈α
are␈α
created␈α
analogously.␈α
 The␈αtypical␈α
act␈α
of␈α
BEING␈α
creation␈α
is␈α
when␈α
a␈αsingle,␈α
general

␈↓ ↓¬␈↓BEING,␈αsay␈αB,␈αrecognizes␈α
that␈αhe␈αcan␈αdo␈αthe␈α
desired␈αtarget␈αtask,␈αbut␈αalso␈α
knows␈αthat␈αhe␈αis␈αtoo␈α
general␈αand

␈↓ ↓¬␈↓ine≠cient␈α⊂to␈α∂be␈α⊂duplicated␈α∂precisely␈α⊂inside␈α∂the␈α⊂target␈α∂program.␈α⊂Then␈α∂B␈α⊂introspects,␈α∂and,␈α⊂together␈α⊂with␈α∂a

␈↓ ↓¬␈↓programmer-BEING,␈α∪assembles␈α∪a␈α∪new,␈α∪specialized,␈α∪streamlined␈α∪version␈α∪of␈α∪itself,␈α∪B',␈α∪which␈α∪can␈α∀do␈α∪the

␈↓ ↓¬␈↓required␈αtask␈αbut␈αnot␈αmuch␈αelse.␈α Other␈αBEINGs␈αtry␈αto␈α≡ll␈αin␈αor␈αextend␈αall␈αthe␈αparts␈αof␈αB',␈αso␈αB'␈αcan␈αanswer

␈↓ ↓¬␈↓any of the 29 possible questions ␈↓↓Q␈↓.


␈↓ ↓¬␈↓The␈α
process␈αof␈α
≡lling␈αin␈α
all␈α29␈α
parts␈αof␈α
all␈α80␈α
BEINGs␈α
was␈αfairly␈α
routine.␈αFor␈α
part␈αp␈α
of␈αBEING␈α
B,␈αwe␈α
simply

␈↓ ↓¬␈↓looked␈αthrough␈αthe␈αsimulated␈αdialogue,␈αcollected␈αall␈αthe␈αinstances␈αin␈αwhich␈αexpert␈αB␈αanswered␈αa␈αquestion␈αin

␈↓ ↓¬␈↓category␈α∂p,␈α∂then␈α∂wrote␈α∂a␈α∂small␈α∂collection␈α⊂of␈α∂facts␈α∂and␈α∂procedures␈α∂which␈α∂(i)␈α∂embodied␈α∂the␈α⊂knowledge␈α∂that

␈↓ ↓¬␈↓seemed␈α
to␈αbe␈α
required␈αto␈α
produce␈α
these␈αresponses,␈α
(ii)␈αensured␈α
that␈α
in␈αeach␈α
of␈αthese␈α
situations,␈α
the␈αresponse

␈↓ ↓¬␈↓given␈αwould␈α
be␈αthe␈αsame␈α
as␈αthe␈α
one␈αin␈αthe␈α
dialogue.␈αThe␈α
≡rst␈αpoint␈αensures␈α
we␈αare␈α
not␈α"cheating",␈αthe␈α
second

␈↓ ↓¬␈↓point␈α
guarantees␈α
that␈α
the␈α∞whole␈α
dialogue␈α
can␈α
be␈α∞reproduced␈α
by␈α
the␈α
BEINGs␈α∞system,␈α
if␈α
the␈α
human␈α∞user␈α
of

␈↓ ↓¬␈↓PUP6 makes exactly the same requests, comments, and replies as the hypothetical user in the dialgoue.


␈↓ ↓¬␈↓␈↓ ↓e␈↓↓␈↓&2.2. Details of the Experimental System␈↓)αβ␈↓


␈↓ ↓¬␈↓Because␈αof␈αits␈α
genesis␈αfrom␈αa␈α
particular␈αprotocol␈αdialogue␈↓#
6␈↓#,␈αit␈α
is␈αnot␈αsurprising␈α
that␈αthe␈αPUP6␈αsystem␈α
actually

␈↓ ↓¬␈↓did␈αmanage␈αto␈αsynthesize␈αthe␈αtarget␈αprogram.␈αWe␈α
shall␈αrefer␈αto␈αthat␈αconcept␈αformation␈αprogram␈↓#
2␈↓#␈αas␈α
CF.␈αBy

␈↓ ↓¬␈↓adding␈αa␈α
few␈αmore␈αBEINGs␈α
to␈αthe␈αsystem,␈α
the␈αuser␈α
was␈αable␈αto␈α
get␈αPUP6␈αto␈α
write␈αa␈α
grammatical␈αinference

␈↓ ↓¬␈↓program␈α(GI)␈αand␈αa␈αsimple␈αproperty-list␈α
maintenance␈αprogram␈α(PL).␈αDespite␈αthese␈αpromising␈αresults,␈α
dialogue

␈↓ ↓¬␈↓∨exibility problems detracted greatly from PUP6.  For a detailed treatment of PUP6, see ␈↓#
4,7␈↓#.


␈↓ ↓¬␈↓A␈αset␈αof␈α29␈αubiquitous␈αquestions␈αwere␈αchosen,␈α
representing␈αeverything␈αone␈αexpert␈αmight␈αwant␈αto␈αask␈α
another.

␈↓ ↓¬␈↓At␈α∞least,␈α∂they␈α∞naturally␈α∂encompass␈α∞those␈α∂questions␈α∞which␈α∞were␈α∂asked␈α∞during␈α∂the␈α∞simulated␈α∂meeting,␈α∞hence

␈↓ ↓¬␈↓should␈α
be␈α
su≠cient␈α
for␈αgenerating␈α
CF.␈α
 ␈↓↓Q␈↓#vc␈↓#␈↓#vf␈↓#␈↓,␈α
this␈α
universal␈αset␈α
of␈α
BEING␈α
parts,␈αis␈α
listed␈α
in␈α
the␈α
box␈αbelow.

␈↓ ↓¬␈↓One␈αpotential␈α∨aw␈αof␈αa␈αuniversal␈αset␈αof␈αparts␈αis␈αthat␈αeach␈αBEING␈αmight␈αreally␈αonly␈αwant␈αto␈αanswer␈αa␈αfew␈αof

␈↓ ↓¬␈↓the␈αquestions;␈αmany␈α
of␈αthem␈αshould␈αnot␈α
be␈αasked␈αof␈αhim.␈α
In␈αfact,␈αthis␈αsuper∨uity␈α
was␈αfound␈αto␈αexist.␈α
It␈αwas

␈↓ ↓¬␈↓not␈αtolerable,␈αand␈αled␈αto␈αthe␈αFamilies␈αconcept,␈α
which␈αwill␈αbe␈αintroduced␈αin␈αsection␈α3.␈α Each␈αpart␈α
was␈αneeded

␈↓ ↓¬␈↓by␈αonly␈αabout␈αa␈αthird␈αof␈αall␈αthe␈αBEINGs␈αin␈αthe␈αPUP6␈αsystem,␈αin␈αorder␈αfor␈αPUP6␈αto␈αbe␈αable␈αto␈αgenerate␈αCF
␈↓ ↓¬␈↓␈↓εDouglas B. Lenat␈↓ ¬;Interacting Knowledge Modules␈↓ 
∩April 15, 1975      page 6␈↓


␈↓ ↓¬␈↓properly.  The percentage of the BEINGs in PUP6 which actually needed each part is also noted below.
␈↓"↓␈↓ ↓¬␈↓π⊂ααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααα␈↓ ¬⊃
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓WHAT␈↓¬ 82%  A brief summary of the global purpose, a template for an English phrase.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓WHY␈↓¬ 77%  A justification of the BEING's existence, partly filled in by the BEING who called it.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓HOW␈↓¬ 72%  A summary of the methods the BEING intends to employ. Global strategies.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓IDEN␈↓¬ 54%  How this BEING is referenced in English phrases. Implemented as productions.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓WHEN␈↓¬ 19%  Why this BEING should be given control now. The sum of weighted factors.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓ARGS␈↓¬ 63%  Number of arguments; which are optional; names of any local variables.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓ARG-CHECK␈↓¬ 81%  Predicates which examine each argument for suitability.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓EVAL-ARGS␈↓¬  4%  Which arguments are to be evaluated, and which quoted.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓REQUISITES␈↓¬ 10%  If this BEING is actually chosen, what must be made true prior to (pre)␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬           during (co), and after (post) execution.  Work to make these true (unlike ARG-CHECK).␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓DEMONS␈↓¬ 7%  Set of demons to be kept active while the BEING is in control.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓INHIBIT-DEMONS␈↓¬  5%  A lock/unlock mechanism, useful when handling demonic interrupts.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓EFFECTS␈↓¬ 27%  What will probably be true after this BEING is through. What it achieves.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓RESULTS␈↓¬ 15%  How many values this returns. What domain each is in. Side effects.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓META-CODE␈↓¬ 70%  The body of the executable code, but with uninstantiated subparts.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓COMMENTS␈↓¬ 16%  Hints for filling in undefined sections of other BEING parts.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓STRUCTURE␈↓¬ 4% Viewing this BEING as a data structure, the operations doable to it.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓AFFECTS␈↓¬ 14%  Other BEINGs which might be called by this BEING, and why.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓COMPLEXITY␈↓¬ 92%  A vector of utility measures, including probable ease of calling,␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬           of success, of (calling)* itself, time/space cost, efficiency of its output results.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓GENERALIZATIONS␈↓¬ 27%  Some other BEINGs, encompassing this one.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓ALTERNATIVES␈↓¬ 16%  Some equivalent BEINGs (to try if this one fails).␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓SPECIALIZATIONS␈↓¬ 40%  How to write a streamlined, special-case version of this BEING.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓¬␈↓π~␈↓¬␈↓↓ENCODABLE␈↓¬ 9%  How to order the evaluation of the other parts for self-streamlining.␈↓ ¬␈↓π~␈↓¬
␈↓"↓␈↓ ↓¬␈↓π%ααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααα␈↓ ¬$

␈↓ ↓¬␈↓Each␈αof␈αthe␈α100␈α
BEINGs␈αultimately␈αpresent␈αin␈αPUP6␈α
should␈αhave␈αhad␈αa␈αvalue␈α
for␈αeach␈αpart␈α(in␈αreality,␈α
only

␈↓ ↓¬␈↓40%␈α∞of␈α∞these␈α∞2900␈α
slots␈α∞were␈α∞≡lled␈α∞in;␈α
only␈α∞75%␈α∞of␈α∞␈↓βthese␈↓␈α
were␈α∞actually␈α∞used␈α∞in␈α
generating␈α∞one␈α∞of␈α∞the␈α
target

␈↓ ↓¬␈↓programs).␈α∂ The␈α∂set␈α∞of␈α∂parts␈α∂breaks␈α∞into␈α∂three␈α∂rough␈α∞categories:␈α∂those␈α∂parts␈α∞which␈α∂are␈α∂useful␈α∂in␈α∞deciding

␈↓ ↓¬␈↓which␈α∩BEING␈α∩gets␈α∩control,␈α∪those␈α∩useful␈α∩primarily␈α∩in␈α∩answering␈α∪the␈α∩user's␈α∩questions␈α∩and␈α∪keeping␈α∩him

␈↓ ↓¬␈↓oriented, and those useful when the BEING gains control.

␈↓ ↓¬␈↓␈↓&CONTROL:␈↓)αβ␈αAt␈αthe␈αhumans'␈αmeeting,␈αonly␈αone␈αexpert␈αspoke␈αat␈αa␈αtime;␈αin␈αthe␈αBEINGs␈αcommunity,␈αonly␈αone
␈↓ ↓¬␈↓BEING␈α∞has␈α∞control␈α∞at␈α∞any␈α∞given␈α∞moment.␈α∞He␈α∞uses␈α∞his␈α∞parts␈α∞to␈α∞do␈α∞things␈α∞(ask,␈α∞create,␈α∞modify),␈α∞and␈α∞yields
␈↓ ↓¬␈↓control␈αeither␈αvoluntarily␈αor␈αthrough␈αinterruption.␈αThe␈αparts␈αuseful␈αhere␈αinclude␈α␈↓¬ARGS,␈αDEMONS,␈αMETA-CODE,
␈↓ ↓¬␈↓¬COMMENTS, ARG-CHECK, and REQUISITES.␈↓

␈↓ ↓¬␈↓␈↓&ORIENTATION:␈↓)αβ␈α⊂In␈α⊂the␈α⊂simulated␈α∂dialogue,␈α⊂the␈α⊂user␈α⊂had␈α∂no␈α⊂trouble␈α⊂whatever␈α⊂understanding␈α⊂what␈α∂the
␈↓ ↓¬␈↓experts␈αasked␈αhim.␈αIn␈αthe␈αactual␈αprogrammed␈αPUP6␈αsystem,␈αthe␈αhuman␈αwho␈αwas␈αsitting␈αat␈αthe␈αteletype␈αquite
␈↓ ↓¬␈↓␈↓βrarely␈↓␈α∂understood␈α∂what␈α∂was␈α∂wanted␈α∂by␈α∂the␈α∂BEINGs.␈α∂He␈α∂frequently␈α∂had␈α∂to␈α∂interrupt␈α∂them␈α∂and␈α∂ask␈α∂them
␈↓ ↓¬␈↓questions␈αabout␈αwho␈αwas␈αin␈αcontrol,␈αwhy,␈αwhat␈αhe␈αwas␈αtrying␈αto␈αdo,␈αwhat␈αhad␈αrecently␈αtranspired,␈αetc.␈αThese
␈↓ ↓¬␈↓ideally␈α
can␈α
be␈α
phrased␈α
as␈α
simple␈α
retrievals␈α∞and␈α
EVALs␈α
of␈α
active␈α
BEINGs'␈α
parts.␈α
 The␈α
BEING␈α∞parts␈α
most
␈↓ ↓¬␈↓often␈α∞called␈α∞for␈α
by␈α∞the␈α∞user␈α∞are␈α
the␈α∞simple␈α∞one-line␈α∞"orientation"␈α
templates.␈α∞These␈α∞include␈α∞WHAT,␈α
HOW,
␈↓ ↓¬␈↓WHY,␈α∩and␈α⊃AFFECTS.␈α∩ Since␈α⊃BEINGs␈α∩can␈α⊃only␈α∩create␈α⊃new␈α∩BEINGs␈α⊃(not␈α∩simple␈α⊃LISP␈α∩functions),␈α⊃the
␈↓ ↓¬␈↓synthesized␈α∂program,␈α∂CF,␈α∂was␈α∂writen␈α⊂as␈α∂a␈α∂pool␈α∂of␈α∂BEINGs␈α∂itself␈α⊂(by␈α∂PUP6,␈α∂but␈α∂not␈α∂during␈α⊂the␈α∂original
␈↓ ↓¬␈↓protocol).␈αAlthough␈αits␈αquestion-answering␈αability␈αis␈αinferior␈αto␈αPUP6,␈αthe␈αfact␈αthat␈αCF␈αhad␈α␈↓βany␈↓␈αsuch␈αpower
␈↓ ↓¬␈↓was␈αsurprising␈αto␈αthe␈αauthor.␈αIn␈αother␈αwords,␈αone␈αcan␈αinterrupt␈αthe␈αsynthesized␈αprogram␈αas␈αit␈αis␈αrunning␈αand
␈↓ ↓¬␈↓ask␈α⊂questions.␈α⊂Any␈α⊂BEING␈α⊂on␈α∂the␈α⊂control␈α⊂stack␈α⊂will␈α⊂provide␈α⊂fully␈α∂instantiated␈α⊂answers␈α⊂to␈α⊂any␈α⊂of␈α⊂its␈α∂29
␈↓ ↓¬␈↓allowable queries (its parts); all other BEINGs will provide only hypothetical answers.


␈↓ ↓¬␈↓␈↓ ↓e␈↓↓␈↓&2.3. Results␈↓)αβ␈↓
␈↓ ↓¬␈↓␈↓εDouglas B. Lenat␈↓ ¬;Interacting Knowledge Modules␈↓ 
∩April 15, 1975      page 7␈↓


␈↓ ↓¬␈↓Four␈αof␈αthe␈αmost␈αsigni≡cant␈αquestions␈αfor␈αautomatic␈αprogramming␈αsystems␈αoperating␈αinformally␈αvia␈αdialogue

␈↓ ↓¬␈↓are␈α(i)␈αwhat␈αkinds␈αof␈αthings␈αthe␈αuser␈αmust␈αtell␈αthe␈αsystem,␈α(ii)␈αhow␈αit␈αmanages␈αto␈αcarry␈αon␈αthat␈αdialogue,␈α(iii)

␈↓ ↓¬␈↓what␈αprograms␈αare␈αsynthesized,␈αand␈α(iv)␈αhow␈αmuch␈αknowledge␈αis␈αused␈αin␈αmore␈αthan␈αone␈αsynthesis?␈α The␈α≡rst

␈↓ ↓¬␈↓two␈α∞questions␈α∞are␈α
answered␈α∞by␈α∞recalling␈α
that␈α∞the␈α∞PUP6␈α
BEINGs␈α∞are␈α∞␈↓βby␈α
construction␈↓␈α∞able␈α∞to␈α∞reproduce␈α
the

␈↓ ↓¬␈↓speci≡c␈α∞experts'␈α∞protocol␈α∞which␈α
resulted␈α∞in␈α∞CF.␈α∞The␈α∞mechanism␈α
is␈α∞simply␈α∞questioning␈α∞and␈α∞answering,␈α
with

␈↓ ↓¬␈↓new␈α∀BEINGs␈α∪generated␈α∀as␈α∀side␈α∪e≥ects␈α∀of␈α∀satisfying␈α∪some␈α∀BEINGs'␈α∀requests.␈α∪The␈α∀range␈α∀of␈α∪programs

␈↓ ↓¬␈↓synthesized␈αincludes␈α
CF,␈αGI␈α
(a␈αsimple␈α
grammatical␈αinference␈α
program),␈αand␈α
PL␈α(an␈α
even␈αsimpler␈αproperty␈α
list

␈↓ ↓¬␈↓manipulation␈α⊂program).␈α⊂ The␈α⊂≡nal␈α⊂question␈α∂can␈α⊂be␈α⊂answered␈α⊂easily␈α⊂in␈α∂a␈α⊂BEINGs␈α⊂system:␈α⊂in␈α⊂the␈α⊂case␈α∂of

␈↓ ↓¬␈↓PUP6,␈α∞39␈α∞BEINGs␈α∞were␈α∞used␈α∞in␈α∞synthesizing␈α∞␈↓βall␈α∞three␈↓␈α∞target␈α∞programs;␈α∞there␈α∞were␈α∞about␈α∞100␈α∞BEINGs␈α∞in

␈↓ ↓¬␈↓total.


␈↓ ↓¬␈↓␈↓ ↓∃␈↓↓␈↓&2.3.1. Character of the Dialogue between User and PUP6␈↓)αβ␈↓


␈↓ ↓¬␈↓By␈α∂lumping␈α∂all␈α∂the␈α∂parts␈α∂of␈α∂the␈α∂PUP6␈α∂system␈α∂together␈α∂conceptually,␈α∂the␈α∂interaction␈α∂is␈α∂seen␈α∂as␈α∂a␈α∞␈↓βdialogue␈↓

␈↓ ↓¬␈↓between␈αa␈αhuman␈αuser␈αand␈αan␈αautomatic␈αprogramming␈αsystem.␈α We␈αpresent␈αbelow␈αthe␈αinitial␈αfew␈αlines␈αof␈αthe

␈↓ ↓¬␈↓actual␈α
300-page␈α
long␈α∞dialogue␈α
which␈α
resulted␈α∞in␈α
CF␈α
being␈α
generated.␈α∞This␈α
is␈α
shown␈α∞as␈α
the␈α
human␈α∞user␈α
of

␈↓ ↓¬␈↓PUP6 might see it.

␈↓"↓␈↓ ↓¬␈↓π⊂ααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααα␈↓ ¬⊃
␈↓"↓␈↓ ↓¬␈↓↓␈↓π~␈↓↓ PUP: I want any task.␈↓ ¬␈↓π~␈↓↓
␈↓"↓␈↓ ↓¬␈↓↓␈↓π~␈↓↓ ␈↓βUSER: Write a program which does concept formation.␈↓↓␈↓ ¬␈↓π~␈↓↓
␈↓"↓␈↓ ↓¬␈↓↓␈↓π~␈↓↓ PUP: I am about to write a program for ␈↓¬specialized CONCEPT-FORMATION␈↓↓.␈↓ ¬␈↓π~␈↓↓
␈↓"↓␈↓ ↓¬␈↓↓␈↓π~␈↓↓ PUP: Give me a proper name for this program. (I suggest CF or CONCEPT)␈↓ ¬␈↓π~␈↓↓
␈↓"↓␈↓ ↓¬␈↓↓␈↓π~␈↓↓ ␈↓βUSER: CF.␈↓↓␈↓ ¬␈↓π~␈↓↓
␈↓"↓␈↓ ↓¬␈↓↓␈↓π~␈↓↓ PUP: I am temporarily replacing the task "␈↓¬specialized form of CONCEPT-FORMATION␈↓↓"␈↓ ¬␈↓π~␈↓↓
␈↓"↓␈↓ ↓¬␈↓↓␈↓π~␈↓↓           by "␈↓¬specialized form of PARTITION-A-DOMAIN␈↓↓".␈↓ ¬␈↓π~␈↓↓
␈↓"↓␈↓ ↓¬␈↓↓␈↓π~␈↓↓ PUP: Must CF mimic di≥erent human subjects? (respond YES, NO, ?, or DETAILS)␈↓ ¬␈↓π~␈↓↓
␈↓"↓␈↓ ↓¬␈↓↓␈↓π~␈↓↓ ␈↓βUSER: NO.␈↓↓␈↓ ¬␈↓π~␈↓↓
␈↓"↓␈↓ ↓¬␈↓π%ααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααααα␈↓ ¬$


␈↓ ↓¬␈↓From␈α⊂the␈α⊂above,␈α⊂one␈α⊂gets␈α⊂the␈α⊂true␈α⊂impression␈α⊂that␈α⊂the␈α⊂dialogue␈α⊂was␈α⊂fairly␈α⊂tedious:␈α⊂choosing␈α⊂names␈α⊂and

␈↓ ↓¬␈↓resolving␈α
design␈α
decisions␈α
␈↓βwas␈↓␈α
the␈α
user's␈α
main␈α
role.␈αThere␈α
were,␈α
in␈α
sum,␈α
about␈α
300␈α
such␈α
responses␈αinvolved␈α
in

␈↓ ↓¬␈↓generating CF.


␈↓ ↓¬␈↓The␈αsystem␈α
was␈αspeci≡ed␈αby␈α
collecting␈αknowledge␈αfrom␈α
one␈αsingle␈αprotocol;␈α
as␈αa␈αresult,␈α
each␈αBEING␈αhad␈α
only

␈↓ ↓¬␈↓those␈α
parts␈α
speci≡ed␈α
which␈α
it␈α∞actually␈α
would␈α
need␈α
in␈α
the␈α
ensuing␈α∞dialogue.␈α
 Part␈α
of␈α
the␈α
di≠culty␈α∞with␈α
new
␈↓ ↓¬␈↓␈↓εDouglas B. Lenat␈↓ ¬;Interacting Knowledge Modules␈↓ 
∩April 15, 1975      page 8␈↓


␈↓ ↓¬␈↓dialogues␈αstemmed␈αfrom␈αthis␈αminimal␈αcompletion.␈α In␈αthe␈α
protocol,␈αwhen␈αa␈αdecision␈αwas␈αmade␈αby␈αexperts,␈α
the

␈↓ ↓¬␈↓knowledge␈αnecessary␈αto␈αfollow␈αthe␈α␈↓βother␈↓␈αalternative␈αbranch␈αwas␈αnot␈αused␈αby␈αthe␈αexpert␈αat␈αthat␈αtime,␈αnor␈αwere

␈↓ ↓¬␈↓such␈αsuper∨uous␈αfacts␈αsupplied␈αto␈α
the␈αBEINGs␈αin␈αPUP6.␈αThus␈αthe␈α
user␈αof␈αPUP6␈αmust␈αalmost␈αalways␈α
resolve

␈↓ ↓¬␈↓each␈αchoice␈αthe␈αway␈α
the␈αsimulated␈α(protocol)␈αuser␈α
did,␈αor␈αelse␈αhe␈αwill␈α
draw␈αon␈αknowledge␈αwhich␈α
the␈αexperts

␈↓ ↓¬␈↓had␈α(but␈α
didn't␈αuse␈α
before)␈αbut␈α
which␈αPUP6␈α
lacks.␈α It␈αis␈α
felt␈αthat␈α
if␈αall␈α
the␈αparts␈α
of␈αall␈α
the␈αBEINGs␈αhad␈α
been

␈↓ ↓¬␈↓faithfully␈α∂≡lled␈α∂in,␈α∂this␈α∂problem␈α∂would␈α∂have␈α∂subsided.␈α∂Basically,␈α∂the␈α∂di≠culty␈α∂is␈α∂one␈α∂of␈α∂modelling␈α∂all␈α∂the

␈↓ ↓¬␈↓possibly␈α
relevant␈α
knowledge␈αan␈α
expert␈α
has,␈α
rather␈αthan␈α
(as␈α
was␈α
done)␈αjust␈α
capturing␈α
enough␈α
of␈αhis␈α
knowledge

␈↓ ↓¬␈↓to do a few given tasks.


␈↓ ↓¬␈↓␈↓ ↓e␈↓↓␈↓&2.4. Conclusions␈↓)αβ␈↓


␈↓ ↓¬␈↓Some␈αof␈αthe␈αproblems␈αencountered␈αin␈αPUP6␈αwere␈αdue␈αto␈αincomplete␈α≡lling␈αin␈αof␈αBEINGs,␈αpoor␈αchoice␈αof␈α␈↓↓Q␈↓,

␈↓ ↓¬␈↓absence␈α⊂of␈α⊃needed␈α⊂BEINGs␈α⊃(like␈α⊂"DEBUG"),␈α⊂overly␈α⊃simple␈α⊂translation␈α⊃mechanisms,␈α⊂limited␈α⊃channels␈α⊂for

␈↓ ↓¬␈↓communication,␈α
the␈αneed␈α
for␈αevery␈α
BEING␈αto␈α
conform␈αto␈α
a␈αsingle␈α
set␈αQ␈α
of␈αquestions,␈α
the␈αcreation␈α
of␈αa␈α
system

␈↓ ↓¬␈↓designed to do just one particular task.


␈↓ ↓¬␈↓Some␈α
of␈αthe␈α
di≠culties␈αstem␈α
from␈αthe␈α
nature␈αof␈α
the␈αtask.␈α
In␈αany␈α
long␈αdialogue,␈α
the␈αuser␈α
often␈αforgets,␈α
changes

␈↓ ↓¬␈↓his␈αmind,␈αerrs,␈αetc.␈α
A␈αvery␈αsophisticated␈αuser␈α
model␈αwould␈αbe␈αnecessary␈α
to␈αaccomodate␈αthis␈αerrorful␈αprocess␈α
in

␈↓ ↓¬␈↓a␈αnon-debugging␈αsystem.␈αWithout␈αsuch␈αabilities,␈αthe␈αsystem␈α
itself␈αmay␈αbe␈αled␈αinto␈αerror.␈α While␈αmost␈αbugs␈α
␈↓βare␈↓

␈↓ ↓¬␈↓avoidable␈αby␈αcareful␈αrecord-keeping,␈αit␈αproved␈αunrealistic␈αto␈αmake␈αno␈αprovision␈αfor␈αdebugging␈αa␈αnew␈αthirty-

␈↓ ↓¬␈↓page program. When a few errors did occur, PUP6 itself had to be altered.


␈↓ ↓¬␈↓The␈αperformance␈α
of␈αthe␈α
BEINGs␈αrepresentation␈α
itself␈αin␈α
PUP6␈αis␈α
mixed.␈α Two␈α
advantages␈αwere␈α
hoped␈αfor

␈↓ ↓¬␈↓by␈αusing␈αa␈αuniform␈αset␈αof␈αBEING␈αparts.␈α Addition␈αof␈αnew␈αBEINGs␈αto␈αthe␈αpool␈αwas␈αnot␈αeasy␈α(for␈αuntrained

␈↓ ↓¬␈↓users)␈α⊃but␈α⊃communication␈α⊃among␈α⊂BEINGs␈α⊃␈↓βwas␈↓␈α⊃easy␈α⊃(fast,␈α⊃natural).␈α⊂Two␈α⊃advantages␈α⊃were␈α⊃hoped␈α⊃for␈α⊂by

␈↓ ↓¬␈↓keeping␈α⊂the␈α⊃BEINGs␈α⊂highly␈α⊃structured.␈α⊂ The␈α⊃interactions␈α⊂(especially␈α⊂with␈α⊃the␈α⊂user)␈α⊃were␈α⊂brittle,␈α⊃but␈α⊂the

␈↓ ↓¬␈↓complex tasks put to the pool ␈↓βwere␈↓ successfully completed.


␈↓ ↓¬␈↓The␈α⊃crippling␈α⊃problems␈α⊂are␈α⊃seen␈α⊃to␈α⊃be␈α⊂with␈α⊃user-system␈α⊃communication,␈α⊃not␈α⊂with␈α⊃the␈α⊃ideas␈α⊃dicussed␈α⊂in

␈↓ ↓¬␈↓Section␈α1.␈α Sophisticated,␈αbug-free␈αprograms␈α␈↓βwere␈↓␈αgenerated,␈αafter␈αhours␈αof␈αfairly␈αhigh␈αlevel␈αdialogue␈αwith␈αan

␈↓ ↓¬␈↓active␈αuser,␈αafter␈αtens␈αof␈α
thousands␈αof␈αmessages␈αpassed␈αamong␈αthe␈α
BEINGs.␈α Part␈αof␈αthis␈αsuccess␈αis␈α
attributed
␈↓ ↓¬␈↓␈↓εDouglas B. Lenat␈↓ ¬;Interacting Knowledge Modules␈↓ 
∩April 15, 1975      page 9␈↓


␈↓ ↓¬␈↓to␈αdistributing␈αthe␈αresponsibility␈αfor␈αwriting␈αcode␈αand␈αfor␈αrecognizing␈αrelevance,␈αto␈αa␈αhundred␈αentities,␈αrather

␈↓ ↓¬␈↓than␈α
having␈α
a␈αfew␈α
central␈α
monitors␈αworry␈α
about␈α
everything.␈α The␈α
standardization␈α
of␈αparts␈α
made␈α
≡lling␈αin␈α
the

␈↓ ↓¬␈↓BEINGs' contents fairly painless.


␈↓ ↓¬␈↓To␈α∀continue␈α∃our␈α∀research,␈α∃we␈α∀must␈α∃≡nd␈α∀a␈α∃task␈α∀where␈α∃BEINGs␈α∀are␈α∃well-suited,␈α∀where␈α∃the␈α∀problems

␈↓ ↓¬␈↓encountered␈α∞in␈α∂PUP6␈α∞won't␈α∞recur.␈α∂What␈α∞␈↓βare␈↓␈α∞BEINGs␈α∂good␈α∞for?␈α∞ The␈α∂idea␈α∞of␈α∞a␈α∂≡xed␈α∞set␈α∞of␈α∂parts␈α∞(which

␈↓ ↓¬␈↓distinguishes␈α∞them␈α∞from␈α∞ACTORs␈↓#
1␈↓#)␈α∞is␈α∞useful␈α∞if␈α∞the␈α∞mass␈α∞of␈α∞knowledge␈α∞is␈α∞too␈α∞huge␈α∞for␈α∞one␈α∞individual␈α
to

␈↓ ↓¬␈↓keep␈α"on␈αtop"␈α
of.␈α It␈αthen␈αshould␈α
be␈αorganized␈αin␈αa␈α
very␈αuniform␈αway␈α(to␈α
simplify␈αpreparing␈αit␈α
for␈αstorage),

␈↓ ↓¬␈↓yet it must also be highly structured (to speed up retrieval).


␈↓ ↓¬␈↓The␈α∩new␈α∩domain␈α⊃should␈α∩be␈α∩one␈α⊃in␈α∩which␈α∩the␈α∩problems␈α⊃centarl␈α∩to␈α∩our␈α⊃research␈α∩are␈α∩isolated␈α∩from␈α⊃the

␈↓ ↓¬␈↓staggering␈α
complexities␈αof␈α
natural␈α
language␈αhandling.␈α
For␈αthese␈α
reasons,␈α
we␈αnext␈α
consider␈αautomated␈α
research

␈↓ ↓¬␈↓in␈α
the␈α
≡eld␈α
of␈α
elementary␈α
number␈α
theory,␈α
as␈α
a␈α
potential␈α
task␈α
domain.␈α
 BEINGs␈α
are␈α
big␈α
and␈α
slow,␈αbut␈α
valuable

␈↓ ↓¬␈↓for␈α∞organizing␈α∞knowledge␈α∞in␈α∞ways␈α∞meaningful␈α∞to␈α∞how␈α∞it␈α∞will␈α∞be␈α∞used.␈α∞In␈α∞the␈α∞system␈α∞described␈α∞in␈α∞the␈α∞next

␈↓ ↓¬␈↓section,␈α∞BEINGs␈α∞will␈α∞be␈α∞one␈α∞--␈α∞but␈α∞not␈α∞the␈α∞only␈α∞--␈α∞internal␈α∞mechanism␈α∞for␈α∞representing␈α∂and␈α∞manipulating

␈↓ ↓¬␈↓knowledge.



␈↓ ↓¬␈↓␈↓ βy␈↓∧␈↓&3. Application to Mathematics Research␈↓)αβ␈↓


␈↓ ↓¬␈↓␈↓ ↓e␈↓↓␈↓&3.1. Correcting di≠culties of the Automatic Programming system␈↓)αβ␈↓


␈↓ ↓¬␈↓The␈αmost␈αdevastating␈αcriticism␈αof␈αBEINGs␈αwas␈αthat,␈αin␈αPUP6,␈αeach␈αBEING␈αwanted␈αonly␈αabout␈α10␈αof␈αthe␈α29

␈↓ ↓¬␈↓possible␈α
parts␈α
≡lled␈α
in;␈α∞that␈α
is,␈α
each␈α
BEING␈α
part␈α∞was␈α
really␈α
needed␈α
only␈α
by␈α∞a␈α
third␈α
of␈α
all␈α
the␈α∞BEINGs,␈α
in

␈↓ ↓¬␈↓order␈α
for␈αPUP6␈α
to␈αsynthesize␈α
the␈α
concept␈αformation␈α
program␈αsuccessfully.␈α
 Analyzing␈αthis␈α
problem␈α
leads␈αus

␈↓ ↓¬␈↓back␈α
to␈α
the␈α
cooperating␈αexperts␈α
analogy.␈α
There,␈α
we␈α
≡nd␈αthe␈α
additional␈α
structural␈α
level␈α
of␈α␈↓βprofessions␈↓.␈α
 Experts

␈↓ ↓¬␈↓within␈α∀a␈α∀given␈α∀profession␈α∀can␈α∀communicate␈α∀some␈α∀specialized␈α∀questions␈α∀and␈α∀answers␈α∀which␈α∀would␈α∪be

␈↓ ↓¬␈↓unintelligible␈αto␈αoutsiders.␈↓#
*␈↓#␈↓#
*␈↓#␈αOf␈αcourse,␈αmany␈αof␈αthe␈αquestions␈αare␈αfound␈αin␈αmore␈αthan␈αone␈αprofession;␈αa␈αfew

␈↓ ↓¬␈↓are really universal.


␈↓ ↓¬␈↓By␈αchanging␈αthe␈αdomain␈αof␈αthe␈αBEINGs'␈αactivities␈αto␈αelementary␈αmathematics␈↓#
1␈↓#␈↓#
0␈↓#,␈αthe␈αnatural␈αcommunication


␈↓ ↓¬␈↓****************************************************************************************
␈↓ ↓¬␈↓ε␈↓#
*␈↓#␈↓#
*␈↓# For example, the hackers' jargon. Thus, our group of experts was not really uniform, but rather structured along occupational lines.
␈↓ ↓¬␈↓␈↓εDouglas B. Lenat␈↓ ¬;Interacting Knowledge Modules␈↓ 
βApril 15, 1975      page 10␈↓


␈↓ ↓¬␈↓problems␈αmentioned␈αin␈αthe␈αlast␈αsection␈αbecome␈αmanageable.␈α A␈αsmall␈αcore␈αof␈αprimitive␈αconstructions␈αmust␈αbe

␈↓ ↓¬␈↓mastered,␈αbut␈αafter␈αthat,␈αit␈αis␈αthe␈αuser's␈α
responsibility␈αto␈αde≡ne␈αany␈αadditional␈αphrases.␈αThis␈αis␈αtrue␈α
in␈α"real"

␈↓ ↓¬␈↓mathematical␈αresearch,␈αand␈αmakes␈αtranslation␈αfairly␈αroutine.␈α A␈αsecond␈αadvantage␈αof␈αthis␈αtask␈αdomain␈αis␈αthat

␈↓ ↓¬␈↓there␈α
is␈αnot␈α
any␈αone␈α
speci≡c␈α
task␈αwhich␈α
the␈αsystem␈α
is␈α
expected␈αto␈α
achieve.␈α A␈α
␈↓βsolution␈↓␈α
to␈αthis␈α
task␈αwould␈α
mean

␈↓ ↓¬␈↓successfully␈α⊃accounting␈α⊃for␈α⊃the␈α⊃␈↓βdriving␈↓␈α⊃and␈α⊃the␈α⊃␈↓βpruning␈↓␈α⊃forces␈α⊃which␈α⊃result␈α⊃in␈α∩interesting␈α⊃mathematical

␈↓ ↓¬␈↓theories␈α
being␈αdeveloped.␈α
Success␈αcould␈α
be␈α
measured␈αin␈α
operational␈αterms,␈α
by␈α
applying␈αthese␈α
forces␈αto␈α
various

␈↓ ↓¬␈↓domains of mathematics, and comparing the results to what is already known in those ≡elds.


␈↓ ↓¬␈↓Let␈α
us␈αconsider,␈α
then,␈α
the␈αtask␈α
of␈αdoing␈α
mathematical␈α
research:␈αde≡ning␈α
potentially␈α
interesting␈αmathematical

␈↓ ↓¬␈↓systems,␈α_and␈α_developing␈α_theorems␈α_about␈α→them.␈α_Each␈α_BEING␈α_will␈α_represent␈α_a␈α→conceptualization:␈α_a

␈↓ ↓¬␈↓mathematical␈αconstruction,␈αan␈αactivity,␈αa␈αmeta-concept,␈αetc.␈α The␈αBEINGs␈αwill␈αbe␈αgrouped␈αinto␈αa␈αfew␈αfamilies

␈↓ ↓¬␈↓or professions; each family f has its own distinctive set ␈↓↓Q␈↓#vf␈↓#␈↓ of questions which its members can answer.


␈↓ ↓¬␈↓␈↓ ↓e␈↓↓␈↓&3.2. Characteristics of a Proposed System␈↓)αβ␈↓

␈↓ ↓¬␈↓Before␈α∪detailing␈α∀such␈α∪a␈α∪system,␈α∀let␈α∪us␈α∀try␈α∪to␈α∪explain␈α∀the␈α∪motivations␈α∪and␈α∀the␈α∪mechanisms␈α∀by␈α∪which
␈↓ ↓¬␈↓mathematicians operate.␈↓#
1␈↓#␈↓#
1␈↓#

␈↓ ↓¬␈↓(i)␈α∀The␈α∀driving␈α∀and␈α∀pruning␈α∃forces␈α∀are␈α∀(in␈α∀decreasing␈α∀order␈α∀of␈α∃importance)␈α∀aesthetics/interestingness,
␈↓ ↓¬␈↓␈↓ ↓Eintuition, utility, analogy, inductive inference (empirical), and deductive inference (formal).

␈↓ ↓¬␈↓(ii) Each of these forces is useful both in generating new conjectures, and in assessing their acceptability.

␈↓ ↓¬␈↓(iii)␈α
If␈α
the␈α
essence␈α
of␈α
these␈α
ideas␈α
can␈α
be␈α
factored␈α
out␈α
into␈α
an␈α
explicit␈α
set␈α
of␈α
BEINGs␈α
and␈α
utility␈αfunctions,␈α
then
␈↓ ↓¬␈↓␈↓ ↓Ethey can be used to develop almost any branch of mathematics, at almost any level.

␈↓ ↓¬␈↓(iv)␈α⊂Each␈α∂mathematical␈α⊂concept␈α∂should␈α⊂be␈α∂represented␈α⊂in␈α∂several␈α⊂ways,␈α∂including␈α⊂declarative,␈α∂operational,
␈↓ ↓¬␈↓␈↓ ↓Edemonic (recognizing its relevance), exemplary (especially boundary examples), and intuitive.

␈↓ ↓¬␈↓(v)␈αA␈αwide␈αfoundation␈αof␈αintuition,␈αspanning␈αseveral␈αmathematical␈αand␈αreal-world␈αconcepts,␈αis␈αprerequisite␈αto
␈↓ ↓¬␈↓␈↓ ↓Esophisticated behavior in ␈↓βany␈↓ branch of mathematics.

␈↓ ↓¬␈↓Now␈αconsider␈αthe␈αcharacteristics␈αof␈αa␈αman-machine␈αsystem␈αwhich␈αcould␈αbe␈αused␈αexperimentally␈αfor␈αthis␈αtask.
␈↓ ↓¬␈↓The␈α⊗initial␈α↔knowledge␈α⊗in␈α⊗the␈α↔system␈α⊗will␈α⊗consist␈α↔of␈α⊗(i)␈α⊗speci≡c␈α↔facts␈α⊗about␈α↔mathematics,␈α⊗reasoning,
␈↓ ↓¬␈↓programming,␈α⊂and␈α⊂communication,␈α⊂(ii)␈α∂strategies␈α⊂for␈α⊂≡lling␈α⊂out␈α∂parts␈α⊂of␈α⊂incomplete␈α⊂BEINGs,␈α⊂(iii)␈α∂opaque
␈↓ ↓¬␈↓functions␈α∞which␈α
simulate␈α∞parts␈α∞of␈α
Nature,␈α∞and␈α∞(iv)␈α
opaque␈α∞judgment␈α∞criteria␈α
for␈α∞aesthetics,␈α∞interest,␈α
utility,
␈↓ ↓¬␈↓complexity,␈αetc.␈α The␈αspeci≡c␈αfacts␈αare␈αorganized␈αinto␈α4␈αfamilies␈αof␈αBeings;␈αeach␈αfamily␈αinitially␈αhas␈αabout␈α35
␈↓ ↓¬␈↓BEINGs,␈αand␈αeach␈αBEING␈αhas␈αabout␈α25␈αparts.␈αThe␈αfamilies␈αare:␈αStatic␈α(eg,␈αsets),␈αActive␈α(eg,␈αrelation),␈αStatic-
␈↓ ↓¬␈↓Meta␈α∩(eg,␈α∩analogy),␈α∩and␈α∩Active-Meta␈α∩(eg,␈α∩prove).␈α∩ For␈α∩uniformity,␈α∩the␈α∩strategies␈α∩form␈α∩a␈α∩≡fth␈α∩family␈α⊃of
␈↓ ↓¬␈↓BEINGs,␈αcalled␈αArchetypical␈αBEINGs.␈α A␈α
strategy␈αBEING␈αis␈αsimply␈αa␈α
collection␈αof␈αfacts␈αfor␈αdealing␈α
with␈αa
␈↓ ↓¬␈↓particular␈αtype␈αof␈αBEING␈αpart␈α(e.g.,␈αthe␈α"Examples"␈αBEING␈αcontains␈αsuggestions␈αfor␈α≡lling␈αin␈αthe␈αExamples
␈↓ ↓¬␈↓part of any BEING).

␈↓ ↓¬␈↓The␈αintuition␈αfunctions␈αrepresent␈αthe␈αenvironment␈αand␈αare␈α␈↓βopaque␈↓.␈αFor␈αexample,␈αa␈αmodel␈αof␈αa␈αseesaw␈αmight
␈↓ ↓¬␈↓exist,␈α∞and␈α∂the␈α∞system␈α∞could␈α∂play␈α∞around␈α∞at␈α∂varying␈α∞the␈α∞weights␈α∂on␈α∞each␈α∞side␈α∂and␈α∞their␈α∞distance␈α∂from␈α∞the
␈↓ ↓¬␈↓fulcrum,␈α∞and␈α∞the␈α∞seesaw␈α∂function␈α∞would␈α∞explain␈α∞which␈α∂side␈α∞sank␈α∞and␈α∞how␈α∂fast.␈α∞ This␈α∞might␈α∞be␈α∂useful␈α∞in
␈↓ ↓¬␈↓getting an intuition about multiplication, substitution, or symmetry.
␈↓ ↓¬␈↓␈↓εDouglas B. Lenat␈↓ ¬;Interacting Knowledge Modules␈↓ 
βApril 15, 1975      page 11␈↓


␈↓ ↓¬␈↓The␈α∩quantity␈α∩of␈α∩this␈α∩corpus␈α∩appears␈α∩large␈α∩(about␈α∩3000␈α∩BEING-parts␈α∩to␈α∩encode,␈α∩each␈α∩as␈α∩a␈α∩little␈α⊃LISP
␈↓ ↓¬␈↓program),␈α∞and␈α
it␈α∞is␈α
of␈α∞some␈α
interest␈α∞to␈α∞hope␈α
that␈α∞the␈α
very␈α∞same␈α
techniques␈α∞which␈α
lead␈α∞to␈α∞discovering␈α
new
␈↓ ↓¬␈↓mathematical␈α
knowledge␈αlater␈α
on␈α
might␈αbe␈α
able␈α
to␈α"grow"␈α
this␈α
knowledge␈αbase␈α
from␈α
a␈αmuch␈α
smaller␈α
core␈α--
␈↓ ↓¬␈↓say␈αa␈αcollection␈αof␈α100␈αBEINGs␈αwith␈αonly␈αa␈αfew␈αparts␈α≡lled␈αin␈αfor␈αeach.␈α The␈α≡rst␈αactivity␈αof␈αsuch␈αa␈αsystem,
␈↓ ↓¬␈↓then, would be ␈↓βcontemplative␈↓: the interaction with the user would be minimal.

␈↓ ↓¬␈↓The␈α∂user␈α∂is␈α∞considered␈α∂slow␈α∂and␈α∞dangerously␈α∂contradicatory,␈α∂hence␈α∞not␈α∂a␈α∂good␈α∞channel␈α∂to␈α∂obtain␈α∂data␈α∞in
␈↓ ↓¬␈↓general.␈α
 But␈α
as␈α
the␈αknown␈α
information␈α
swells,␈α
the␈αneed␈α
for␈α
guidance␈α
also␈αgrows.␈α
At␈α
some␈α
point␈α
the␈αsystem
␈↓ ↓¬␈↓may␈α∂simply␈α∞be␈α∂swamped␈α∂by␈α∞a␈α∂multitude␈α∞of␈α∂equally-mediocre␈α∂alternatives␈α∞to␈α∂investigate.␈α∞It␈α∂will␈α∂then␈α∞(abeit
␈↓ ↓¬␈↓reluctantly)␈α∞request␈α∞direction␈α∂from␈α∞a␈α∞human␈α∂user,␈α∞in␈α∞what␈α∂is␈α∞to␈α∞be␈α∂an␈α∞␈↓βassimilative␈↓␈α∞phase.␈α∂These␈α∞teachings
␈↓ ↓¬␈↓should␈αbe␈αthe␈αcore␈αde≡nitions␈αof␈αa␈αspeci≡c␈α≡eld,␈αand␈αof␈αcourse␈αshould␈αbe␈αbased␈αon␈αwhat␈αis␈αalready␈αmastered.
␈↓ ↓¬␈↓The␈α≡rst␈αexperiences␈αcould␈αbe␈αin␈αset␈αtheory,␈αBoolean␈αalgebra,␈αabstract␈αalgebra,␈αlogic,␈αor␈αarithmetic.␈α This␈αwill
␈↓ ↓¬␈↓probably be the level ≡nally attained by the actual system.

␈↓ ↓¬␈↓One␈αhigher␈α
mode␈αof␈α
interaction␈αis␈α
conceivable:␈αthat␈α
of␈αa␈α
colleague␈αin␈α
research.␈α In␈α
conjunction␈αwith␈αa␈α
human
␈↓ ↓¬␈↓adviser,␈αthe␈αsystem␈αwould␈αpropose␈αand␈αexplore␈αinteresting␈αnew␈αrelationships,␈αdecide␈αwhich␈αcreations␈αto␈αname,
␈↓ ↓¬␈↓explore␈α∞the␈α∞intuitive␈α∞meanings␈α∞of␈α∞statements,␈α∂etc.␈α∞ Hopefully,␈α∞the␈α∞reader␈α∞has␈α∞balked,␈α∞complaining␈α∂that␈α∞this
␈↓ ↓¬␈↓sounds␈αjust␈αlike␈αthe␈αearlier␈α
phases.␈αIn␈αfact,␈αthe␈αsystem␈α
will␈αnot␈αring␈αa␈αbell␈α
and␈αsuddenly␈αswitch␈αits␈αactivities;␈α
it
␈↓ ↓¬␈↓has␈αno␈αway␈αof␈αknowing␈αthat␈α
its␈αdiscovery␈αof␈αPLUS␈αis␈αnot␈α
new␈αto␈αMankind.␈α The␈αdriving␈αand␈αpruning␈α
forces
␈↓ ↓¬␈↓in␈αall␈α
phases␈αare␈α
the␈αsame:␈αuse␈α
aesthetics␈αand␈α
utility␈αjudgments␈α
to␈α≡ll␈αout␈α
parts␈αof␈α
incomplete␈αBEINGs.␈α If␈α
the
␈↓ ↓¬␈↓guidance␈αof␈αthe␈αhuman␈αturns␈αout␈αto␈αbe␈αimportant,␈αhowever,␈αthen␈αit␈αwill␈αcome␈αas␈αno␈αsurprise␈αif␈αthe␈α∨avor␈αof
␈↓ ↓¬␈↓the interactions changes as the system enters a realm unfamiliar to the user.

␈↓ ↓¬␈↓The␈α
overall␈α
control␈α
∨ow␈α
would␈α
be␈α
a␈α
series␈αof␈α
␈↓β"Work␈α
on␈α
part␈α
p␈α
of␈α
BEING␈α
B"␈↓␈α
directives.␈α The␈α
driving/pruning
␈↓ ↓¬␈↓forces␈αwould␈αeach␈αtime␈αselect,␈αas␈α
the␈αnext␈α(p,B)␈αpair,␈αthe␈αone␈α
which␈αthen␈αseemed␈αmost␈αpromising.␈α During␈α
the
␈↓ ↓¬␈↓course␈α
of␈α
such␈αcompletions,␈α
new␈α
BEINGs␈α
might␈αbe␈α
called␈α
for␈α(split␈α
o≥␈α
rich␈α
subparts␈αof␈α
part␈α
of␈α
an␈αalready-
␈↓ ↓¬␈↓exisiting BEING).

␈↓ ↓¬␈↓The␈αbasic␈αmechanism␈αis␈αthe␈α≡lling␈αin␈αand␈αthe␈αrunning␈αof␈αparts␈αof␈αBEINGs.␈α But␈αBEING␈αparts␈αare␈αgenerally
␈↓ ↓¬␈↓procedural␈αknowledge,␈α
so␈αthis␈α
task␈αreally␈αmeans␈α
automatic␈αcode␈α
synthesis.␈αKnowledge␈αis␈α
stored␈αwith␈α
an␈αidea
␈↓ ↓¬␈↓toward␈α
future␈αusage,␈α
both␈αin␈α
where␈α
it␈αis␈α
placed␈αand␈α
how␈αit␈α
is␈α
recorded.␈α This␈α
is␈αthe␈α
logical␈α
continuation␈αof
␈↓ ↓¬␈↓the␈α∞usage-oriented␈α∞storage␈α∞originating␈α∞in␈α∞PUP1␈↓#
4␈↓#␈α∞and␈α∂developed␈α∞into␈α∞BEINGs␈α∞in␈α∞PUP6␈α∞(Section␈α∞2␈α∂of␈α∞this
␈↓ ↓¬␈↓paper).

␈↓ ↓¬␈↓For␈α⊂automatic␈α⊂programming,␈α⊂␈↓↓Q␈↓#vc␈↓#␈↓#vf␈↓#␈↓␈α⊂divided␈α⊂into␈α⊂three␈α⊂groups␈α⊂of␈α⊂questions:␈α⊂recognizing␈α⊂relevance,␈α∂exerting
␈↓ ↓¬␈↓control,␈α
and␈α
general␈α
orientation␈α
information.␈α
 For␈α
math␈αtheory␈α
formation,␈α
a␈α
set␈α
␈↓↓Q␈↓#vmath␈↓#␈↓␈α
of␈α
25␈α
questions␈αhas
␈↓ ↓¬␈↓been␈α⊂formulated,␈α∂divided␈α⊂into␈α⊂four␈α∂groups:␈α⊂␈↓↓recognizing␈α∂relevance␈↓␈α⊂␈↓ε(Changes,␈α⊂Final,␈α∂Past,␈α⊂Iden)␈↓,␈α⊂␈↓↓altering␈α∂oneself␈↓
␈↓ ↓¬␈↓␈↓ε(Generalizations,␈α
Specializations,␈α	Boundary,␈α
Domain/Range,␈α
Ordering,␈α	Worth,␈α
Interestingness,␈α	Justification,␈α
Operations/Properties)␈↓,␈α
␈↓↓acting␈α	upon
␈↓ ↓¬␈↓↓another␈α⊂BEING␈↓␈α⊂␈↓ε(Boundary-operations,␈α⊃Fillin,␈α⊂Structure,␈α⊂Algorithms,␈α⊂Check,␈α⊃Representation,␈α⊂Views)␈↓,␈α⊂and␈α⊃␈↓↓general␈α⊂information␈↓
␈↓ ↓¬␈↓␈↓ε(Definition,␈α
Intuition,␈αTies␈α
to␈α
related␈αBEINGs,␈α
Examples,␈α
Contents)␈↓.␈α Each␈α
family␈α
or␈αprofession␈α
has␈α
its␈αown␈α
distinctive␈αsubset␈α
of
␈↓ ↓¬␈↓these␈α⊃parts␈α⊂which␈α⊃all␈α⊂its␈α⊃members␈α⊂must␈α⊃relate␈α⊂to.␈α⊃For␈α⊂example,␈α⊃BEINGs␈α⊂representing␈α⊃activities␈α⊃relate␈α⊂to
␈↓ ↓¬␈↓Domain/Range, whereas static BEINGs don't.

␈↓ ↓¬␈↓It␈α∩is␈α⊃␈↓βnot␈↓␈α∩optimal␈α⊃to␈α∩have␈α⊃only␈α∩one␈α∩representation␈α⊃of␈α∩knowledge␈α⊃in␈α∩a␈α⊃system;␈α∩while␈α∩multiple␈α⊃knowledge
␈↓ ↓¬␈↓formalisms␈α∂create␈α∂interfacing␈α∂di≠culties,␈α∞the␈α∂advantages␈α∂of␈α∂expressing␈α∂a␈α∞piece␈α∂of␈α∂information␈α∂in␈α∂the␈α∞best-
␈↓ ↓¬␈↓suited␈α∪way␈α∪is␈α∩considered␈α∪worth␈α∪the␈α∪headaches␈α∩of␈α∪interfacing.␈α∪ BEINGs␈α∩are␈α∪≡ne␈α∪for␈α∪storing␈α∩structured
␈↓ ↓¬␈↓information in an accessable manner, but much of the system will be inaccessable (e.g., intuition).

␈↓ ↓¬␈↓The␈αspeci≡c␈αmath␈αknowledge␈αhas␈αmany␈αsophisticated␈αduties,␈αso␈αit␈αseems␈αappropriate␈αthat␈αBEINGs␈αbe␈αused␈αto
␈↓ ↓¬␈↓hold␈α
and␈αorganize␈α
this␈α
information.␈αThe␈α
main␈α
cost,␈αthat␈α
of␈α
slowness,␈αis␈α
not␈α
critical␈αhere,␈α
since␈αeach␈α
individual
␈↓ ↓¬␈↓chunk␈α
is␈α
used␈α
infrequently,␈α∞and␈α
a␈α
wrong␈α
usage␈α
is␈α∞far␈α
more␈α
serious␈α
than␈α
a␈α∞slow␈α
usage.␈α
 One␈α
≡nal␈α∞factor␈α
in
␈↓ ↓¬␈↓favor␈α∞of␈α∂using␈α∞BEINGs␈α∞here␈α∂is␈α∞that␈α∞all␈α∂the␈α∞knowledge␈α∞that␈α∂is␈α∞available␈α∞at␈α∂the␈α∞time␈α∞of␈α∂creation␈α∞of␈α∂a␈α∞new
␈↓ ↓¬␈↓BEING␈α⊃will␈α⊃≡nd␈α⊃its␈α⊃way␈α⊃to␈α⊃the␈α⊃right␈α⊃place;␈α⊂any␈α⊃missing␈α⊃knowledge␈α⊃will␈α⊃be␈α⊃conspicuous␈α⊃as␈α⊃a␈α⊃blank␈α⊂or
␈↓ ↓¬␈↓incomplete BEING part.

␈↓ ↓¬␈↓The␈αcontents␈αof␈αeach␈αpart␈αof␈αeach␈αBEING␈αis␈αcomposed␈αof␈αspecialized␈αrules,␈αassertions,␈αand␈αpointers␈αto␈αother
␈↓ ↓¬␈↓parts␈α∩and␈α∪other␈α∩BEINGs.␈α∪The␈α∩knowledge␈α∪may␈α∩have␈α∪to␈α∩be␈α∪altered␈α∩from␈α∪time␈α∩to␈α∪time,␈α∩hence␈α∪must␈α∩be
␈↓ ↓¬␈↓inspectable␈α
and␈α
interpretable␈α
meaningfully␈α
and␈α
easily,␈α
so␈α
compiled␈α
code␈α
is␈α
ruled␈α
out.␈α
To␈α
demand␈α
that␈α
each
␈↓ ↓¬␈↓␈↓εDouglas B. Lenat␈↓ ¬;Interacting Knowledge Modules␈↓ 
βApril 15, 1975      page 12␈↓


␈↓ ↓¬␈↓part␈α∞of␈α∞each␈α
BEING␈α∞be␈α∞itself␈α
a␈α∞BEING␈α∞would␈α∞trivially␈α
cause␈α∞an␈α∞in≡nite␈α
regress.␈α∞Hence␈α∞the␈α∞reliance␈α
upon
␈↓ ↓¬␈↓"intermediate" representations and strategy BEINGs.


␈↓ ↓¬␈↓␈↓ ↓e␈↓↓␈↓&3.3. Projected Results␈↓)αβ␈↓


␈↓ ↓¬␈↓There␈α⊂are␈α⊂several␈α⊂possible␈α⊂outcomes␈α⊂of␈α⊂all␈α⊂this.␈α∂Ideally,␈α⊂the␈α⊂system␈α⊂will␈α⊂≡nd␈α⊂a␈α⊂useful␈α⊂redivision␈α⊂of␈α∂some

␈↓ ↓¬␈↓concepts,␈α
and␈α
new␈αconcepts␈α
and␈α
theories␈α
overlooked␈αby␈α
mathematicians.␈α
 The␈α
next␈αbest␈α
result␈α
would␈α
be␈αthe

␈↓ ↓¬␈↓re-discovery␈α
and␈αre-development␈α
of␈αsome␈α
existing␈αmathematical␈α
theories,␈αperhaps␈α
by␈αbeing␈α
carefully␈αpushed

␈↓ ↓¬␈↓along␈α
the␈α
"right"␈α
path.␈α
 Even␈α
if␈αthe␈α
system␈α
never␈α
gets␈α
beyond␈α
the␈αmost␈α
elementary␈α
levels␈α
in␈α
each␈α
≡eld,␈αthat

␈↓ ↓¬␈↓very failure will indicate for the ≡rst time a lower bound on the magnitude of the theory formation problem.


␈↓ ↓¬␈↓␈↓ ↓e␈↓↓␈↓&3.4. Synthesis␈↓)αβ␈↓


␈↓ ↓¬␈↓Before␈α
closing,␈α
a␈α
brief␈α
review␈α
is␈α
in␈α
order.␈α
We␈α
began␈α
by␈α
examining␈α
why␈α
biological␈α
entities␈α∞dominate␈α
Earth,

␈↓ ↓¬␈↓progressed␈α∞to␈α∞observing␈α∞cooperating␈α∞human␈α∞experts,␈α∞and␈α∞≡nally␈α∞abstracted␈α∞out␈α∞a␈α∞manageable␈α∞set␈α∞of␈α
design

␈↓ ↓¬␈↓constraints␈α⊃for␈α⊃a␈α⊃viable␈α⊂system.␈α⊃These␈α⊃were:␈α⊃discrete␈α⊂modules␈α⊃containing␈α⊃interrelated␈α⊃knowledge␈α⊃on␈α⊂very

␈↓ ↓¬␈↓specialized␈α
topics,␈α∞each␈α
with␈α
the␈α∞same␈α
anatomy␈α∞but␈α
with␈α
a␈α∞di≥erent␈α
area␈α
of␈α∞expertise,␈α
capable␈α∞of␈α
querying,

␈↓ ↓¬␈↓answering, and reproducing.


␈↓ ↓¬␈↓The␈α≡rst␈αimplementation,␈αin␈αthe␈αdomain␈αof␈αAutomatic␈αProgramming,␈αwas␈αpartially␈αsuccessful.␈αThe␈αdi≠culties

␈↓ ↓¬␈↓promted␈α∂us␈α∂to␈α∂choose␈α∂a␈α∂new␈α⊂domain␈α∂(Elementary␈α∂Number␈α∂Theory␈α∂Research)␈α∂and␈α∂weaken␈α⊂our␈α∂constraints

␈↓ ↓¬␈↓(uniform␈αanatomy␈α
only␈αwithin␈α
each␈αfamily).␈α The␈α
role␈αof␈α
knowledge␈αmodules␈αis␈α
central␈αto␈α
the␈αnew␈αsystem,␈α
but

␈↓ ↓¬␈↓not universal: wherever a more e≠cient representation of knowledge is possible, it should be employed.
␈↓ ↓¬␈↓␈↓εDouglas B. Lenat␈↓ ¬;Interacting Knowledge Modules␈↓ 
βApril 15, 1975      page 13␈↓


␈↓ ↓¬␈↓␈↓ ¬m␈↓∧␈↓&References␈↓)αβ␈↓

␈↓"β␈↓ ↓¬␈↓␈↓#
1␈↓#␈α
C.␈α
Hewitt,␈α
"A␈α
Universal␈α
Modular␈αACTOR␈α
Formalism␈α
for␈α
Arti≡cial␈α
Intelligence,"␈α
␈↓β3rd␈α
International␈αJoint
␈↓"↓␈↓ ↓¬␈↓β␈↓ ↓UConference on Arti≡cial Intelligence␈↓, 1973, pp. 235-245.
␈↓"β␈↓ ↓¬␈↓␈↓#
2␈↓#␈αP.␈αWinston,␈α␈↓βLearning␈αStructural␈αDescriptions␈αfrom␈αExamples␈↓,␈αPh.D.␈αthesis,␈αDept.␈αof␈αElectrical␈αEngineering,
␈↓"↓␈↓ ↓¬␈↓␈↓ ↓UTR-76,␈α
Project␈α
MAC,␈α
TR-231,␈α
Arti≡cial␈α
Intelligence␈α
Laboratory,␈α
Massachusetts␈α
Institute␈α
of␈α
Technology,
␈↓"↓␈↓ ↓¬␈↓␈↓ ↓UCambridge, Massachusetts, September, 1970.
␈↓"β␈↓ ↓¬␈↓␈↓#
3␈↓#␈αC.␈α
G.␈αHempel,␈α
␈↓βFundamentals␈αof␈α
Concept␈αFormation␈αin␈α
Empirical␈αScience␈↓,␈α
University␈αof␈α
Chicago,␈αChicago,
␈↓"↓␈↓ ↓¬␈↓␈↓ ↓UIllinois, 1952.
␈↓"β␈↓ ↓¬␈↓␈↓#
4␈↓#␈αC.␈αGreen,␈αR.␈αWaldinger,␈αD.␈αBarstow,␈αR.␈αElshlager,␈αD.␈αLenat,␈αB.␈αMcCune,␈αD.␈αShaw,␈αL.␈αSteinberg,␈α"Progress
␈↓"↓␈↓ ↓¬␈↓␈↓ ↓UReport␈α∪on␈α∩Program-Understanding␈α∪Systems",␈α∩␈↓βSAIL␈α∪Memo␈α∩AIM-240,␈α∪CS␈α∪Report␈α∩STAN-CS-74-444,␈↓
␈↓"↓␈↓ ↓¬␈↓␈↓ ↓UArti≡cial Intelligence Laboratory, Stanford University, August, 1974.
␈↓"β␈↓ ↓¬␈↓␈↓#
5␈↓#␈α∂W.␈α∂A.␈α∂Woods␈α∂and␈α∂J.␈α⊂Makhoul,␈α∂"Mechanical␈α∂Inference␈α∂Problems␈α∂in␈α∂Continuous␈α⊂Speech␈α∂Understanding,
␈↓"↓␈↓ ↓¬␈↓␈↓ ↓U␈↓βThird International Joint Conference on Arti≡cial Intelligence␈↓, 1973, pp. 200-207.
␈↓"β␈↓ ↓¬␈↓␈↓#
6␈↓# R. Floyd, "Toward Interactive Design of Correct Programs", ␈↓βIFIP␈↓ 1971, V. 1, pp. 7-10.
␈↓"β␈↓ ↓¬␈↓␈↓#
7␈↓#␈αD.␈α
Lenat,␈α"Synthesis␈αof␈α
Large␈αPrograms␈α
from␈αSpeci≡c␈αDialogues",␈α
␈↓βProceedings␈αof␈α
the␈αSymposium␈αon␈α
Proving
␈↓"↓␈↓ ↓¬␈↓β␈↓ ↓Uand Improving Programs␈↓, Institut de Recherche d'Informatique et d'Automatique, July, 1975.
␈↓"β␈↓ ↓¬␈↓␈↓#
8␈↓#␈αP.␈αWinston,␈αed.,␈α"New␈αProgress␈αin␈αArti≡cial␈αIntelligence",␈α␈↓βMIT␈αAI␈αLab␈αMemo␈αAI-TR-310␈↓,␈αJune,␈α1974.␈αNote
␈↓"↓␈↓ ↓¬␈↓␈↓ ↓Uespecially the summaries of work on Frames, Demons, Hacker, Heterarchy, Dialogue, and Belief.
␈↓"β␈↓ ↓¬␈↓␈↓#
9␈↓#␈αW.␈αTeitelman,␈α␈↓βINTERLISP␈αReference␈αManual␈↓,␈αXEROX␈αPARC,␈α1974.␈α Use␈αwas␈αalso␈αmade␈αof␈αfeatures␈αof
␈↓"↓␈↓ ↓¬␈↓␈↓ ↓UCLISP, QA4 (J. Rulifson), and QLISP (R. Reboh and E. Sacerdoti).
␈↓"β␈↓ ↓¬␈↓␈↓#
1␈↓#␈↓#
0␈↓#␈αR.␈αB.␈αKershner␈αand␈αL.␈αR.␈αWilcox,␈α␈↓βThe␈αAnatomy␈αof␈αMathematics␈↓,␈αThe␈αRonald␈αPress␈αCompany,␈αNew␈αYork,
␈↓"↓␈↓ ↓¬␈↓␈↓ ↓U1950.
␈↓"β␈↓ ↓¬␈↓␈↓#
1␈↓#␈↓#
1␈↓# J. Hadamard, ␈↓βThe Psychology of Invention in the Mathematical Field␈↓, Dover Publications, New York, 1945.
␈↓"β␈↓ ↓¬␈↓␈↓#
1␈↓#␈↓#
2␈↓#␈α⊂R.␈α⊂Hilpinen,␈α⊂"Rules␈α⊂of␈α⊂Acceptance␈α⊂and␈α⊂Inductive␈α⊂Logic",␈α⊂␈↓βActa␈α⊂Philosophica␈α⊂Fennica␈↓,␈α⊂Fasc.␈α⊃22,␈α⊂North-
␈↓"↓␈↓ ↓¬␈↓␈↓ ↓UHolland␈α∞Publishing␈α∞Company,␈α∂Amsterdam,␈α∞1968.␈α∞Also␈α∞see␈α∂J.␈α∞Pietarinen,␈α∞"Lawlikeness,␈α∂Analogy,␈α∞and
␈↓"↓␈↓ ↓¬␈↓␈↓ ↓UInductive Logic", in Fasc. 26 of the same journal, 1972.


␈↓ ↓¬␈↓␈↓ ↓e␈↓↓␈↓&Acknowledgements␈↓)αβ␈↓


␈↓ ↓¬␈↓The␈α
ideas␈α
and␈α
the␈α
system␈α
described␈α
are␈αbuilt␈α
upon␈α
recent␈α
researches.␈α
Many␈α
hours␈α
of␈α
creative␈αdiscussions␈α
were

␈↓ ↓¬␈↓equally␈α⊃important.␈α⊃ In␈α⊃particular,␈α∩the␈α⊃author␈α⊃acknowledges␈α⊃the␈α⊃contributions␈α∩by␈α⊃C.␈α⊃Green,␈α⊃A.␈α∩Cohn,␈α⊃R.

␈↓ ↓¬␈↓Waldinger,␈αE.␈αSacerdoti,␈αand␈αM.␈αWilber.␈α Computer␈α
time␈αfor␈αthe␈αPUP6␈αresearch␈αwas␈αgenerously␈αprovided␈α
by

␈↓ ↓¬␈↓the Arti≡cial Intelligence Center of SRI.