/LMAR=0/XLINE=3/FONT#0=BAXL30/FONT#1=BAXM30/FONT#2=BAXB30/FONT#3=SUB/FONT#4=SUP/FONT#5=BASL35/FONT#6=NGR25/FONT#7=MATH30/FONT#8=FIX25/FONT#9=GRK30



␈↓ ↓H␈↓α␈↓ βXEPISTEMOLOGICAL PROBLEMS OF COGNOLOGY
















␈↓ ↓H␈↓Abstract:␈αThis␈αpaper␈αattempts␈αto␈αidentify␈αthe␈αepistemological␈αproblems␈αof␈α␈↓↓artificial␈αintelligence␈↓,␈α
or,
␈↓ ↓H␈↓as␈αwe␈α
might␈αrename␈α
the␈αscience,␈α
␈↓↓cognology␈↓.␈α An␈α
admittedly␈αincomplete␈α
classification␈αis␈α
given,␈αand
␈↓ ↓H␈↓some␈α
partial␈α
results␈α
on␈α∞a␈α
two␈α
problems␈α
are␈α
given.␈α∞ These␈α
are:␈α
formalization␈α
of␈α
concepts␈α∞and␈α
the
␈↓ ↓H␈↓reasoning involved in jumping to conclusions.
␈↓ ↓H␈↓␈↓ εH␈↓ 91


␈↓ ↓H␈↓α␈↓ ¬IINTRODUCTION

␈↓ ↓H␈↓␈↓ α_One's␈αopinion␈αon␈αhow␈αlong␈αit␈αwill␈αtake␈αto␈αachieve␈αhuman-level␈αartificial␈αintelligence␈αlargely
␈↓ ↓H␈↓depends␈α∂on␈α∞what␈α∂problems␈α∞one␈α∂can␈α∞see␈α∂between␈α∂the␈α∞present␈α∂state␈α∞of␈α∂our␈α∞science␈α∂and␈α∂this␈α∞goal.
␈↓ ↓H␈↓Unless␈α
one␈αis␈α
convinced␈αthat␈α
the␈αproblems␈α
he␈αcan␈α
see␈αare␈α
all␈αthe␈α
important␈αproblems,␈α
one␈αcan␈α
only
␈↓ ↓H␈↓give␈α∞lower␈α∞bounds␈α∞on␈α∞the␈α∞time␈α∞required␈α∞to␈α∞overcome␈α∞them.␈α∞ Since␈α∞1958␈α∞(see␈α∞McCarthy␈α∞1959),␈α∞I
␈↓ ↓H␈↓have␈αseen␈αepistemological␈αproblems␈αas␈αmajor␈αobstacles␈αto␈αartificial␈αintelligence,␈αand␈αI␈αhave␈αalways
␈↓ ↓H␈↓been␈αdisappointed␈α
at␈αhow␈α
few␈αpeople␈α
work␈αseriously␈α
on␈αthem.␈α
 For␈αexample,␈α
when␈αErik␈α
Sandewall
␈↓ ↓H␈↓and␈αI␈αannounced␈αa␈α
special␈αissue␈αof␈αthe␈α
AI␈αJournal␈αdevoted␈αto␈α
these␈αproblems,␈αwe␈αreceived␈αtoo␈α
few
␈↓ ↓H␈↓relevant papers to justify the special issue.

␈↓ ↓H␈↓␈↓ α_The␈α⊂neglect␈α⊂of␈α⊂epistemology␈α⊂seems␈α⊂related␈α⊂to␈α∂the␈α⊂fact␈α⊂that␈α⊂95%␈α⊂of␈α⊂the␈α⊂work␈α⊂in␈α∂artificial
␈↓ ↓H␈↓intelligence␈αis␈αengineering␈α(trying␈α
to␈αmake␈αa␈αprogram␈αwith␈α
specified␈αperformance),␈αand␈αonly␈αa5%␈α
is
␈↓ ↓H␈↓science␈α(studying␈αa␈αnatural␈αphenomenon).␈α I␈αthink␈αit␈αshould␈αbe␈αabout␈α75-25.␈α For␈αthis␈αreason,␈αand
␈↓ ↓H␈↓also␈α∩to␈α⊃get␈α∩a␈α⊃name␈α∩more␈α⊃parallel␈α∩to␈α∩those␈α⊃of␈α∩other␈α⊃sciences,␈α∩I␈α⊃suggest␈α∩renaming␈α∩our␈α⊃subject
␈↓ ↓H␈↓␈↓↓cognology␈↓ - ␈↓αthe science of intellectual processes and their relation to the problems they solve␈↓.

␈↓ ↓H␈↓␈↓ α_␈↓↓Epistemological␈↓␈α⊂problems␈α∂of␈α⊂cognology␈α∂are␈α⊂clarified␈α∂by␈α⊂trying␈α∂to␈α⊂separate␈α∂them␈α⊂from␈α∂the
␈↓ ↓H␈↓␈↓↓heuristic␈↓␈αproblems.␈α ␈↓↓Epistemology␈↓␈αstudies␈α
what␈αkinds␈αof␈αfacts␈α
are␈αavailable␈αfor␈αattaining␈αgoals,␈α
and
␈↓ ↓H␈↓␈↓↓heuristics␈↓␈αstudies␈αhow␈α
to␈αuse␈αthe␈αfacts.␈α
 Epistemology␈αis␈αalso␈αconcerned␈α
with␈αthe␈αvalid␈α
and␈αuseful
␈↓ ↓H␈↓modes␈αof␈αreasoning␈αthat␈αallow␈αconcluding␈αthat␈αa␈αproblem␈αhas␈αa␈αcertain␈αsolution.␈α Admittedly␈αthis
␈↓ ↓H␈↓separation␈α∀can␈α∀only␈α∀be␈α∪partial,␈α∀because␈α∀facts␈α∀about␈α∀heuristics␈α∪may␈α∀be␈α∀needed␈α∀to␈α∀solve␈α∪the
␈↓ ↓H␈↓subproblem␈α∃of␈α∃what␈α∃heuristics␈α∃to␈α∃use␈α∀(whenever␈α∃finding␈α∃heuristics␈α∃reaches␈α∃the␈α∃level␈α∃of␈α∀a
␈↓ ↓H␈↓subproblem),␈α
and␈α
the␈α
heuristic␈αmethods␈α
for␈α
dealing␈α
with␈α
facts␈αdepend␈α
on␈α
what␈α
kinds␈α
of␈αfacts␈α
there
␈↓ ↓H␈↓are.␈α⊂ Moreover,␈α⊂the␈α⊃problem␈α⊂of␈α⊂how␈α⊃to␈α⊂represent␈α⊂facts␈α⊃has␈α⊂both␈α⊂epistemological␈α⊃and␈α⊂heuristic
␈↓ ↓H␈↓aspects.␈α Nevertheless,␈αwe␈αcan␈αidentify␈αmany␈αspecifically␈αepistemological␈αproblems␈αthat␈αlie␈αbetween
␈↓ ↓H␈↓us and human-level artificial intelligence.

␈↓ ↓H␈↓␈↓ α_Our␈α⊂first␈α⊂observation␈α∂about␈α⊂the␈α⊂information␈α∂available␈α⊂for␈α⊂solving␈α∂problems␈α⊂is␈α⊂that␈α⊂it␈α∂is
␈↓ ↓H␈↓usually partial knowledge.

␈↓ ↓H␈↓Outline of epistemology

␈↓ ↓H␈↓Time, events, states, persistence

␈↓ ↓H␈↓Objects, creation, persistence, destruction Material

␈↓ ↓H␈↓Space, location, motion

␈↓ ↓H␈↓Causality and ability

␈↓ ↓H␈↓Approximations, models, and correspondences

␈↓ ↓H␈↓Persons - ability, goals

␈↓ ↓H␈↓Information - knowledge, belief, desires and goals.  Experimental problems

␈↓ ↓H␈↓␈↓ α_The␈α
traders␈α
buy␈α
and␈α
sell␈α
wheat␈α
by␈α
telephone.␈α
 There␈α
are␈α
at␈α
least␈α
farmers␈α
and␈α
bakers,␈α
but
␈↓ ↓H␈↓perhaps␈αalso␈αwarehouse␈αowners,␈αwholesale␈αdealers␈αand␈αspeculators.␈α We␈αmust␈αrepresent␈αsuch␈αfacts
␈↓ ↓H␈↓␈↓ εH␈↓ 92


␈↓ ↓H␈↓as,␈α"If␈α
I␈αdon't␈αaccept␈α
his␈αoffer␈αnow,␈α
he␈αmay␈α
buy␈αwhat␈αhe␈α
wants␈αfrom␈αsomebody␈α
else␈αbefore␈α
I␈αcan
␈↓ ↓H␈↓call him back".

␈↓ ↓H␈↓␈↓ α_Missionaries␈αand␈αcannibals␈αdialog␈αwith␈αthe␈αheckler␈αwho␈αsuggests␈αthat␈αthe␈αboat␈αmay␈αleak␈αor
␈↓ ↓H␈↓there may be a bridge.

␈↓ ↓H␈↓␈↓ α_More blocks.

␈↓ ↓H␈↓␈↓ α_I␈α
need␈αto␈α
book␈αmy␈α
flights␈αwith␈α
the␈αtravel␈α
agent,␈αbut␈α
I␈αdon't␈α
need␈αto␈α
know␈αnow␈α
the␈αgate␈α
from
␈↓ ↓H␈↓which␈α
my␈α
plane␈α
will␈α
leave,␈α
but␈α
I␈α
need␈α
to␈α
be␈α
able␈α
to␈α
find␈α
out␈α
in␈α
Bahrein␈α
even␈α
without␈αspeaking
␈↓ ↓H␈↓Arabic.

␈↓ ↓H␈↓General notes

␈↓ ↓H␈↓␈↓ α_Include a summary of concepts paper.

␈↓ ↓H␈↓␈↓ α_The problem of refinable knowledge.

␈↓ ↓H␈↓John McCarthy
␈↓ ↓H␈↓Artificial Intelligence Laboratory
␈↓ ↓H␈↓Computer Science Department
␈↓ ↓H␈↓Stanford University
␈↓ ↓H␈↓Stanford, California 94305

␈↓ ↓H␈↓ARPANET: MCCARTHY@SU-AI

␈↓ ↓H␈↓␈↓εThis draft of

␈↓ ↓H␈↓εPUBbed at 18:30 on June 16, 1977.␈↓