/FONT#0=BASL30/FONT#1=BASI30/FONT#2=BASB30/FONT#3=MISC30
␈↓␈↓↓␈↓α␈↓β␈↓
␈↓ ↓H
␈↓ ↓HArtificial␈α∀Intelligence:␈α∀A␈α∀General␈α∀Survey␈α∀by␈α∀Professor␈α∀Sir␈α∀James␈α∀Lighthill,␈α∀FRS,␈α∀in␈α∪␈↓↓Artificial
␈↓ ↓HIntelligence:␈α
a␈α
paper␈α
symposium␈↓,␈α
Science␈α
Research␈α
Council␈α
1973.

␈↓ α_Professor␈α∪Lighthill␈α∪of␈α∪Cambridge␈α∪University␈α∩is␈α∩a␈α∩famous␈α∩hydrodynamicist␈α∩with␈α∩a␈α∩recent
␈↓ ↓Hinterest␈α∞in␈α∞applications␈α∞to␈α∞biology.␈α∞His␈α∞review␈α∞of␈α
artificial␈α
intelligence␈α
was␈α
at␈α
the␈α
request␈α
of␈α
Brian
␈↓ ↓HFlowers,␈αhead␈αof␈αthe␈αScience␈αResearch␈αCouncil␈αof␈αGreat␈αBritain,␈αthe␈αmain␈αfunding␈αbody␈αfor␈αBritish
␈↓ ↓Huniversity␈α
scientific␈α
research.␈α
Its␈α
purpose␈α
was␈α
to␈α
help␈α
the␈α
Science␈αResearch␈αCouncil␈αdecide␈αrequests
␈↓ ↓Hfor␈αsupport␈αof␈αwork␈αin␈αAI.␈αLighthill␈αclaims␈αno␈αprevious␈αacquaintance␈αwith␈αthe␈αfield,␈αbut␈αrefers␈αto␈αa
␈↓ ↓Hlarge␈α
number␈α
of␈α
authors␈α
whose␈α
works␈α
he␈α
consulted,␈α
though␈α
not␈α
to␈α
any␈α
specific␈α
papers.

␈↓ α_The␈α
␈↓↓Lighthill␈α
Report␈↓␈α
is␈α
organized␈α
around␈α
a␈α
classification␈α
of␈α
AI␈α
research␈α
into␈α
three␈α
categories:

␈↓ α_Category␈α∪A␈α∪is␈α∪␈↓↓advanced␈α∪automation␈↓␈α∪or␈α∪␈↓↓applications␈↓,␈α∪and␈α∪he␈α∪approves␈α∩of␈α∩it␈α∩in␈α∩principle.
␈↓ ↓HIncluded␈α
in␈α
A␈α
are␈α
some␈α
activities␈α
that␈α
are␈α
obviously␈α
applied␈αbut␈αalso␈αactivities␈αlike␈αcomputer␈αchess
␈↓ ↓Hplaying␈α⊂that␈α⊂are␈α⊂often␈α⊂done␈α⊂not␈α⊂for␈α∂themselves␈α∂but␈α∂in␈α∂order␈α∂to␈α∂study␈α∂the␈α∂structure␈α∂of␈α∂intelligent
␈↓ ↓Hbehavior.

␈↓ α_Category␈α∂C␈α∞comprises␈α∞studies␈α∞of␈α∞the␈α∞␈↓↓central␈α∞nervous␈α∞system␈↓␈α∞including␈α∞computer␈α∞modeling␈α∞in
␈↓ ↓Hsupport␈α
of␈α
both␈α
neurophysiology␈α
and␈α
psychology.

␈↓ α_Category␈α⊂B␈α⊂is␈α⊂defined␈α⊂as␈α⊂"building␈α⊂robots"␈α∂and␈α∂"bridge"␈α∂between␈α∂the␈α∂other␈α∂two␈α∂categories.
␈↓ ↓HLighthill␈αdefines␈αa␈αrobot␈αas␈αa␈αprogram␈αor␈αdevice␈αbuilt␈αneither␈αto␈αserve␈α
a␈α
useful␈α
purpose␈α
nor␈α
to␈α
study
␈↓ ↓Hthe␈α∪central␈α∩nervous␈α∩system,␈α∩which␈α∩obviously␈α∩would␈α∩exclude␈α∩Unimates,␈α∩etc.␈α∩which␈α∩are␈α∩generally
␈↓ ↓Hreferred␈α
to␈α
as␈α
industrial␈α
robots.␈α
Emphasizing␈α
the␈αbridge␈αaspect␈αof␈αthe␈αdefinition,␈αLighthill␈αstates␈αas
␈↓ ↓Hobvious␈α
that␈α
work␈α
in␈α
category␈α
B␈α
is␈α
worthwhile␈α
only␈α
in␈α
so␈α
far␈α
as␈α
it␈α
contributes␈α
to␈α
the␈α
other␈α
categories.

␈↓ α_If␈α
we␈α
take␈αthis␈αcategorization␈αseriously,␈αthen␈αmost␈αAI␈αresearchers␈αlose␈αintellectual␈αcontact␈αwith
␈↓ ↓HLighthill␈αimmediately,␈αbecause␈αhis␈αthree␈αcategories␈αhave␈αno␈αplace␈αfor␈αwhat␈αis␈αor␈αshould␈αbe␈αour␈αmain
␈↓ ↓Hscientific␈α∞activity␈α∞-␈α∞␈↓αstudying␈α∞the␈α∞structure␈α∞of␈α∞information␈α∞and␈α∞the␈α
structure␈α
of␈α
problem␈α
solving
␈↓ ↓Hprocesses␈α∪independently␈α∪of␈α∪applications␈α∪and␈α∪independently␈α∩of␈α∩its␈α∩realization␈α∩in␈α∩animals␈α∩or
␈↓ ↓Hhumans␈↓.␈α
This␈α
study␈α
is␈α
based␈α
on␈α
the␈α
following␈α
ideas:

␈↓ α_1.␈α∩Intellectual␈α∩activity␈α∩takes␈α∩place␈α∩in␈α∩a␈α⊃world␈α⊃that␈α⊃has␈α⊃a␈α⊃certain␈α⊃physical␈α⊃and␈α⊃intellectual
␈↓ ↓Hstructure:␈α∀Physical␈α∪objects␈α∪exist,␈α∪move␈α∪about,␈α∪are␈α∪created␈α∪and␈α∪destroyed.␈α∪Actions␈α∪that␈α∪may␈α∪be
␈↓ ↓Hperformed␈α
have␈α
effects␈αthat␈αare␈αpartially␈αknown.␈αEntities␈αwith␈αgoals␈αhave␈αavailable␈αto␈αthem␈αcertain
␈↓ ↓Hinformation␈α⊃about␈α⊃this␈α⊃world.␈α⊂Some␈α⊂of␈α⊂this␈α⊂information␈α⊂may␈α⊂be␈α⊂built␈α⊂in,␈α⊂and␈α⊂some␈α⊂arises␈α⊂from
␈↓ ↓Hobservation,␈α
from␈α
communication,␈α
from␈α
reasoning,␈α
and␈α
by␈α
more␈α
or␈α
less␈α
complex␈α
processes␈α
of␈α
retrieval
␈↓ ↓Hfrom␈α∂information␈α∂bases.␈α∂Much␈α∞of␈α∞this␈α∞structure␈α∞is␈α∞common␈α∞to␈α∞the␈α∞intellectual␈α∞position␈α∞of␈α∞animals,
␈↓ ↓Hpeople,␈α∞and␈α∞machines␈α
which␈α
we␈α
may␈α
design,␈α
e.g.␈α
the␈α
effects␈α
of␈α
physical␈α
actions␈α
on␈α
material␈α
objects
␈↓ ↓Hand␈αalso␈αthe␈αinformation␈αthat␈αmay␈αbe␈αobtained␈αabout␈αthese␈αobjects␈αby␈αvision.␈αThe␈αgeneral␈αstructure
␈↓ ↓Hof␈α⊂the␈α⊂intellectual␈α⊂world␈α⊂is␈α⊂far␈α⊂from␈α∂understood,␈α∂and␈α∂it␈α∂is␈α∂often␈α∂quite␈α∂difficult␈α∂to␈α∂decide␈α∂how␈α∂to
␈↓ ↓Hrepresent␈αeffectively␈αthe␈αinformation␈αavailable␈αabout␈αa␈αquite␈αlimited␈α
domain␈α
of␈α
action␈α
even␈α
when␈α
we
␈↓ ↓Hare␈α
quite␈α
willing␈α
to␈α
treat␈α
a␈α
particular␈α
problem␈α
in␈α
an␈α
␈↓↓ad␈α
hoc␈↓␈α
way.

␈↓ α_2.␈αThe␈αprocesses␈αof␈αproblem␈αsolving␈αdepend␈αon␈αthe␈αclass␈αof␈αproblems␈αbeing␈αsolved␈αmore␈αthan
␈↓ ↓Hon␈α∞the␈α∞solver.␈α∞Thus␈α∞playing␈α∞chess␈α∞seems␈α∞to␈α∞require␈α∞look-ahead␈α∞whether␈α∞the␈α∞apparatus␈α∞is␈α∞made␈α∞of
␈↓ ↓Hneurons␈α∩or␈α∩transistors.␈α∩Isolation␈α∩of␈α∩the␈α∩information␈α∩relevant␈α∩to␈α∩a␈α∩problem␈α∩from␈α∩the␈α⊃totality␈α⊃of
␈↓ ↓Hprevious␈αexperience␈αis␈αrequired␈αwhether␈α
the␈α
solver␈α
is␈α
man␈α
or␈α
machine,␈α
and␈α
so␈α
is␈α
the␈α
ability␈α
to␈α
divide
␈↓ ↓Ha␈α⊃problem␈α⊃into␈α⊃weakly␈α⊃connected␈α⊃subproblems␈α⊃that␈α⊃can␈α⊃be␈α⊃thought␈α⊃about␈α⊂separately␈α⊂before␈α⊂the
␈↓ ↓Hresults␈α
are␈α
combined.

␈↓ α_3.␈αExperiment␈αis␈αuseful␈αin␈αdetermining␈αwhat␈αrepresentations␈α
of␈α
information␈α
and␈α
what␈α
problem
␈↓ ↓Hsolving␈α
processes␈α
are␈α
needed␈α
to␈α
solve␈α
a␈α
given␈α
class␈α
of␈α
problems.␈α
We␈α
can␈α
illustrate␈αthis␈αpoint␈αby␈αan
␈↓ ↓Hexample␈α
from␈α
the␈α
␈↓↓Lighthill␈α
Report␈↓␈α
which␈α
asserts␈α
(p.␈α
15)␈α
that␈α
the␈α
heuristics␈αof␈αa␈αchess␈αprogram␈αare
␈↓ ↓Hembodied␈αin␈αthe␈αevaluation␈αfunction.␈αThis␈αis␈αplausible␈αand␈αwas␈αassumed␈α
by␈α
the␈α
first␈α
writers␈α
of␈α
chess
␈↓ ↓Hprograms.␈αExperiment␈α
showed,␈α
however,␈α
that␈α
the␈α
procedures␈α
that␈α
select␈α
what␈α
part␈α
of␈α
the␈α
move␈α
tree␈α
is
␈↓ ↓Hexamined␈α
are␈α
even␈α
more␈α
important,␈α
i.e.␈α
when␈α
the␈α
program␈α
errs␈α
it␈α
is␈α
usually␈α
because␈α
it␈α
didn't␈α
examine
␈↓ ↓Ha␈α∪line␈α∪of␈α∪play␈α∪rather␈α∪than␈α∪because␈α∪it␈α∪mis-evaluated␈α∪a␈α∪final␈α∪position.␈α∪Modern␈α∪chess␈α∩programs
␈↓ ↓Hconcentrate␈α
on␈α
this␈α
and␈α
often␈α
have␈α
simpler␈α
evaluators␈α
than␈α
the␈α
earlier␈α
programs.

␈↓ α_4.␈α⊃The␈α⊃experimental␈α⊃domain␈α⊃should␈α⊃be␈α⊃chosen␈α⊃to␈α⊃test␈α⊃the␈α⊃adequacy␈α⊂of␈α⊂representations␈α⊂of
␈↓ ↓Hinformation␈α
and␈α
of␈αproblem␈αsolving␈αmechanisms.␈αThus␈αchess␈αhas␈αcontributed␈αmuch␈αto␈αthe␈αstudy␈αof
␈↓ ↓Htree␈αsearch;␈αone␈αSoviet␈αcomputer␈αscientist␈αrefers␈αto␈αchess␈αas␈αthe␈α␈↓↓Drosophila␈↓␈αof␈αartificial␈αintelligence.␈αI
␈↓ ↓Hthink␈αthere␈αis␈αmuch␈αmore␈αto␈αbe␈αlearned␈αfrom␈αchess,␈αbecause␈αmaster␈αlevel␈α
play␈α
will␈α
require␈α
more␈α
than
␈↓ ↓Hjust␈αimproving␈αthe␈αpresent␈αmethods␈αof␈αsearching␈αtrees.␈αNamely,␈αit␈αwill␈αrequire␈αthe␈αability␈αto␈αidentify,
␈↓ ↓Hrepresent,␈α
and␈α
recognize␈α
the␈α
patterns␈α
of␈α
position␈α
and␈α
play␈α
that␈α
correspond␈α
to␈α
"chess␈α
ideas",␈α
the␈α
ability
␈↓ ↓Hto␈αsolve␈αsome␈αabstractions␈αof␈αpositions␈α(e.g.␈αhow␈αto␈αmake␈αuse␈αof␈αa␈αpassed␈αpawn␈αand␈αa␈αseventh␈αrank
␈↓ ↓Hrook␈α
jointly)␈α
and␈α
to␈α
apply␈α
the␈α
result␈α
to␈α
actual␈α
positions.␈α
It␈α
will␈α
probably␈α
also␈α
require␈α
the␈αability␈αto
␈↓ ↓Hanalyze␈α⊃a␈α⊃problem␈α⊃into␈α⊃subproblems␈α⊃and␈α⊃combine␈α⊂the␈α⊂separate␈α⊂results.␈α⊂(This␈α⊂ability␈α⊂is␈α⊂certainly
␈↓ ↓Hrequired␈α
for␈α
a␈α
successful␈α
␈↓↓Go␈↓␈α
program).

␈↓ α_Having␈αignored␈αthe␈αpossibility␈αthat␈αAI␈αhas␈αgoals␈αof␈αits␈αown,␈αLighthill␈αgoes␈αon␈α
to␈α
document␈α
his
␈↓ ↓Hclaim␈α
that␈αit␈αhas␈αnot␈αcontributed␈αto␈αapplications␈αor␈αto␈αpsychology␈αand␈αphysiology.␈αHe␈αexaggerates␈αa
␈↓ ↓Hbit␈αhere,␈αit␈αseems␈αworthwhile␈αto␈αspend␈αsome␈αeffort␈αdisputing␈αhis␈αclaims␈αthat␈αAI␈αhas␈αnot␈αcontributed
␈↓ ↓Hto␈α
these␈α
other␈α
subjects.

␈↓ α_In␈α⊃my␈α⊃opinion,␈α⊃AI's␈α⊃contribution␈α⊃to␈α⊃practical␈α⊃applications␈α⊃has␈α⊃been␈α⊃significant␈α⊃but␈α⊂so␈α⊂far
␈↓ ↓Hmostly␈α∞peripheral␈α∞to␈α∞the␈α∞central␈α∞ideas␈α∞and␈α∞problems␈α∞of␈α∞AI.␈α∞Thus␈α∞the␈α∞LISP␈α∞language␈α∞for␈α
symbolic
␈↓ ↓Hcomputing␈α∞was␈α∞developed␈α∞for␈α∞AI␈α∞use,␈α∞but␈α∞has␈α∞had␈α∞applications␈α∞to␈α
symbolic␈α
computations␈α
in␈α
other
␈↓ ↓Hareas,␈α∞e.g.␈α∞physics.␈α∞Moreover,␈α
some␈α
ideas␈α
from␈α
LISP␈α
such␈α
as␈α
conditional␈α
expressions␈α
and␈α
recursive
␈↓ ↓Hfunction␈αdefinitions␈αhave␈αbeen␈αused␈αin␈αother␈αprogramming␈αlanguages.␈αHowever,␈αthe␈αideas␈αthat␈αhave
␈↓ ↓Hbeen␈α⊂applied␈α⊂elsewhere␈α⊂don't␈α⊂have␈α⊂a␈α⊂specifically␈α⊂AI␈α⊂character␈α⊂and␈α∂might␈α∂have␈α∂been␈α∂but␈α∂weren't
␈↓ ↓Hdeveloped␈α
without␈α
AI␈α
in␈α
mind.␈α
Other␈α
examples␈α
include␈α
time-sharing,␈α
the␈αfirst␈αproposals␈αfor␈αwhich
␈↓ ↓Hhad␈α⊃AI␈α⊃motivations␈α⊃and␈α⊃some␈α⊃techniques␈α⊃of␈α⊃picture␈α⊂processing␈α⊂that␈α⊂were␈α⊂first␈α⊂developed␈α⊂in␈α⊂AI
␈↓ ↓Hlaboratories␈α⊂and␈α⊂have␈α⊂been␈α⊂used␈α⊂elsewhere.␈α⊂Even␈α⊂the␈α⊂current␈α⊂work␈α⊂in␈α∂automatic␈α∂assembly␈α∂using
␈↓ ↓Hvision␈αmight␈αhave␈αbeen␈αdeveloped␈αwithout␈αAI␈αin␈αmind.␈αHowever,␈αthe␈αDendral␈αwork␈αhas␈αalways␈αhad
␈↓ ↓Ha␈α⊗specifically␈α⊗AI␈α⊗character,␈α⊗and␈α⊗many␈α⊗of␈α∃the␈α∃recent␈α∃developments␈α∃in␈α∃programming␈α∃such␈α∃as
␈↓ ↓HPLANNER␈α
and␈α
CONNIVER␈α
have␈α
an␈α
AI␈α
motivation.

␈↓ α_AI's␈αcontributions␈αto␈αneurophysiology␈αhave␈αbeen␈αsmall␈αand␈αmostly␈αof␈αa␈αnegative␈αcharacter,␈αi.e.
␈↓ ↓Hshowing␈αthat␈αcertain␈αmechanisms␈αthat␈αneurophysiologists␈αpropose␈αare␈αnot␈α
well␈α
defined␈α
or␈α
inadequate
␈↓ ↓Hto␈αcarry␈αout␈αthe␈αbehavior␈αthey␈αare␈αsupposed␈αto␈αaccount␈αfor.␈αI␈αhave␈αin␈αmind␈αHebb's␈αproposals␈αin␈αhis
␈↓ ↓Hbook␈α
␈↓↓The␈α
Organization␈α
of␈α
Behavior␈↓.␈α
No-one␈α
today␈α
would␈α
believe␈α
that␈αthe␈αgaps␈αin␈αthose␈αideas␈αcould
␈↓ ↓Hbe␈αfilled␈αwithout␈αadding␈αsomething␈αmuch␈αlarger␈αthan␈αthe␈αoriginal␈αwork.␈αMoreover,␈αthe␈αlast␈α20␈αyears
␈↓ ↓Hexperience␈α⊂in␈α⊂programming␈α⊂machines␈α∂to␈α∂learn␈α∂and␈α∂solve␈α∂problems␈α∂makes␈α∂it␈α∂implausible␈α∂that␈α∂cell
␈↓ ↓Hassemblies␈α↔␈↓↓per␈α↔se␈↓␈α↔would␈α↔learn␈α↔much␈α↔without␈α↔putting␈α⊗in␈α⊗some␈α⊗additional␈α⊗organization,␈α⊗and
␈↓ ↓Hphysiologists␈αtoday␈αwould␈αbe␈αunlikely␈αto␈αpropose␈αsuch␈αa␈αtheory.␈αHowever,␈αmerely␈αshowing␈αthat␈αsome
␈↓ ↓Hthings␈α⊂are␈α⊂unlikely␈α⊂to␈α∂work␈α∂is␈α∂not␈α∂a␈α∂␈↓↓positive␈↓␈α∂contribution.␈α∂I␈α∂think␈α∂there␈α∂will␈α∂be␈α∂more␈α∂interaction
␈↓ ↓Hbetween␈α⊂AI␈α⊂and␈α⊂neurophysiology␈α⊂as␈α∂soon␈α∂as␈α∂the␈α∂neurophysiologists␈α∂are␈α∂in␈α∂a␈α∂position␈α∂to␈α∂compare
␈↓ ↓Hinformation␈α∩processing␈α∩models␈α∩of␈α∩higher␈α∩level␈α⊃functions␈α⊃with␈α⊃physiological␈α⊃data.␈α⊃There␈α⊃is␈α⊃little
␈↓ ↓Hcontact␈α∞at␈α∞the␈α∞nerve␈α∞cell␈α
level,␈α
because,␈α
as␈α
Minsky␈α
showed␈α
in␈α
his␈α
PhD␈α
dissertation␈α
in␈α
1954,␈α
almost
␈↓ ↓Hany␈α∂of␈α∂the␈α∂proposed␈α∂models␈α∂of␈α∂the␈α∂neuron␈α∂is␈α∂a␈α∂universal␈α∂computing␈α∂element,␈α∞so␈α∞that␈α∞there␈α∞is␈α∞no
␈↓ ↓Hconnection␈α
between␈α
the␈α
structure␈α
of␈α
the␈α
neuron␈α
and␈α
what␈α
higher␈α
level␈α
processes␈α
are␈α
possible.

␈↓ α_On␈α⊃the␈α⊂other␈α⊂hand,␈α⊂the␈α⊂effects␈α⊂of␈α⊂artificial␈α⊂intelligence␈α⊂research␈α⊂on␈α⊂psychology␈α⊂have␈α⊂been
␈↓ ↓Hlarger␈α∞as␈α∞attested␈α∞by␈α∞various␈α∞psychologists.␈α∞First␈α∞of␈α∞all,␈α∞psychologists␈α∞have␈α∞begun␈α∞to␈α
use␈α
models␈α
in
␈↓ ↓Hwhich␈α∞complex␈α∞internal␈α∞data␈α∞structures␈α∞that␈α∞cannot␈α∞be␈α∞observed␈α∞directly␈α∞are␈α∞attributed␈α∞to␈α∞animals
␈↓ ↓Hand␈αpeople.␈αPsychologists␈αhave␈αcome␈αto␈αuse␈αthese␈αmodels,␈αbecause␈αthey␈αexhibit␈αbehavior␈αthat␈αcannot
␈↓ ↓Hbe␈α∩exhibited␈α∩by␈α∩models␈α∩conforming␈α∩to␈α∩the␈α∩tenets␈α∩of␈α∩behaviorism␈α∩which␈α∩essentially␈α⊃allows␈α⊃only
␈↓ ↓Hconnections␈α∂between␈α∂externally␈α∂observable␈α∂variables.␈α∂Information␈α∂processing␈α∂models␈α∞in␈α∞psychology
␈↓ ↓Hhave␈α∂also␈α∂induced␈α∂dissatisfaction␈α∂with␈α∂psychoanalytic␈α∂and␈α∂related␈α∂theories␈α∂of␈α∂emotional␈α∞behavior.
␈↓ ↓HNamely,␈α∞these␈α∞information␈α∞processing␈α∞models␈α∞of␈α∞emotional␈α∞states␈α∞can␈α∞yield␈α∞predictions␈α
that␈α
can␈α
be
␈↓ ↓Hcompared␈α⊂with␈α⊂experiment␈α⊂or␈α⊂experience␈α⊂in␈α⊂a␈α⊂more␈α⊂definite␈α⊂way␈α∂than␈α∂can␈α∂the␈α∂vague␈α∂models␈α∂of
␈↓ ↓Hpsychoanalysis␈α
and␈α
its␈α
offspring.

␈↓ α_Contributions␈α∞of␈α
AI␈α
to␈α
psychology␈α
are␈α
further␈α
discussed␈α
in␈α
the␈α
paper␈α
␈↓↓Some␈α
Comments␈α
on␈α
the
␈↓ ↓HLighthill␈αReport␈↓␈αby␈αN.␈αS.␈αSutherland␈αwhich␈αwas␈αincluded␈α
in␈α
the␈α
same␈α
book␈α
with␈α
the␈α
Lighthill␈α
report
␈↓ ↓Hitself.

␈↓ α_Systematic␈αcomment␈αon␈αthe␈αmain␈αsection,␈αentitled␈α␈↓↓Past␈αDisappointments␈↓␈αis␈αdifficult␈αbecause␈αof
␈↓ ↓Hthe␈α
strange␈α
way␈α
the␈α
subject␈α
is␈α
divided␈α
up␈α
but␈α
here␈α
are␈α
some␈α
remarks:

␈↓ α_1.␈α∩Automatic␈α∩landing␈α∩systems␈α∩for␈α∩airplanes␈α∩are␈α∩offered␈α∩as␈α∩a␈α∩field␈α∩in␈α⊃which␈α⊃conventional
␈↓ ↓Hengineering␈αtechniques␈αhave␈αbeen␈αmore␈αsuccessful␈αthan␈αAI␈αmethods.␈αIndeed,␈αno-one␈αwould␈αadvocate
␈↓ ↓Happlying␈αthe␈αscene␈αanalysis␈αor␈αtree␈αsearch␈αtechniques␈αdeveloped␈αin␈αAI␈αresearch␈αto␈αautomatic␈αlanding
␈↓ ↓Hin␈α
the␈α
context␈αin␈αwhich␈αautomatic␈αlanding␈αhas␈αbeen␈αdeveloped.␈αNamely,␈αradio␈αsignals␈αare␈αavailable
␈↓ ↓Hto␈α
determine␈α
the␈α
precise␈α
position␈α
of␈α
the␈α
airplane␈α
in␈α
relation␈α
to␈α
a␈α
straight␈α
runway␈α
which␈α
is␈α
guaranteed
␈↓ ↓Hclear␈α
of␈α
interfering␈α
objects.␈α
AI␈α
techniques␈α
would␈α
be␈α
necessary␈αto␈αmake␈αa␈αsystem␈αcapable␈αof␈αlanding
␈↓ ↓Hon␈α∂an␈α∂unprepared␈α∂dirt␈α∂strip␈α∞with␈α∞no␈α∞radio␈α∞aids␈α∞which␈α∞had␈α∞to␈α∞be␈α∞located␈α∞and␈α∞distinguished␈α∞from
␈↓ ↓Hroads␈α⊂visually␈α⊂and␈α⊂which␈α⊂might␈α⊂have␈α⊂cows␈α⊂or␈α⊂potholes␈α⊂or␈α⊂muddy␈α∂places␈α∂on␈α∂it.␈α∂The␈α∂problem␈α∂of
␈↓ ↓Hautomatically␈αdriving␈αan␈αautomobile␈αin␈αan␈αuncontrolled␈αenvironment␈αis␈αeven␈αmore␈αdifficult␈αand␈αwill
␈↓ ↓Hdefinitely␈α
require␈α
AI␈α
techniques,␈αwhich,␈αhowever,␈αare␈αnot␈αnearly␈αready␈αfor␈αa␈αfull␈αsolution␈αof␈αsuch␈αa
␈↓ ↓Hdifficult␈α
problem.

␈↓ α_2.␈α⊂Lighthill␈α⊂is␈α⊂disappointed␈α⊂that␈α⊂detailed␈α⊂knowledge␈α⊂of␈α⊂subject␈α⊂matter␈α⊂has␈α⊂to␈α⊂be␈α∂put␈α∂in␈α∂if
␈↓ ↓Hprograms␈α
are␈α
to␈α
be␈αsuccessful␈αin␈αtheorem␈αproving,␈αinterpreting␈αmass␈αspectra,␈αand␈αgame␈αplaying.␈αHe
␈↓ ↓Huses␈α⊂the␈α⊂word␈α∂␈↓↓heuristics␈↓␈α∂in␈α∂a␈α∂non-standard␈α∂way␈α∂for␈α∂this.␈α∂He␈α∂misses␈α∂the␈α∂fact␈α∂that␈α∂there␈α∂are␈α∂great
␈↓ ↓Hdifficulties␈αin␈αfinding␈αways␈αof␈αrepresenting␈αknowledge␈αof␈αthe␈α
world␈α
in␈α
computer␈α
programs␈α
and␈α
much
␈↓ ↓HAI␈α∂research␈α∂and␈α∞internal␈α∞controversy␈α∞are␈α∞directed␈α∞to␈α∞this␈α∞problem.␈α∞Moreover,␈α∞most␈α∞AI␈α∞researchers
␈↓ ↓Hfeel␈αthat␈αmore␈αprogress␈αon␈αthis␈α␈↓↓representation␈αproblem␈↓␈αis␈αessential␈αbefore␈αsubstantial␈αprogress␈αcan␈αbe
␈↓ ↓Hmade␈α⊂on␈α∂the␈α∂problem␈α∂of␈α∂automatic␈α∂acquisition␈α∂of␈α∂knowledge.␈α∂Of␈α∂course,␈α∂missing␈α∂these␈α∂particular
␈↓ ↓Hpoints␈αis␈αa␈αconsequence␈αof␈αmissing␈αthe␈αexistence␈αof␈αthe␈αAI␈αproblem␈αas␈αdistinct␈αfrom␈αautomation␈αand
␈↓ ↓Hstudy␈α
of␈α
the␈α
central␈α
nervous␈α
system.

␈↓ α_3.␈αA␈α
further␈α
disappointment␈α
is␈α
that␈α
chess␈α
playing␈α
programs␈α
have␈α
only␈α
reached␈α
an␈α
"experienced
␈↓ ↓Hamateur"␈α
level␈α
of␈α
play.␈α
Well,␈α
if␈α
programs␈α
can't␈αdo␈αbetter␈αthan␈αthat␈αby␈α1978,␈αI␈αshall␈αlose␈αa␈α␈↓βB␈↓250␈αbet
␈↓ ↓Hand␈αwill␈αbe␈αdisappointed␈αtoo␈αthough␈αnot␈αextremely␈αsurprised.␈αThe␈αpresent␈αlevel␈αof␈αcomputer␈αchess␈α
is
␈↓ ↓Hbased␈α
on␈α
the␈α
incorporation␈α
of␈α
certain␈α
intellectual␈α
mechanisms␈α
in␈α
the␈α
programs.␈αSome␈αimprovement
␈↓ ↓Hcan␈α
be␈α
made␈α
by␈α
further␈α
refinement␈α
of␈α
the␈α
heuristics␈α
in␈α
the␈α
programs,␈α
but␈α
probably␈α
master␈α
level␈α
chess
␈↓ ↓Hawaits␈α∞the␈α∞ability␈α∞to␈α∞put␈α∞general␈α∞configuration␈α∞patterns␈α∞into␈α∞the␈α∞programs␈α∞in␈α
an␈α
easy␈α
and␈α
flexible
␈↓ ↓Hway.␈α∩I␈α∩don't␈α∩see␈α∩how␈α∩to␈α∩set␈α∩a␈α∩date␈α∩by␈α∩which␈α∩this␈α∩problem␈α∩must␈α⊃be␈α⊃solved␈α⊃in␈α⊃order␈α⊃to␈α⊃avoid
␈↓ ↓Hdisappointment␈α
in␈α
the␈α
field␈α
of␈α
artificial␈α
intelligence␈α
as␈α
a␈α
whole.

␈↓ α_4.␈αLighthill␈αdiscusses␈αthe␈α␈↓↓combinatorial␈αexplosion␈↓␈αproblem␈αas␈αthough␈αit␈αwere␈αa␈αrelatively␈αrecent
␈↓ ↓Hphenomenon␈α∞that␈α∞disappointed␈α∞hopes␈α∞that␈α∞unguided␈α∞theorem␈α∞provers␈α
would␈α
be␈α
able␈α
to␈α
start␈α
from
␈↓ ↓Haxioms␈α_representing␈α_knowledge␈α_about␈α_the␈α↔world␈α↔and␈α↔solve␈α↔difficult␈α↔problems.␈α↔In␈α↔fact,␈α↔the
␈↓ ↓H␈↓↓combinatorial␈α∩explosion␈↓␈α∩problem␈α∩has␈α∩been␈α∩recognized␈α∩in␈α∩AI␈α∩from␈α∩the␈α⊃beginning,␈α⊃and␈α⊃the␈α⊃usual
␈↓ ↓Hmeaning␈α∞of␈α∞␈↓↓heuristic␈↓␈α
is␈α
a␈α
device␈α
for␈α
reducing␈α
this␈α
explosion.␈α
Regrettably,␈α
some␈α
people␈α
were␈α
briefly
␈↓ ↓Hover-optimistic␈α⊂about␈α⊂what␈α⊂general␈α⊂purpose␈α⊂heuristics␈α⊂for␈α⊂theorem␈α⊂proving␈α⊂could␈α⊂do␈α⊂in␈α∂problem
␈↓ ↓Hsolving.

␈↓ ↓HDid␈α
We␈α
Deserve␈α
It?

␈↓ α_Lighthill␈αhad␈αhis␈αshot␈αat␈αAI␈αand␈αmissed,␈αbut␈αthis␈αdoesn't␈αprove␈αthat␈αeverything␈αin␈α
AI␈α
is␈α
ok.␈α
In
␈↓ ↓Hmy␈α
opinion,␈α
present␈α
AI␈α
research␈α
suffers␈α
from␈α
some␈α
major␈αdeficiencies␈αapart␈αfrom␈αthe␈αfact␈αthat␈αany
␈↓ ↓Hscientists␈α
would␈α
achieve␈α
more␈α
if␈α
they␈α
were␈α
smarter␈α
and␈α
worked␈α
harder.

␈↓ α_1.␈αMuch␈αwork␈αin␈αAI␈αhas␈α
the␈α
"look␈α
ma,␈α
no␈α
hands"␈α
disease.␈α
Someone␈α
programs␈α
a␈α
computer␈α
to␈α
do
␈↓ ↓Hsomething␈α
no␈α
computer␈α
has␈α
done␈α
before␈α
and␈α
writes␈α
a␈α
paper␈α
pointing␈αout␈αthat␈αthe␈αcomputer␈αdid␈αit.
␈↓ ↓HThe␈α⊃paper␈α⊃is␈α⊃not␈α⊃directed␈α⊃to␈α⊃the␈α⊃identification␈α⊃and␈α⊃study␈α⊃of␈α⊃intellectual␈α⊂mechanisms␈α⊂and␈α⊂often
␈↓ ↓Hcontains␈α∂no␈α∂coherent␈α∞account␈α∞of␈α∞how␈α∞the␈α∞program␈α∞works␈α∞at␈α∞all.␈α∞As␈α∞an␈α∞example,␈α∞consider␈α∞that␈α∞the
␈↓ ↓HSIGART␈α
Newsletter␈α
prints␈α
the␈α
scores␈αof␈αthe␈αgames␈αin␈αthe␈αACM␈αComputer␈αChess␈αTournament␈αjust
␈↓ ↓Has␈αthough␈αthe␈αprograms␈αwere␈αhuman␈αplayers␈αand␈αtheir␈αinnards␈αwere␈αinaccessible.␈αWe␈αneed␈αto␈αknow
␈↓ ↓Hwhy␈α
one␈α
program␈α
missed␈α
the␈α
right␈α
move␈α
in␈α
a␈α
position␈α
-␈α
what␈α
was␈α
it␈α
thinking␈α
about␈α
all␈α
that␈α
time?␈α
We
␈↓ ↓Halso␈α∞need␈α∞an␈α∞analysis␈α∞of␈α∞what␈α∞class␈α∞of␈α∞positions␈α∞the␈α∞particular␈α∞one␈α∞belonged␈α∞to␈α∞and␈α∞how␈α∞a␈α
future
␈↓ ↓Hprogram␈α
might␈α
recognize␈α
this␈α
class␈α
and␈α
play␈α
better.

␈↓ α_2.␈α∞A␈α∞second␈α∞disease␈α
is␈α
to␈α
work␈α
only␈α
on␈α
theories␈α
that␈α
can␈α
be␈α
expressed␈α
mathematically␈α
in␈α
the
␈↓ ↓Hpresent␈α
state␈α
of␈α
knowledge.␈α
Mathematicians␈α
are␈α
often␈α
attracted␈αto␈αthe␈αartificial␈αintelligence␈αproblem
␈↓ ↓Hby␈α
its␈α
intrinsic␈α
interest.␈α
Unfortunately␈α
for␈α
the␈α
mathematicians,␈α
however,␈α
many␈α
plausible␈α
mathematical
␈↓ ↓Htheories␈α
with␈α
good␈αtheorems␈αsuch␈αas␈αcontrol␈αtheory␈αor␈αstatistical␈αdecision␈αtheory␈αhave␈αturned␈αout␈αto
␈↓ ↓Hhave␈α→little␈α_relevance␈α_to␈α_AI.␈α_Even␈α_worse,␈α_the␈α_applicability␈α_of␈α_statistical␈α_decision␈α_theory␈α_to
␈↓ ↓Hdiscriminating␈α∃among␈α∃classes␈α∀of␈α∀signals␈α∀led␈α∀to␈α∀the␈α∀mistaken␈α∀identification␈α∀of␈α∀perception␈α∀with
␈↓ ↓Hdiscrimination␈α∞rather␈α∞than␈α∞with␈α∞description␈α∞which␈α∞so␈α∞far␈α∞has␈α∞not␈α∞led␈α∞to␈α∞much␈α∞mathematics.␈α
More
␈↓ ↓Hrecently,␈α∃however,␈α∃problems␈α∃of␈α∀theorem␈α∀proving␈α∀and␈α∀problems␈α∀of␈α∀representation␈α∀have␈α∀led␈α∀to
␈↓ ↓Hinteresting␈α
mathematical␈α
problems␈α
in␈α
logic␈α
and␈α
mathematical␈α
theory␈α
of␈α
computation.

␈↓ α_3.␈α⊂Every␈α⊂now␈α⊂and␈α⊂then,␈α⊂some␈α⊂AI␈α⊂scientist␈α⊂gets␈α∂an␈α∂idea␈α∂for␈α∂a␈α∂general␈α∂scheme␈α∂of␈α∂intelligent
␈↓ ↓Hbehavior␈αthat␈αcan␈αbe␈αapplied␈αto␈αany␈αproblem␈αprovided␈αthe␈αmachine␈αis␈αgiven␈αthe␈αspecific␈αknowledge
␈↓ ↓Hthat␈αa␈αhuman␈αhas␈αabout␈αthe␈αdomain.␈αExamples␈αof␈αthis␈αhave␈αincluded␈αthe␈αGPS␈αformalism,␈αa␈αsimple
␈↓ ↓Hpredicate␈αcalculus␈αformalism,␈αand␈αmore␈αrecently␈αthe␈αPLANNER␈αformalism␈αand␈αperhaps␈αthe␈αcurrent
␈↓ ↓HCarnegie-Mellon␈α⊂production␈α⊂formalism.␈α⊂In␈α⊂the␈α⊂first␈α⊂and␈α⊂third␈α⊂cases,␈α⊂the␈α∂belief␈α∂that␈α∂any␈α∂problem
␈↓ ↓Hsolving␈α
ability␈α
and␈α
knowledge␈α
could␈α
be␈α
fitted␈α
into␈αthe␈αformalisms␈αled␈αto␈αpublished␈αpredictions␈αthat
␈↓ ↓Hcomputers␈αwould␈αachieve␈αcertain␈αlevels␈αof␈αperformance␈αin␈αcertain␈αtime␈αscales.␈αIf␈αthe␈αinventors␈αof␈αthe
␈↓ ↓Hformalisms␈α∂had␈α∂been␈α∞right␈α∞about␈α∞them,␈α∞the␈α∞goals␈α∞might␈α∞have␈α∞been␈α∞achieved,␈α∞but␈α∞regrettably␈α∞they
␈↓ ↓Hwere␈αmistaken.␈αSuch␈αgeneral␈αpurpose␈α
formalisms␈α
will␈α
be␈α
invented␈α
from␈α
time␈α
to␈α
time,␈α
and,␈α
most␈α
likely,
␈↓ ↓Hone␈α
of␈α
them␈α
will␈α
eventually␈α
prove␈α
adequate.␈α
However,␈α
it␈α
would␈α
be␈α
a␈α
great␈α
relief␈α
to␈α
the␈α
rest␈α
of␈α
the
␈↓ ↓Hworkers␈α⊂in␈α⊂AI␈α⊂if␈α⊂the␈α⊂inventors␈α⊂of␈α⊂new␈α⊂general␈α⊂formalisms␈α∂would␈α∂express␈α∂their␈α∂hopes␈α∂in␈α∂a␈α∂more
␈↓ ↓Hguarded␈α
form␈α
than␈α
has␈α
sometimes␈α
been␈α
the␈α
case.

␈↓ α_4.␈αAt␈αpresent,␈αthere␈αdoes␈αnot␈αexist␈αa␈αcomprehensive␈αgeneral␈αreview␈αof␈αAI␈αthat␈αdiscusses␈αall␈αthe
␈↓ ↓Hmain␈α⊃approaches␈α⊃and␈α⊃achievements␈α⊂and␈α⊂issues.␈α⊂Most␈α⊂likely,␈α⊂this␈α⊂is␈α⊂not␈α⊂merely␈α⊂because␈α⊂the␈α⊂field
␈↓ ↓Hdoesn't␈α∂have␈α∂a␈α∂first␈α∂rate␈α∂reviewer␈α∂at␈α∂present,␈α∂but␈α∂because␈α∂the␈α∂field␈α∂is␈α∂confused␈α∂about␈α∂what␈α∞these
␈↓ ↓Happroaches␈α
and␈α
achievements␈α
and␈α
issues␈α
are.␈α
The␈α
production␈α
of␈α
such␈α
a␈α
review␈α
will␈α
therefore␈αbe␈αa
␈↓ ↓Hmajor␈α
creative␈α
work␈α
and␈α
not␈α
merely␈α
a␈α
work␈α
of␈α
scholarship.

␈↓ α_5.␈α∪While␈α∩it␈α∩is␈α∩far␈α∩beyond␈α∩the␈α∩scope␈α∩of␈α∩this␈α∩review␈α∩to␈α∩try␈α∩to␈α∩summarize␈α∩what␈α∩has␈α∩been
␈↓ ↓Haccomplished␈α∞in␈α∞AI␈α∞since␈α
Turing's␈α
1950␈α
paper,␈α
here␈α
is␈α
a␈α
five␈α
sentence␈α
try:␈α
Many␈α
approaches␈α
have
␈↓ ↓Hbeen␈α∃explored␈α∃and␈α∃tentatively␈α∃rejected␈α∃including␈α∀automaton␈α∀models,␈α∀random␈α∀search,␈α∀sequence
␈↓ ↓Hextrapolation,␈αand␈αmany␈αothers.␈αMany␈αheuristics␈αhave␈αbeen␈αdeveloped␈αfor␈α
reducing␈α
various␈α
kinds␈α
of
␈↓ ↓Htree␈α
search;␈α
some␈α
of␈α
these␈α
are␈α
quite␈α
special␈αto␈αparticular␈αapplications,␈αbut␈αothers␈αare␈αgeneral.␈αMuch
␈↓ ↓Hprogress␈αhas␈αbeen␈αmade␈αin␈αdiscovering␈αhow␈αvarious␈αkinds␈αof␈αinformation␈αcan␈αbe␈αrepresented␈αin␈αthe
␈↓ ↓Hmemory␈α⊃of␈α⊃a␈α⊃computer,␈α⊃but␈α⊃a␈α⊂fully␈α⊂general␈α⊂representation␈α⊂is␈α⊂not␈α⊂yet␈α⊂available.␈α⊂The␈α⊂problem␈α⊂of
␈↓ ↓Hperception␈αof␈αspeech␈αand␈αvision␈αhas␈αbeen␈αexplored␈α
and␈α
recognition␈α
has␈α
been␈α
found␈α
feasible␈α
in␈α
many
␈↓ ↓Hinstances.␈α
A␈α
beginning␈α
has␈α
been␈α
made␈α
in␈α
understanding␈α
the␈α
semantics␈α
of␈α
natural␈α
language.

␈↓ α_These␈αaccomplishments␈αnotwithstanding,␈αI␈αthink␈αthat␈αartificial␈αintelligence␈αresearch␈αhas␈αso␈αfar
␈↓ ↓Hbeen␈αonly␈αmoderately␈αsuccessful;␈αits␈αrate␈αof␈αsolid␈αprogress␈αis␈α
perhaps␈α
greater␈α
than␈α
most␈α
social␈α
sciences
␈↓ ↓Hand␈αless␈αthan␈α
many␈α
physical␈α
sciences.␈α
This␈α
is␈α
perhaps␈α
to␈α
be␈α
expected␈α
considering␈α
the␈α
difficulty␈α
of␈α
the
␈↓ ↓Hproblem.

␈↓ ↓H

␈↓ ↓H

␈↓ ↓H
␈↓ α_␈↓ αh␈↓ β8␈↓ ∧λJohn McCarthy - 9 March 1974
␈↓ ↓H