/LMAR=0/XLINE=3/FONT#0=BAXL30/FONT#1=BAXM30/FONT#2=BASB30/FONT#3=SUB/FONT#4=SUP/FONT#5=BASL35/FONT#6=NGR25/FONT#7=MATH30/FONT#8=FIX25/FONT#9=GRKB30
␈↓ α∧␈↓␈↓ u1


␈↓ α∧␈↓␈↓ ∧6␈↓αNotes on Searle's "Notes on artificial intelligence"␈↓

␈↓ α∧␈↓␈↓ αT(These␈α
notes␈α
are␈αdependent␈α
on␈α
posessing␈α
the␈αpaper,␈α
and␈α
part␈α
of␈αthem␈α
were␈α
written␈α
on␈αthe
␈↓ α∧␈↓basis of the first draft).

␈↓ α∧␈↓␈↓ αTI␈α
think␈αthat␈α
suitable␈α
computer␈αprograms␈α
can␈α
␈↓↓understand␈↓.␈α However,␈α
such␈α
a␈αprogram␈α
is␈αnot␈α
a
␈↓ α∧␈↓theory␈α∩in␈α∩itself␈α∩any␈α∩more␈α∩than␈α∩a␈α∩man␈α⊃who␈α∩understands␈α∩is␈α∩a␈α∩theory.␈α∩ Admittedly␈α∩there␈α∩is␈α⊃a
␈↓ α∧␈↓bobtailed␈α
theory␈α
of␈α
understanding␈α
based␈α
on␈αany␈α
program␈α
or␈α
man␈α
who␈α
understands.␈α
 The␈αtheory␈α
is
␈↓ α∧␈↓␈↓↓understanding␈αis␈αthe␈αinternal␈αstructure␈αrequired␈αto␈αaccount␈αfor␈αthe␈αbehavior␈αof␈αthe␈αprogram␈αor␈αman␈↓.
␈↓ α∧␈↓In␈αthe␈α
case␈αof␈αthe␈α
program,␈αthe␈αreader␈α
is␈αinvited␈αto␈α
read␈αthe␈αlisting,␈α
and␈αin␈αthe␈α
case␈αof␈α
the␈αman,
␈↓ α∧␈↓the␈α∞reader␈α∞is␈α∞invited␈α
to␈α∞do␈α∞physiology.␈α∞ Such␈α∞theories␈α
of␈α∞understand␈α∞won't␈α∞be␈α∞very␈α
illuminating
␈↓ α∧␈↓unless␈α∂the␈α⊂program␈α∂is␈α∂more␈α⊂comprehensible␈α∂than␈α⊂any␈α∂of␈α∂today's␈α⊂substantial␈α∂program␈α⊂or␈α∂unless
␈↓ α∧␈↓physiology becomes much easier than it has been both technically and conceptually.

␈↓ α∧␈↓␈↓ αTIt␈α∞is␈α∞much␈α∞more␈α∞reasonable␈α∞to␈α∞regard␈α∞a␈α∞program␈α∞that␈α∞understands␈α∞as␈α∞an␈α∞illustration␈α∂of␈α∞a
␈↓ α∧␈↓theory␈α∂of␈α∂understanding␈α∂and␈α∞an␈α∂existence␈α∂proof␈α∂that␈α∂the␈α∞form␈α∂of␈α∂␈↓↓understanding␈↓␈α∂built␈α∂into␈α∞the
␈↓ α∧␈↓program␈α⊃meets␈α⊃some␈α⊃of␈α⊃our␈α⊃intuitive␈α⊃desiderata␈α∩or␈α⊃at␈α⊃least␈α⊃meets␈α⊃the␈α⊃empirical␈α⊃claims␈α∩of␈α⊃its
␈↓ α∧␈↓designer.␈α To␈αserve␈αas␈αsuch␈αan␈α
illustration,␈αthe␈αprogram␈αmust␈αbe␈αaccompanied␈αby␈α
an␈αexplanation
␈↓ α∧␈↓of how it embodies the concept of understanding being illustrated.

␈↓ α∧␈↓␈↓ αTIt␈αis␈αconceivable␈αthat␈αsomeone␈αwill␈αmake␈αa␈αprogram␈αthat␈αunderstands␈αwithout␈αbeing␈αable␈αto
␈↓ α∧␈↓explain␈α
it␈α
satisfactorily.␈α∞ I␈α
think␈α
that␈α
is␈α∞unlikely␈α
and␈α
would␈α
be␈α∞very␈α
disappointed␈α
should␈α∞it␈α
turn
␈↓ α∧␈↓out␈α⊗that␈α⊗by␈α↔tinkering␈α⊗humanity␈α⊗can␈α⊗produce␈α↔artificial␈α⊗intelligence␈α⊗but␈α↔cannot␈α⊗understand
␈↓ α∧␈↓intelligence even with the aid of the intelligent programs themselves.

␈↓ α∧␈↓␈↓ αTTherefore,␈αit␈αwould␈αbe␈αbetter␈αfor␈αme␈αif␈αSearle␈αwould␈αsplit␈αhis␈αstronger␈αclaim␈αabout␈αAI␈αinto
␈↓ α∧␈↓two␈α⊂versions␈α⊃-␈α⊂one␈α⊂that␈α⊃says␈α⊂programs␈α⊂can␈α⊃potentially␈α⊂understand␈α⊂and␈α⊃the␈α⊂other␈α⊂that␈α⊃such␈α⊂a
␈↓ α∧␈↓program would constitute a theory of understanding all by itself.

␈↓ α∧␈↓1a␈α∂I␈α∂have␈α∂reservations␈α∂about␈α∂Schank's␈α∞program␈α∂and␈α∂also␈α∂about␈α∂Schank's␈α∂doctrines␈α∂about␈α∞what
␈↓ α∧␈↓constitutes a theory of AI.

␈↓ α∧␈↓␈↓ αTI␈αam␈αdoubtful␈αthat␈α
Schank's␈αprogram␈αunderstands,␈αbut␈αI␈α
don't␈αhave␈αa␈αformal␈α
definition␈αof
␈↓ α∧␈↓understanding␈αwith␈αwhich␈αto␈αcompare␈αit.␈α I␈αam␈α
not␈αconvinced␈αby␈αthe␈αdialogs␈αcited,␈αand␈αI␈α
think␈αI
␈↓ α∧␈↓could␈αinvent␈αother␈αquestions␈αusing␈αthe␈αsame␈αvocabulary␈αthat␈αwould␈αbuffalo␈αthe␈αprogram.␈α I␈αdon't
␈↓ α∧␈↓know␈αif␈αthe␈αprogram␈αwould␈αbe␈αable␈αto␈αanswer␈αwhether␈αthe␈αpleased␈αman␈αwould␈αhave␈αaccepted␈αan
␈↓ α∧␈↓offer␈α∂from␈α∞the␈α∂waitress␈α∞to␈α∂eat␈α∞the␈α∂hamburger␈α∞again␈α∂or␈α∞whether␈α∂the␈α∞offended␈α∂man␈α∂would␈α∞have
␈↓ α∧␈↓accepted␈α
an␈α
offer␈αto␈α
uncook␈α
the␈αhamburger.␈α
 I␈α
am,␈α
however,␈αuncertain␈α
whether␈α
these␈αquestions␈α
are
␈↓ α∧␈↓really␈αcriterial␈αfor␈αunderstanding␈αabout␈αrestaurants.␈α What␈αif␈αthe␈αprogram␈αwere␈αasked,␈α"Might␈αthe
␈↓ α∧␈↓offended␈αman␈αhave␈αdecided␈αthat␈αhe␈αwould␈αnever␈αeat␈αagain".␈α I␈αchoose␈αthese␈αoutlandish␈αquestions,
␈↓ α∧␈↓because␈αI␈αsuspect␈αthat␈α
the␈αprogram,␈αlike␈αmany␈α
AI␈αprograms,␈αis␈αfragile.␈α
 It␈αseems␈αto␈αmake␈α
sense␈αif
␈↓ α∧␈↓you␈αrespect␈αit,␈αbut␈αif␈αyou␈αare␈αskeptical␈αenough,␈αyou␈αcan␈αelicit␈αridiculous␈αbehavior.␈α Unfortunately,
␈↓ α∧␈↓this␈αoff-the-cuff␈αskepticism␈αis␈αnot␈αbased␈αon␈αhaving␈αread␈αenough␈αof␈αSchank's␈αpapers␈α
well␈αenough
␈↓ α∧␈↓to␈α
have␈αa␈α
fully␈αinformed␈α
opinion.␈α It␈α
would␈α
be␈αworthwhile␈α
to␈αhave␈α
the␈αopinion␈α
of␈α
someone␈αwho
␈↓ α∧␈↓has␈αread␈αmore,␈αand,␈α
if␈αdoubt␈αremains,␈αto␈α
perform␈αthe␈αexperiment.␈α (I␈α
remember␈αa␈α␈↓↓go␈↓␈αprogram␈α
that
␈↓ α∧␈↓made␈α
rather␈α∞good␈α
moves␈α∞if␈α
you␈α∞made␈α
reasonable␈α
opening␈α∞moves,␈α
but␈α∞collapsed␈α
if␈α∞you␈α
launched
␈↓ α∧␈↓unsound␈α⊃direct␈α⊃attacks␈α⊃on␈α⊃its␈α⊃groups.␈α⊃ This␈α⊂collapse␈α⊃contrasts␈α⊃with␈α⊃the␈α⊃behavior␈α⊃of␈α⊃even␈α⊂the
␈↓ α∧␈↓microcomputer␈α⊃chess␈α⊃games␈α⊃that␈α⊃will␈α⊃punish␈α⊂sufficiently␈α⊃unsound␈α⊃play␈α⊃without␈α⊃regard␈α⊃to␈α⊂any
␈↓ α∧␈↓expectations of sound play).
␈↓ α∧␈↓␈↓ u2


␈↓ α∧␈↓␈↓ αTThere␈α
is␈αa␈α
level␈αof␈α
performance␈α
that␈αwould␈α
leave␈αme␈α
betting␈αthat␈α
Schank's␈α
program␈αreally
␈↓ α∧␈↓understands,␈α∃but␈α∃without␈α∃more␈α∀of␈α∃a␈α∃theory␈α∃of␈α∀understanding,␈α∃I␈α∃wouldn't␈α∃easily␈α∃be␈α∀finally
␈↓ α∧␈↓convinced.

␈↓ α∧␈↓I like the Berkeley answer that the system knows and will defend it against Searle's objections.

␈↓ α∧␈↓1.␈α
 Suppose␈αthat␈α
the␈α
human␈αdoesn't␈α
know␈α
Chinese,␈αonly␈α
the␈αset␈α
of␈α
rules␈αfor␈α
making␈α
responses␈αto
␈↓ α∧␈↓the␈αstrings␈αof␈αChinese␈αcharacters.␈α Suppose␈αfurther␈αthat␈αthe␈αrules␈αare␈αelaborate␈αenough␈αto␈αbehave
␈↓ α∧␈↓like an excellent Chinese scholar.

␈↓ α∧␈↓I␈α⊃would␈α⊃then␈α⊃say␈α⊃that␈α⊃this␈α⊃system,␈α⊃though␈α⊃not␈α⊃the␈α⊃"underlying␈α⊃mental␈α⊃system"␈α⊃of␈α⊃the␈α⊃person
␈↓ α∧␈↓understands␈α∂Chinese.␈α∂ That␈α⊂a␈α∂person␈α∂could␈α⊂carry␈α∂out␈α∂rules␈α∂elaborate␈α⊂enough␈α∂to␈α∂behave␈α⊂like␈α∂a
␈↓ α∧␈↓Chinese␈α∃scholar␈α⊗is␈α∃implausible␈α⊗given␈α∃the␈α⊗limitations␈α∃of␈α⊗human␈α∃data␈α⊗processing␈α∃capability.
␈↓ α∧␈↓Moreover, AI is not ready to formulate the required rules.

␈↓ α∧␈↓2.␈α
 There␈α
are␈α
some␈α
weakly␈α
analogous␈α
cases,␈αhowever.␈α
Some␈α
schizophrenics␈α
are␈α
said␈α
to␈α
have␈αsplit
␈↓ α∧␈↓personalities - the same brain carries out the mental processes of the several personalities.

␈↓ α∧␈↓3.␈α Searle's␈αproblem␈αcan␈αbe␈αmapped␈αinto␈αa␈αpurely␈αcomputer␈αframework.␈α Suppose␈αa␈αprogram,␈αlike
␈↓ α∧␈↓the␈α
proposed␈α
Advice␈α
Taker,␈α
formulates␈αwhat␈α
it␈α
knows␈α
as␈α
sentences␈αof␈α
logic␈α
and␈α
decides␈α
what␈αto␈α
do
␈↓ α∧␈↓by␈αreasoning␈α
in␈αthis␈α
logic.␈α Suppose␈αfurther␈α
that␈αthis␈α
program␈αcan␈α
engage␈αin␈αreasonably␈α
intelligent
␈↓ α∧␈↓dialog␈α⊂-␈α∂in␈α⊂the␈α⊂logical␈α∂language␈α⊂or␈α∂even␈α⊂in␈α⊂English.␈α∂ Suppose␈α⊂further␈α∂that␈α⊂someone␈α⊂gives␈α∂the
␈↓ α∧␈↓program␈α-␈αin␈αlogic␈αor␈αin␈αEnglish␈α-␈αa␈αset␈αof␈αrules␈αfor␈αmanipulating␈αpictographs␈αbut␈αgives␈αno␈αrules
␈↓ α∧␈↓for␈α⊂translating␈α⊂the␈α⊂pictographs␈α⊂into␈α⊂its␈α⊂basic␈α⊂logical␈α⊂language.␈α⊂ Suppose␈α⊂further␈α⊂that␈α⊂the␈α⊂rules
␈↓ α∧␈↓again␈αamount␈αto␈αthe␈αbehavior␈αof␈αa␈αChinese␈αscholar.␈α Then␈αthe␈αinterpreted␈αsystem␈αknows␈αChinese
␈↓ α∧␈↓and many facts about Chinese literature while the basic system doesn't.

␈↓ α∧␈↓Actually,␈αthe␈αsame␈αproblems␈αwould␈αarise␈αif␈αthe␈αinterpreted␈αlanguage␈αwere␈αEnglish,␈αexcept␈αthat␈αthe
␈↓ α∧␈↓basic␈αsystem␈αwould␈αhave␈αto␈αbe␈αrather␈αobtuse␈αto␈αobey␈αthe␈αrules␈αgiven␈αin␈αEnglish␈αfor␈αmanipulating
␈↓ α∧␈↓quoted␈αstrings␈αof␈αcharacters␈αwithout␈αnoticing␈αthat␈αthese␈αstrings␈αcould␈αthemselves␈αbe␈αinterpreted␈αin
␈↓ α∧␈↓its base language.

␈↓ α∧␈↓Searle's␈αpaper␈αpoints␈αup␈αthe␈αfact␈αthat␈αa␈αgiven␈αbrain␈αcould␈αbe␈αhost␈αto␈αseveral␈αminds.␈α These␈αminds
␈↓ α∧␈↓could␈α
either␈α
be␈α
parallel␈α
in␈α
the␈α
brain␈α
in␈αthat␈α
the␈α
␈↓↓hardware␈↓␈α
interprets␈α
them␈α
separately␈α
or␈α
they␈αcan␈α
be
␈↓ α∧␈↓organized␈α
in␈α∞layers␈α
-␈α
where␈α∞one␈α
mind␈α
knows␈α∞how␈α
to␈α
obey␈α∞the␈α
rules␈α
of␈α∞another␈α
with␈α∞or␈α
without
␈↓ α∧␈↓being able to translate the other's sentences into its own language.

␈↓ α∧␈↓If␈α
such␈α
a␈αphenomenon␈α
were␈α
common,␈αordinary␈α
language␈α
would␈α
not␈αmake␈α
the␈α
"category␈αmistake"␈α
of
␈↓ α∧␈↓saying␈α"John␈αsaw␈αhimself␈αin␈αthe␈αmirror",␈αsince␈αit␈αwould␈αnot␈αidentify␈αthe␈αpersonality␈αJohn␈αwith␈α
the
␈↓ α∧␈↓physical␈α∂object.␈α∂ Since␈α∂the␈α∂phenomenon␈α∂probably␈α∂doesn't␈α∂actually␈α∂occur␈α∂except␈α⊂with␈α∂computers,
␈↓ α∧␈↓ordinary␈α∞language␈α
users␈α∞make␈α
this␈α∞mistake␈α∞only␈α
in␈α∞connection␈α
with␈α∞computers␈α∞and␈α
officialdoms.
␈↓ α∧␈↓The␈α∂statement␈α∂"The␈α∞computer␈α∂knows␈α∂X"␈α∂or␈α∞"The␈α∂government␈α∂knows␈α∞X"␈α∂often␈α∂elicits␈α∂the␈α∞reply
␈↓ α∧␈↓"What␈α∞program␈α∞are␈α∞you␈α∞talking␈α∞about,␈α∞do␈α∞you␈α∞mean␈α∞the␈α∞personnel␈α∞program␈α∞that␈α∞has␈α∞just␈α
been
␈↓ α∧␈↓moved from an IBM 370/168 to an IBM 3033?" or "What agency do you mean?".

␈↓ α∧␈↓␈↓ αTI␈αturn␈αnow␈αto␈αthe␈αinterpolation␈αon␈αunderstanding.␈α In␈αmy␈αpaper␈α␈↓↓Ascribing␈αMental␈αQualities
␈↓ α∧␈↓↓to␈α⊃Machines␈↓,␈α⊃I␈α⊃advocate␈α∩ascribing␈α⊃limited␈α⊃beliefs␈α⊃even␈α∩to␈α⊃thermostats,␈α⊃even␈α⊃though␈α∩it␈α⊃doesn't
␈↓ α∧␈↓contribute␈α∞to␈α∞understanding␈α∞them.␈α∞ (It␈α∞might␈α∞for␈α∞a␈α∞person␈α∞with␈α∞a␈α∞sufficiently␈α∞limited␈α
knowledge
␈↓ α∧␈↓faced␈αwith␈α
a␈αslightly␈αcomplex␈α
thermostatic␈αsystem).␈α I␈α
regard␈αsome␈αof␈α
the␈αarguments␈α
against␈αsuch
␈↓ α∧␈↓␈↓ u3


␈↓ α∧␈↓ascription␈αas␈α
resembling␈αproposals␈αto␈α
begin␈αthe␈α
number␈αsystem␈αwith␈α
2␈αon␈αthe␈α
grounds␈αthat␈α
if␈αwe
␈↓ α∧␈↓only␈αhad␈α
to␈αconsider␈αsets␈α
with␈α0␈α
and␈α1␈αelements,␈α
we␈αwouldn't␈α
have␈αany␈αreal␈α
use␈αfor␈αnumbers.␈α
 The
␈↓ α∧␈↓answer␈α∃is␈α∃that␈α∀starting␈α∃with␈α∃2␈α∃would␈α∀give␈α∃the␈α∃general␈α∃laws␈α∀dealing␈α∃with␈α∃numbers␈α∃a␈α∀less
␈↓ α∧␈↓comprehensible␈αform.␈α
 Indeed␈αfor␈α
a␈αlong␈α
time␈α0␈αwas␈α
mistakenly␈αomitted␈α
from␈αthe␈α
number␈αsystem
␈↓ α∧␈↓on␈α
the␈α
grounds␈αthat␈α
it␈α
wasn't␈α
really␈αa␈α
number.␈α
 My␈αintuition␈α
is␈α
that␈α
a␈αgood␈α
theory␈α
of␈α
belief␈αwill
␈↓ α∧␈↓also␈α∞admit␈α∞trivial␈α∞cases␈α∞that␈α∞wouldn't␈α∞in␈α∞themselves␈α∞justify␈α∞the␈α∞theory␈α∞but␈α∞which␈α∞form␈α∞the␈α∞base
␈↓ α∧␈↓case of inductive arguments in the theory.

␈↓ α∧␈↓␈↓ αTI␈α∞agree␈α
with␈α∞Newell␈α∞and␈α
Simon␈α∞that␈α
AI␈α∞should␈α∞try␈α
to␈α∞make␈α
programs␈α∞that␈α∞understand␈α
in
␈↓ α∧␈↓exactly the same sense as humans do, but I am skeptical about the specifics.

␈↓ α∧␈↓␈↓ αT␈↓αIt␈α⊃would␈α⊂be␈α⊃extremely␈α⊃helpful␈α⊂if␈α⊃Searle␈α⊃could␈α⊂put␈α⊃some␈α⊃of␈α⊂his␈α⊃intuitions␈α⊃about␈α⊂the
␈↓ α∧␈↓αinability␈α
of␈αmachines␈α
to␈αunderstand␈α
into␈αa␈α
form␈αthat␈α
challenges␈αthe␈α
possibility␈αof␈α
a␈αspecific
␈↓ α∧␈↓αperformance.␈α∂ Is␈α∂this␈α∂possible?␈α∞ I␈α∂found␈α∂John␈α∂Haugeland's␈α∞challenges␈α∂to␈α∂AI␈α∂very␈α∂helpful␈α∞in
␈↓ α∧␈↓αdeveloping technical problems␈↓.

␈↓ α∧␈↓19 I agree that this is a good question to ask but take the other side.

␈↓ α∧␈↓␈↓ αTAfter␈α
all␈αthe␈α
fulminations␈αof␈α
the␈α
remainder␈αof␈α
the␈αpaper,␈α
I␈α
am␈αstill␈α
uncertain␈αwhether␈α
Searle
␈↓ α∧␈↓thinks␈αa␈αcomputer␈αprogram␈αfor␈αthe␈αPDP-60␈αwritten␈αby␈α(say)␈αRoger␈αSchank␈αIV␈αcould␈αin␈αprinciple
␈↓ α∧␈↓leave␈αJohn␈αSearle␈αIII␈α
undecided␈αwhether␈αa␈αman␈αor␈α
a␈αmachine␈αwas␈αat␈α
the␈αother␈αend␈αof␈αthe␈α
teletype
␈↓ α∧␈↓line,␈α⊂giving␈α⊂John␈α⊂Searle␈α⊂III␈α⊂repeated␈α⊂tries␈α⊃and␈α⊂permission␈α⊂to␈α⊂consult␈α⊂with␈α⊂experts.␈α⊂ I␈α⊃am␈α⊂also
␈↓ α∧␈↓undecided as to whether success in this would count as true intelligence.