␈↓␈↓ β
ARTIFICIAL INTELLIGENCE LABORATORY
␈↓␈↓ ∧"STANFORD UNIVERSITY
␈↓␈↓ ∧@CALIFORNIA,94305.␈↓↓


␈↓↓November 23rd.



␈↓↓Dear Sirs,


␈↓↓        Bernard␈α∞Williams'␈α∞review␈α∞of␈α∞Dreyfus␈α∞WHAT␈α∞COMPUTERS␈α∞ CAN'T␈α
 DO

␈↓↓was␈α→ very␈α~ welcome,␈α→ because␈α~ workers␈α→ in␈α~ the␈α→ field␈α~ of␈α→ Artificial

␈↓↓Intelligence␈α
[AI]␈α
seem␈α
 unable␈α or␈α
 unwilling␈α
 to␈α
 defend␈α
 themselves␈αagainst

␈↓↓Dreyfus'␈α→ attacks␈α→on␈α→their␈α→work,␈α→and␈α→so␈α→ought␈α→to␈α→be␈α→grateful␈α~for␈α→a

␈↓↓philosopher␈αprepared␈αto␈αchampion␈αthem.

␈↓↓        However,␈αthere␈α
is␈αa␈α
 slight␈α confusion␈α near␈α
 the␈α center␈α
 of␈αWilliams'

␈↓↓discussion␈α∀ that␈α∀ needs␈α∪comment,␈α∀since␈α∀it␈α∪is␈α∀precisely␈α∀the␈α∀elimination␈α∪of

␈↓↓endemuc␈α→miuddle␈α→that␈α→is␈α~the␈α→refreshment␈α→a␈α→ philosopher␈α~might␈α→ have

␈↓↓brought␈α∂to␈α∂the␈α⊂laborers␈α∂in␈α∂the␈α∂mechanical␈α⊂vineyard␈α∂of␈α∂AI.␈α⊂ Dreyfus'␈α∂ thesis

␈↓↓is␈α⊃ that␈α⊃ distinctive␈α∩ human␈α⊃ abilities␈α⊃ cannot␈α∩ be␈α⊃simulated␈α⊃ by␈α∩ a␈α⊃ digital

␈↓↓computer,␈α→ and␈α→Williams␈α→meets␈α→much␈α→of␈α→this␈α→argument␈α→head␈α~on.␈α→ But

␈↓↓throughout␈α~ his␈α~ review␈α→ he␈α~ makes␈α~ use␈α→ of␈α~ a␈α~distinction␈α→ between

␈↓↓"simulating␈α
a␈α
human␈α
activity␈α
",␈α
on␈α∞the␈α
one␈α
hand,␈α
,␈α
in␈α
the␈α
sense␈α∞of␈α
 producing

␈↓↓the␈α≤ relevant␈α≠ behavior␈α≤ in␈α≠ a␈α≤ machine␈α≠(playing␈α≤ chess␈α≠ ,␈α≤say)␈α≠and

␈↓↓"producing␈αthe␈αbehavior␈αin␈αthe␈αway␈αhumans␈αdo␈αit"␈αon␈αthe␈αother.





␈↓␈↓ ¬c1␈↓ 
_



␈↓↓        This␈α∪distinction,␈α∪though␈α∪widely␈α∪accepted␈α∪ in␈α∪ AI,␈α∪ is␈α∪ almost␈α∪wholly

␈↓↓without␈α→ content,␈α→and␈α→moreover␈α→was␈α→not␈α→needed␈α→by␈α→Williams␈α→for␈α_his

␈↓↓excellent␈αattack␈αon␈αDreyfus'␈αmuch␈αpublicised␈αarguments.

␈↓↓        Williams␈α" introduced␈α" the␈α" distinction␈α" (p.␈α"  36)␈α"  when␈α!  he

␈↓↓contrasted␈α∂ AI␈α⊂ and␈α∂Cognitive␈α∂Simulation␈α⊂[CS]␈α∂:␈α∂  where␈α⊂CS␈α∂is␈α∂said␈α⊂to␈α∂take

␈↓↓"tips␈α∞from␈α∞the␈α∞ ways␈α∞ humans␈α∞ actually␈α∞ solve␈α∞ problems".␈α∞  That␈α∞should␈α∞ of

␈↓↓course␈α∩have␈α∩been:␈α∪   "takes␈α∩tips␈α∩from␈α∩the␈α∪ways␈α∩humans␈α∩SAY␈α∪they␈α∩solve

␈↓↓problems"␈α∞since␈α∞no␈α∞one␈α∞has␈α∞the␈α∞least␈α∞idea␈α∞how␈α∞brains␈α∞solve␈α∂problems,␈α∞ nor

␈↓↓even␈α∞any␈α
clear␈α∞idea␈α
of␈α∞what␈α
it␈α∞would␈α
be␈α∞like␈α
to␈α∞know.␈α
 Given␈α∞that␈α∞fact,␈α
the

␈↓↓AI-CS␈α∞distinction␈α∞begins␈α∂to␈α∞look␈α∞rather␈α∞ weak␈α∂ .␈α∞Next␈α∞ Williams␈α∞ says␈α∂ ".␈α∞  .

␈↓↓.␈α
 for␈αAI␈α
researchers␈α
the␈αaim␈α
is␈αnot␈α
to␈α
solve␈αproblems␈α
"the␈αway␈α
we␈α
do",␈αbut

␈↓↓just␈αto␈αget␈αa␈α machine␈α to␈α solve␈αproblems"(p.␈α 38).

␈↓↓        By␈α∂ his␈α∞ suggestion␈α∂ that␈α∞ one␈α∂ MIGHT␈α∞try␈α∂to␈α∞get␈α∂a␈α∞machine␈α∂to␈α∞solve

␈↓↓problems␈α∞the␈α∞way␈α∞ we␈α∞ do,␈α
 Williams␈α∞ endorses␈α∞ the␈α∞ distinction␈α∞again,␈α
 and

␈↓↓is␈α⊂also␈α⊂utterly␈α∂unfair␈α⊂to␈α⊂the␈α∂motivation␈α⊂of␈α⊂many,␈α∂if␈α⊂not␈α⊂most,␈α⊂ AI␈α∂ workers.

␈↓↓For␈α⊂ their␈α∂ aim␈α⊂ is␈α∂ to␈α⊂ simulate␈α∂  intelligent␈α⊂problem-solving␈α∂ activity␈α⊂as␈α∂a

␈↓↓way␈α↔of␈α_understanding␈α↔it,␈α↔not␈α_just␈α↔as␈α↔a␈α_way␈α↔of␈α↔solving␈α_the␈α↔problems.

␈↓↓Moreover␈α∞,␈α∞at␈α
present␈α∞that␈α∞is␈α
the␈α∞ only␈α∞way␈α
 in␈α∞ which␈α∞we␈α∞CAN␈α
understand

␈↓↓mysterious␈α∂processes␈α∂like␈α⊂thinking␈α∂:␈α∂by␈α∂simulating␈α⊂them.␈α∂ It␈α∂may␈α⊂be␈α∂second

␈↓↓best␈α∂but␈α∂it␈α∂is␈α∂ all␈α∂ there␈α∂ is.␈α∂ And␈α∂ when␈α∂ Williams␈α∂ writes␈α∂ near␈α∂ the␈α∂ end

␈↓↓(p.␈α 40)␈α".␈α
 .␈α .␈α and␈α
if␈αconstruct␈α[an␈αartificial␈α
organism]␈αis␈αwhat␈α
we␈αdo.␈α .␈α
 .␈α .

␈↓↓.␈α  then␈αwe␈αunderstand␈αit"␈αhe␈αseems␈αto␈αappreciate␈αthe␈αlast␈αpoint.



␈↓␈↓ ¬c2␈↓ 
_



␈↓↓        But␈α⊃ immediately␈α⊃ after␈α⊃ the␈α⊃ last␈α⊃ quotation␈α⊃ Williams␈α⊃falls␈α⊃straight

␈↓↓back␈α
into␈α
the␈α
muddle␈α
in␈αhis␈α
peroration␈α
".␈α
 .␈α
 .␈α This␈α
is␈α
the␈α
direction␈α
 in␈αwhich

␈↓↓in␈α
which␈α
content␈α
can␈α
be␈α
found␈α
for␈α
the␈α∞notion␈α
that␈α
a␈α
machine.␈α
 .␈α
 .␈α
 .␈α
 .␈α
 .␈α∞ .␈α
 .

␈↓↓might␈α∪solve␈α∀problems␈α∪.␈α∀ .␈α∪ .␈α∀  .␈α∪  .␈α∪as␈α∀ we␈α∪ do".␈α∀   But␈α∪ this␈α∀ is␈α∪ merely

␈↓↓rhetorical,␈αfor␈α
he␈αhas␈α
shown␈αno␈α
direction␈αat␈α
all.␈α Indeed,␈α
if␈αhe␈α
had␈α,␈α
he␈αwould

␈↓↓have␈α∃given␈α∃conceptual␈α∃thinking␈α∃ a␈α∃ gigantic␈α∃ push␈α∃forward,␈α∃since␈α∃at␈α∃the

␈↓↓present␈αtime␈αno␈αone␈αhas␈αany␈αserious␈αsuggestion␈αas␈αto␈αwhat␈αit␈αwould␈α
be␈α like

␈↓↓to␈α
 discover␈α
the␈α
 "language␈α
 in␈α∞which␈α
the␈α
brain␈α
processes␈α
information"␈α∞,␈α
and

␈↓↓hence␈α
how␈αwe␈α
"really"␈αsolve␈α
problems"?␈α
   To␈α see␈α
 this␈α one␈α
 only␈α
 has␈α to

␈↓↓realise␈α∞ the␈α∞ problem␈α∞of␈α∞discovering␈α∂the␈α∞language␈α∞in␈α∞which␈α∞a␈α∂computer␈α∞was

␈↓↓really␈α∞processing␈α
a␈α∞problem,␈α
if␈α∞one␈α∞had␈α
no␈α∞more␈α
to␈α∞go␈α
on␈α∞than␈α∞the␈α
contents

␈↓↓of␈α∂ the␈α⊂ computer's␈α∂ registers.␈α⊂  It␈α∂ would␈α∂ be␈α⊂ an␈α∂ almost␈α⊂impossible␈α∂code

␈↓↓cracking␈α∞job,␈α∞even␈α∞though␈α∞we␈α∂know␈α∞all␈α∞the␈α∞high␈α∞ level␈α∂computer␈α∞ languages

␈↓↓to␈α which␈αthe␈αchanges␈αof␈αstate␈αcould␈αconceivably␈αcorrespond.␈α In␈αthe␈α
case␈αof

␈↓↓the␈αbrain␈αwe␈αhave␈αno␈αsuch␈αinformation.

␈↓↓        Williams␈α≠seems␈α≠in␈α≠his␈α≠ review␈α~ to␈α≠ give␈α≠ his␈α≠ blessing␈α≠ to␈α~ a

␈↓↓fundamental␈α⊃ muddle␈α⊃ in␈α∩AI:␈α⊃  he␈α⊃could␈α∩have␈α⊃ignored␈α⊃it␈α∩,␈α⊃since␈α⊃he␈α∩did␈α⊃not

␈↓↓need␈α∂it␈α∂to␈α∂deal␈α∂with␈α∂Dreyfus,␈α∞or␈α∂have␈α∂cleared␈α∂it␈α∂up.␈α∂ Instead␈α∂ he␈α∞endorsed

␈↓↓it.

␈↓↓        One␈α
 additional␈α
point␈α
is␈α
germane:␈α
  Williams␈α
refers␈α
more␈α
than␈α
once␈α
to

␈↓↓the␈α↔"result"␈α_that␈α↔any␈α↔analog␈α_process␈α↔can␈α↔be␈α_ simulated␈α↔ on␈α_ a␈α↔digital





␈↓␈↓ ¬c3␈↓ 
_



␈↓↓computer.␈α⊃  This␈α⊃ is␈α∩ very␈α⊃ important␈α⊃ to␈α∩ any␈α⊃discussion␈α⊃of␈α∩Dreyfus,␈α⊃who

␈↓↓holds(1)␈α
that␈α
the␈α
human␈α
activities␈α
in␈α
question␈α
cannot␈α
be␈α
simulated␈α
by␈α
digital

␈↓↓means␈α_,␈α_though␈α↔(2)␈α_they␈α_can␈α_perhaps␈α↔be␈α_simulated␈α_by␈α_analog␈α↔means.

␈↓↓Clearly␈α⊂ this␈α∂ "result"␈α⊂ closes␈α⊂ the␈α∂ gap␈α⊂ between␈α⊂Dreyfus'␈α∂ (1)␈α⊂and␈α⊂(2)␈α∂to

␈↓↓vanishing␈αpoint␈α,␈αand␈αso␈αsuggests␈αDreyfus␈αis␈αcontradicting␈αhimself.␈α .

␈↓↓        But␈αthere␈αmay␈α
be␈αa␈αloophole␈αin␈α
 this␈α anti-Dreyfus␈α argument:␈α
for␈αlike

␈↓↓all␈α∀such␈α∪formal␈α∀proofs,␈α∪the␈α∀result␈α∪inquestion␈α∀holds␈α∪only␈α∀for␈α∪the␈α∀class␈α∪of

␈↓↓objects␈α→ defined␈α→ in␈α→ the␈α→ proof.␈α_  There␈α→ may␈α→ well␈α→ be␈α→entities␈α_ or

␈↓↓systems␈α⊃ that␈α⊃ are␈α∩ analog␈α⊃ devices␈α⊃in␈α⊃the␈α∩common␈α⊃sense␈α⊃meaning␈α∩of␈α⊃the

␈↓↓term,␈α∞though␈α∞ are␈α∞ not␈α∞ covered␈α∞ by␈α
 the␈α∞ result.␈α∞  In␈α∞particular,␈α∞ it␈α∞ is␈α
 still

␈↓↓an␈α
 open␈α
 question␈α
 whether␈α
 or␈α not␈α
 the␈α
three-body␈α
problem␈α
can␈α
even␈αbe

␈↓↓recursively␈α approximated␈α by␈α digital␈αmeans.␈α If␈αit␈αshould␈αturn␈αout␈αnot␈α
to␈αbe

␈↓↓so,␈α
then␈α
any␈α∞three␈α
bodies,␈α
in␈α
virtue␈α∞of␈α
their␈α
mutual␈α
gravitational␈α∞attraction␈α
,

␈↓↓would␈α
form␈αa␈α
device␈α
not␈αsimulable␈α
by␈α
a␈αdigital␈α
computer.␈α
  It␈αmay␈α
be␈α
a␈αlittle

␈↓↓hard␈α
to␈α
see␈α
how␈α∞such␈α
a␈α
device␈α
could␈α∞be␈α
used␈α
to␈α
simulate␈α∞ human␈α
 behavior

␈↓↓but␈α⊂ it␈α⊂would␈α⊂  constitute␈α⊂  an␈α⊂  important␈α⊂ theoretical␈α⊂ looophole␈α⊂ in␈α∂ that

␈↓↓particular␈αargumant␈αagainst␈αDreyfus'␈αgeneral␈αclaims.

␈↓↓Sincerely,



␈↓↓Yorick Wilks.




␈↓↓To: The editors
␈↓↓New York Review of Books.



␈↓␈↓ ¬c4␈↓ 
_