/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␈↓Comments on "The State of Technology in Artificial Intelligence" by Duda, Nilsson, and Raphael

␈↓ ↓H␈↓␈↓ α_As␈αa␈αsummary␈αof␈αthe␈αstate␈αof␈αresearch␈αin␈αartificial␈αintelligence,␈αthis␈αpaper␈αreminds␈αme␈αof␈αa
␈↓ ↓H␈↓scene␈αin␈αa␈αWestern␈α
novel␈αin␈αwhich␈αthe␈αbad␈α
guys␈αare␈αmaking␈αtheir␈α
victim␈αdance␈αby␈αshooting␈αat␈α
his
␈↓ ↓H␈↓feet.␈α
 Each␈α
one-paragraph␈α
capsuled␈α
accomplishment␈α
represents␈αone␈α
leap␈α
in␈α
the␈α
air␈α
in␈α
response␈αto␈α
a
␈↓ ↓H␈↓demand␈α
for␈α
a␈αlist␈α
of␈α
the␈α
breakthroughs␈αaccomplished␈α
in␈α
the␈α
last␈αfiscal␈α
year.␈α
 The␈α
style␈αhas␈α
evolved
␈↓ ↓H␈↓by␈α∂survival␈α⊂of␈α∂the␈α⊂fittest␈α∂in␈α∂an␈α⊂atmosphere␈α∂of␈α⊂competition␈α∂for␈α∂bureaucratic␈α⊂favor.␈α∂ As␈α⊂such␈α∂it
␈↓ ↓H␈↓allows␈α
the␈α
recipient␈αto␈α
read␈α
the␈α
article␈αbetween␈α
telephone␈α
calls␈α
at␈αthirty␈α
seconds␈α
per␈αparagraph.␈α
 At
␈↓ ↓H␈↓the␈α∞end␈α
he␈α∞can't␈α
decide␈α∞whether␈α
it␈α∞is␈α∞a␈α
total␈α∞fake,␈α
great␈α∞work␈α
or␈α∞somewhere␈α
in␈α∞between,␈α∞but␈α
he
␈↓ ↓H␈↓knows␈α
he␈α
has␈α
to␈α
make␈α
up␈α
his␈α
mind␈α
in␈α
a␈αfew␈α
months␈α
on␈α
whether␈α
to␈α
fund␈α
it␈α
but␈α
also␈α
knows␈αhe␈α
can't
␈↓ ↓H␈↓or␈α
won't␈α
spend␈α
more␈αthan␈α
another␈α
hour␈α
in␈αunderstanding␈α
it␈α
better.␈α
 What␈αhe␈α
␈↓↓can␈↓␈α
do␈α
is␈α
point␈αto
␈↓ ↓H␈↓the␈αseventeenth␈αaccomplishment␈αand␈αask␈αhis␈αunderlings,␈α"Tell␈αme␈αmore␈αabout␈αthat␈αone"␈αor␈α"What
␈↓ ↓H␈↓are we doing now different from what my predecessor was supporting three years ago?".

␈↓ ↓H␈↓␈↓ α_Different␈αstyles␈α
of␈αpresentation␈αhave␈α
been␈αtried␈α
but␈αhaven't␈αshown␈α
themselves␈αfit␈αto␈α
survive.
␈↓ ↓H␈↓The␈α⊃trouble␈α⊃with␈α⊃a␈α⊃general␈α⊃␈↓↓Scientific␈α∩American␈↓␈α⊃style␈α⊃essay␈α⊃on␈α⊃artificial␈α⊃intelligence␈α⊃is␈α∩that␈α⊃if
␈↓ ↓H␈↓honestly␈α
done,␈α
this␈α
year's␈α
essay␈α
reads␈α
very␈α
much␈α
like␈α
last␈α
year's;␈α
I'm␈α
still␈α
getting␈α
royalties␈α
for␈αthe
␈↓ ↓H␈↓book␈α∀version␈α∀of␈α∀a␈α∀␈↓↓Scientific␈α∀American␈↓␈α∀article␈α∪I␈α∀wrote␈α∀eleven␈α∀years␈α∀ago.␈α∀ The␈α∀trouble␈α∪with
␈↓ ↓H␈↓concentrating␈α∪on␈α∪a␈α∪single␈α∪topic␈α∪is␈α∪that␈α∪it's␈α∪very␈α∪risky;␈α∪whole␈α∪programs␈α∪get␈α∪shot␈α∪down␈α∀if␈α∪a
␈↓ ↓H␈↓Congressional␈α∞staffer␈α∞or␈α∞an␈α∞ARPA␈α∞official␈α∞isn't␈α∞turned␈α∞on.␈α∞ Once␈α∞upon␈α∞a␈α∞time,␈α∞DoD␈α
supported
␈↓ ↓H␈↓research␈α∪in␈α∪on␈α∩a␈α∪less␈α∪hectic␈α∩basis,␈α∪remembering␈α∪that␈α∩rare␈α∪breakthroughs␈α∪sometimes␈α∩produce
␈↓ ↓H␈↓decisive␈αmilitary␈α
advantages␈α-␈α
like␈αradar,␈α
code␈αbreaking,␈α
the␈αjet␈α
engine,␈αor␈α
the␈αatomic␈αbomb.␈α
 Now
␈↓ ↓H␈↓three breakthroughs per year per project is the norm.

␈↓ ↓H␈↓␈↓ α_The␈α⊂authors␈α∂of␈α⊂this␈α⊂report␈α∂have␈α⊂a␈α∂wide␈α⊂acquaintance␈α⊂with␈α∂AI␈α⊂work,␈α∂and␈α⊂there␈α⊂is␈α∂good
␈↓ ↓H␈↓evidence␈αthey␈αhave␈αread␈αall␈αthe␈αworks␈αto␈αwhich␈αthey␈αrefer.␈α Nevertheless,␈αthough␈αI␈αhave␈αread␈αless,
␈↓ ↓H␈↓I cannot accept their picture of the field or its history.

␈↓ ↓H␈↓␈↓ α_As␈α∂they␈α∞correctly␈α∂report,␈α∂much␈α∞AI␈α∂research␈α∂consists␈α∞of␈α∂a␈α∂sequence␈α∞of␈α∂pilot␈α∂projects.␈α∞ The
␈↓ ↓H␈↓project␈α⊃need␈α⊂not␈α⊃have␈α⊃a␈α⊂directly␈α⊃applicable␈α⊂goal,␈α⊃but␈α⊃it␈α⊂must␈α⊃show␈α⊂a␈α⊃capability␈α⊃analogous␈α⊂to
␈↓ ↓H␈↓something␈αthat␈αwould␈αbe␈αapplicable.␈α Projects,␈αwhether␈αtheoretical␈αor␈αexperimental,␈αaimed␈αonly␈αat
␈↓ ↓H␈↓discovering facts are not supportable except as a sideline.

␈↓ ↓H␈↓␈↓ α_I␈α
don't␈αagree␈α
with␈α
the␈αhistory␈α
of␈α
artificial␈αintelligence␈α
presented␈α
in␈αthe␈α
paper.␈α
 The␈αphases
␈↓ ↓H␈↓depicted␈αarose␈αpartly␈αas␈αfads␈αand␈αwere␈αamplified␈αby␈αthe␈αneed␈αto␈αwrite␈αan␈αnew␈αgeneral␈αpicture␈αfor
␈↓ ↓H␈↓the sponsors every few years.  At the risk of being out-of-date, I will claim:

␈↓ ↓H␈↓␈↓ α_1.␈α∩We␈α∪have␈α∩not␈α∪learned␈α∩all␈α∩we␈α∪can␈α∩from␈α∪some␈α∩of␈α∩the␈α∪problems␈α∩which␈α∪were␈α∩attacked
␈↓ ↓H␈↓experimentally␈α⊂in␈α⊂the␈α∂late␈α⊂fifties.␈α⊂ There␈α∂is␈α⊂still␈α⊂more␈α∂to␈α⊂be␈α⊂learned␈α∂from␈α⊂chess,␈α⊂from␈α∂theorem
␈↓ ↓H␈↓proving, and learning for example.

␈↓ ↓H␈↓␈↓ α_2.␈α∩The␈α⊃tendency␈α∩for␈α⊃each␈α∩research␈α∩project␈α⊃to␈α∩build␈α⊃a␈α∩complete␈α⊃system␈α∩with␈α∩a␈α⊃"natural
␈↓ ↓H␈↓language"␈α∩front␈α∩end,␈α∪a␈α∩collection␈α∩of␈α∪domain␈α∩dependent␈α∩experts,␈α∪and␈α∩an␈α∩"explainer"␈α∪leads␈α∩to
␈↓ ↓H␈↓superficial work - even if it does make for better demonstrations.

␈↓ ↓H␈↓␈↓ α_3.␈α
It␈α∞is␈α
important␈α∞for␈α
artificial␈α∞intelligence␈α
to␈α∞isolate␈α
technical␈α∞problems␈α
within␈α∞the␈α
subject
␈↓ ↓H␈↓for␈α
separate␈α
treatment.␈α
 For␈α
example,␈α
the␈α
separation␈α
of␈α
the␈α
AI␈α
problem␈α
into␈α∞epistemological␈α
and
␈↓ ↓H␈↓heuristic parts is necessary.
␈↓ ↓H␈↓␈↓ εH␈↓ 91


␈↓ ↓H␈↓␈↓ α_4.␈α∞The␈α∞field␈α∞is␈α∞now␈α∞95%␈α∂experimental␈α∞and␈α∞5%␈α∞theoretical␈α∞if␈α∞one␈α∞excludes␈α∂theoretical␈α∞work
␈↓ ↓H␈↓that isn't really about AI at all.  It should be about 20% theoretical.

␈↓ ↓H␈↓␈↓ α_5. The field requires a renewed concentration on basic research.

␈↓ ↓H␈↓␈↓ α_6.␈α∞The␈α∞connections␈α∞of␈α∞AI␈α∞with␈α∞philosophical␈α∞logic␈α∞will␈α∞make␈α∞a␈α∞large␈α∞contribution␈α∞to␈α∞both
␈↓ ↓H␈↓fields.

␈↓ ↓H␈↓[This␈αdraft␈αwill␈αbe␈αimproved␈αin␈αthe␈αlist␈αof␈αspecific␈αpoints.␈α The␈αfirst␈αpart␈αmay␈αbe␈αrevised␈αif␈αI␈αcan
␈↓ ↓H␈↓be␈αconvinced␈αit␈αpresents␈αan␈αunfair␈αpicture␈αof␈αscience␈αdistorted␈αby␈αbureaucratic␈αnecessity.]␈αThe␈αfile
␈↓ ↓H␈↓is AI[W77,JMC] at SU-AI and this draft was PUBbed on April 15, 1977.