COMMENT ⊗   VALID 00020 PAGES
C REC  PAGE   DESCRIPTION
C00001 00001
C00003 00002	.DEVICE XGP
C00004 00003	Epist. problems of AI
C00005 00004	.once center
C00006 00005	.once center
C00007 00006	.once center
C00008 00007	.once center
C00009 00008	.begin center
C00010 00009	.once center
C00011 00010	.once center
C00012 00011	.once center
C00013 00012	.once center
C00014 00013	.once center
C00015 00014	.ONCE CENTER
C00016 00015	.ONCE CENTER
C00017 00016	.ONCE CENTER
C00018 00017	.ONCE CENTER
C00019 00018	.ONCE CENTER
C00020 00019	.ONCE CENTER
C00021 00020	  %2denotation(Safe1) = safe1%1
C00022 ENDMK
C⊗;
.DEVICE XGP
.
.
.FONT 1 "NGB40"
.FONT 2 "BDI40"
.FONT 3 "MATH40"
.FONT 4 "MATHX[PUB,PAT]"
.TURN ON "%,α"
.
.ODDLEFTBORDER←EVENLEFTBORDER←700
.PAGE FRAME 53 HIGH 93 WIDE
.AREA TEXT LINES 2 TO 53
.at "⊗" ⊂"%4α⊗%*"⊃
.PLACE TEXT
.NOFILL; PREFACE 0
.SELECT 1
Epist. problems of AI

Epistemology and heuristics

First order logic as a language for stating facts

Epistemology is theory of knowledge

result formalism

My object is to establish the importance of separate 
consideration of epistemological problems, identify 
some of the problems, and show a little progress.
.next page
.once center
EPISTEMOLOGICAL ADEQUACY OF A FORMALISM


 ⊗ THE FACTS ARE EXPRESSIBLE IN THE FORMALISM

 ⊗ SPECIFIC FACTS AND GENERAL COMMON SENSE FACTS

 ⊗ "I SHOULD DO X" IS INFERABLE IF AND ONLY IF IT FOLLOWS

 ⊗ THE LAST POINT MAY HAVE TO BE WEAKENED
.next page
.once center
FIRST ORDER LOGIC


 ⊗ PREDICATE CALCULUS + FUNCTIONS, EQUALITY, 
  CONDITIONAL EXPRESSIONS + ...

 ⊗ THERE ARE MANY WAYS OF FORMALIZING THE SAME FACTS

 ⊗ PREDICATE CALCULUS AS PROGRAMMING LANGUAGE IS A 
   SEPARATE ISSUE

 ⊗ FIRST ORDER LOGIC IS IN SOME SENSE SEMANTICALLY UNIVERSAL

 ⊗ INFORMATION IS MORE EASILY CREATED AND TRANSMITTED IN 
  LOGICAL FORM

 ⊗ INFORMATION IS USED FASTEST IN HARDWARE OR PROGRAM FORM

 ⊗ INTERMEDIATE FORMS EXIST
.next page
.once center
THE EPISTEMOLOGICAL PART OF AI (COGNOLOGY)


 ⊗ WHAT FACTS ABOUT THE WORLD ARE AVAILABLE TO AN OBSERVER

 ⊗ WHAT MODES OF REASONING LEAD TO CORRECT CONCLUSIONS AND 
   USEFUL CONJECTURES

 ⊗ WHAT GENERAL INFORMATION ABOUT THE WORLD CONSTITUTES COMMON 
   SENSE
.next page
.once center
ADVANTAGES OF SEPARATING EPISTEMOLOGY FROM HEURISTICS


 ⊗ MANY EPISTEMOLOGICAL PROBLEMS ARE TASK INDEPENDENT

 ⊗ ONE EPISTEMOLOGICAL FORMALISM CAN SUPPORT A VARIETY OF 
   HEURISTIC APPROACHES

 ⊗ CONCENTRATING ON EPISTEMOLOGY MAKES SOLVING THE 
   EPISTEMOLOGICAL PROBLEMS MORE PROBABLE

 ⊗ AI WORKERS CHOOSE EPISTEMOLOGICALLY TRIVIAL FORMALISMS, BUT
   MANY REAL PROBLEMS ARE EPISTEMOLOGICALLY NON-TRIVIAL
.next page
.begin center
THE SHORTEST PATH TO HUMAN LEVEL INTELLIGENCE
PROBABLY INVOLVES SOLVING THE EPISTEMOLOGICAL
PROBLEMS
.end


 ⊗ CONCURRENT EVENTS

 ⊗ KNOWLEDGE BELIEF AND DESIRES

 ⊗ LOCATION AND SHAPES OF OBJECTS

 ⊗ MATERIALS AND THE CONSTRUCTION AND DESTRUCTION OF OBJECTS

 ⊗ WHAT ARE THE SIMPLEST HYPOTHESES
.next page
.once center
EPISTEMOLOGY IN PHILOSOPHY AND IN COGNOLOGY


 ⊗ GENERAL PROBLEMS THE SAME

 ⊗ WHAT INFORMATION IS ACTUALLY AVAILABLE TO A GIVEN OBSERVER

 ⊗ ANALYTIC PHILOSOPHERS HAVE USEFUL FORMALIZATON OF NECESSITY,
   KNOWLEDGE AND BELIEF AND SOME FORMALIZATION OF ACTION

 ⊗ AI REQUIREMENTS SHOULD HELP FOCUS THE PHILOSOPHY
.next page
.once center
INTERACTION WITH OTHER STRIVERS


 ⊗ CO-OPERATION AND COMPETITION

 ⊗ NEED IT KNOW THEIR GOALS OR ONLY STIMULUS-RESPONSE?

 ⊗ "IT WILL DO WHAT IT THINKS WILL ACHIEVE ITS GOALS"

 ⊗ DOES IT LIKE ME?  WHAT DOES IT THINK I WANT?

 ⊗ "HE WILL GIVE ME A GOOD GRADE IF I DO GOOD WORK"
.next page
.once center
KNOWLEDGE ABOUT KNOWLEDGE


 ⊗ THE AIRLINE CLERK KNOWS THE SCHEDULE

 ⊗ THE TELEPHONE NUMBERS ARE IN THE BOOK

 ⊗ WHEN I GET TO LIMA, INFORMATION WILL BE AVAILABLE IN 
   ENGLISH ABOUT WHERE TO TAKE MY NEXT PLANE

 ⊗ I DON'T KNOW WHETHER PRESIDENT CARTER IS SITTING OR 
   STANDING, AND THINKING WON'T HELP
.next page
.once center
CONCURRENT VS. SEQUENTIAL EVENTS


 ⊗ AI WORK HAS BEEN BASED ON SEQUENTIAL EVENTS WITH DISCRETE TIME

 ⊗  %2s%3'%2 = result(e,s)%1

 ⊗ HOW TO EXPRESS PARTIAL INFORMATION ABOUT CONCURRENT PROCESSES

 ⊗ PARALLEL PROGRAMS?  INDETERMINATE PARALLEL PROGRAMS?  
   PETRI NETS?

 ⊗ CONTINUOUS AND DISCRETE EVENTS
.next page
.once center
OBJECTS, MATERIALS AND SPACE


 ⊗ WHAT DO WE KNOW ABOUT SHAPES OF OBJECTS

 ⊗ WHAT DO WE KNOW ABOUT A PERSON WHEN WE SEE ONLY PART OF 
   HIS FRONT

 ⊗ WHAT DO WE KNOW WITH OUR EYES OPEN AND WITH OUR EYES CLOSED

 ⊗ HOW DO YOU KNOW A RAT CAN'T FIT THROUGH THAT HOLE?

 ⊗ A STATUE OF A WOMAN CAN BE CARVED FROM THAT ELEPHANT TUSK
.NEXT PAGE
.ONCE CENTER
OBJECTS AND APPEARANCES


 ⊗ A DOG IS ONE THING

 ⊗ AN APPEARANCE OF A DOG IS ANOTHER

 ⊗ AI NEEDS BOTH AND THEIR RELATIONS

 ⊗ THREE DIMENSIONAL PATTERNS "PROJECT" INTO TWO 
   DIMENSIONAL PATTERNS
.NEXT PAGE
.ONCE CENTER
CAUSALITY AND ABILITY


 ⊗ THE CAUSE OF THE BROKEN WINDOW WAS THE STONE THROWN 
   THROUGH IT

 ⊗ A COW CAN'T JUMP OVER THE MOON (PHYSICAL)

 ⊗ A COW CAN'T BREAK THE ELECTRIC FENCE BY PUSHING A LOG 
   AGAINST IT (INTELLECTUAL)

.NEXT PAGE
.ONCE CENTER
FRAME PROBLEM


 ⊗ HOW TO STATE WHAT REMAINS UNCHANGED BY AN ACTION

 ⊗ IF THERE ARE %2m%1 OBJECTS AND %2n%1 ACTIONS, %2mn%1 SENTENCES ARE
   REQUIRED WITHOUT FRAMES AND ONLY %2n%1 WITH FRAMES.  BUT WHAT
   GOES WITH WHAT FRAME?

 ⊗ AS A TECHNICAL DEVICE, FRAMES WORK, BUT DON'T SEEM TO
   CORRESPOND TO HUMAN THOUGHT.

 ⊗  %2∀x l s.(at(x,l,result(go(x,l),s)) 
      ∧ ∀y l%3'%2. (y≠x ⊃ at(y,l%3'%2,result(go(x,l),s)) ≡ at(y,l%3'%2,s)))%1

 ⊗ MINSKY SWIPED THE NAME FOR SOMETHING QUITE DIFFERENT
.NEXT PAGE
.ONCE CENTER
ADDITIONAL PROBLEMS



 ⊗ FRAME PROBLEM - HOW TO STATE WHAT REMAINS UNCHANGED 
   AFTER AN EVENT

 ⊗ QUALIFICATION PROBLEM - HOW TO STATE THAT "OTHER THINGS 
   BEING EQUAL, A BOAT CAN BE USED TO CROSS A RIVER"
.NEXT PAGE
.ONCE CENTER
CIRCUMSCRIPTION


 ⊗ THE %2P%1'S THAT MUST EXIST ARE ALL THE %2P%1'S THERE ARE

 ⊗ %2a, b, %1AND %2c %1ARE %2P%1's

 ⊗ %2f(x) %1AND %2g(x,y) %1TAKE %2P%1'S INTO %2P%1'S

 ⊗ %2P(a) ∧ P(b) ∧ P(c) ∧ (∀x)(P(x) ⊃ P(f(x))) ∧ (∀x y)(P(x)
   ∧ P(y) ⊃ P(g(x,y)))

 ⊗ %2qF(a) ∧ qF(b) ∧ qF(c) ∧ (∀x)(qF(x) ⊃ qF(f(x)))
   ∧ (∀x y)(qF(x) ∧ qF(y) ⊃ qF(g(x,y))) ⊃ (∀x)(qF(x) ⊃ P(x))%1

.NEXT PAGE
.ONCE CENTER
CONCEPTS AS OBJECTS


 @ %2combination(safe1) = "35-27-45"%1

 ⊗ %2combination(safe2) = "35-27-45"%1

 @ %2*knows(pat,combination(safe1))%1

 @ %2*knows(pat,combination(safe2))%1

 ⊗ %2knows(pat,Combination(Safe1))%1

.NEXT PAGE
 ⊗ %2denotation(Safe1) = safe1%1

 @ %2denotes(Safe1,safe1)%1

 ⊗ %2(∀X)(denotation Combination X = combination denotation x)%1

 ⊗ %2true Know(Pat, Combination Safe1)%1

 @ %2true K(Mike,Know(Pat,Combination Safe1))%1

 ⊗ %2true P ∧ true Q ≡ true(P and Q)%1

 ⊗ %2P and Q =? Q and P%1