L4 Knowledge Representation

CSE 3201
Artificial Intelligence and Machine Learning
Artificial Intelligence
Knowledge Representation
• When we use search to solve a problem we
must
– Capture the knowledge needed to formalize the

problem
– Apply a search technique to solve problem
– Execute the problem solution

Role of KR
• The first step is the role of “knowledge
representation” in AI.
• Formally,
– The intended role of knowledge representation in artificial
intelligence is to reduce problems of intelligent action to
search problems.
• A good description, developed within the conventions

of a good KR, is an open door to problem solving
• A bad description, using a bad representation, is a
brick wall preventing problem solving
A Knowledge-Based Agent
• We previously talked about applications of search but not about
methods of formalizing the problem.
• Now we look at extended capabilities to general logical reasoning.
• Here is one knowledge representation: logical expressions.
• A knowledge-based agent must be able to
– Represent states, actions, etc.
– Incorporate new percepts
– Update internal representations of the world
– Deduce hidden properties about the world
– Deduce appropriate actions
• We will
– Describe properties of languages to use for logical reasoning
– Describe techniques for deducing new information from current
information
– Apply search to deduce (or learn) specifically needed information
The Wumpus World Environment
Percepts
WW Agent Description
• Performance measure
– gold +1000, death -1000
– -1 per step, -10 for using arrow
• Environment
– Squares adjacent to wumpus are
smelly
– Squares adjacent to pit are breezy
– Glitter iff gold is in same square
– Shooting kills wumpus if agent facing
it
– Shooting uses up only arrow
– Grabbing picks up gold if in same
square
– Releasing drops gold in same square
• Actuators
– Left turn, right turn, forward, grab,
release, shoot
• Sensors
–
WW Environment Properties
• Observable? • Static?
– Partial – Yes (for now), wumpus
• Deterministic? and pits do not move
– Yes • Discrete?
• Episodic? – Yes
– Sequential • Single agent?
– Multi (wumpus,
eventually other agents)
Sample Run
Sample Run
Sample Run
Sample Run
Sample Run
Sample Run
Sample Run
Sample Run
Sample Run
Now we look at
• How to represent facts / beliefs

 “There is a pit in (2,2) or (3,1)”
• How to make inferences
 “No breeze in (1,2), so pit in (3,1)”
Representation, Reasoning and Logic
• Sentence: Individual piece of • Semantics: Mapping from sentences to
knowledge facts in the world
- English sentence forms one piece - They define the truth of a sentence in a
of knowledge in English language “possible world”
- Add the values of 2 and 3, store them in
- Statement in C forms one piece of
the memory location indicated by
knowledge in C programming variable a
language • In the language of arithmetic:
• Syntax: Form used to represent x + 2 >= y is a sentence

sentences x2 + y > is not a sentence
- Syntax of C indicates legal x + 2 >= y is true in all worlds where
combinations of symbols the number x + 2 is
- a = 2 + 3; is legal no less than the number y
x + 2 >= y is true in a world where
- a = + 2 3 is not legal
x = 7, y = 1
- Syntax alone does not indicate x + 2 >= y is false in a world where
meaning x = 0, y = 6
Entailment
• There can exist a relationship between items in the language
– Sentences “entail” sentences (representation level)
– Facts “follow” from facts (real world)
• Entail / Follow mean the new item is true if the old items are true
• A collection of sentences, or knowledge base (KB), entail a sentence
– KB |= sentence
– KB entails the sentence iff the sentence is true in all worlds where the KB is
true
Entails
Sentences Sentence
S S
e e
Representation m m
a a
n n
World t t
i i
c c
s s
Follows
Facts Fact
Entailment Examples
• KB • KB
– The Giants won – CookLectures ->
– The Reds won TodayIsTuesday v
TodayIsThursday
• Entails – - TodayIsThursday
– Either the Giants won or – TodayIsSaturday -> SleepLate
the Reds won – Rainy -> GrassIsWet
– CookLectures v TodayIsSaturday
• KB – - SleepLate
– To get a perfect score your
• Which of these are correct
program must be turned in
entailments?
today – - Sleeplate
– I always get perfect scores – GrassIsWet
• Entails – - SleepLate v GrassIsWet
– I turned in my program – TodayIsTuesday
– True
today
Models
• Logicians frequently use models, which are
formally structured worlds with respect to which
truth can be evaluated. These are our “possible
worlds”.
• M is a model of a sentence s if s is true in M.
• M(s) is the set of all models of s.
• KB entails s (KB |= s) if and only if M(KB) is a
subset of M(s)
• For example, KB = Giants won and Reds won,
s = Giants won
Entailment in the Wumpus World
• Situation after detecting nothing in [1,1], moving right,

breeze in [2,1]
• Consider possible models for the situation (the region
around the visited squares) assuming only pits, no wumpi
• 3 Boolean choices, so there are 8 possible models
Wumpus Models
Wumpus Models
KB = wumpus world rules + observations

Wumpus Models

Possible conclusion: alpha1 = “[1,2] is safe”
KB |= alpha1, proved by model checking
Wumpus Models
alpha2 = “[2,2] is safe”, KB |= alpha2

Inference
• Use two different ways:
– Generate new sentences that are entailed by KB
– Determine whether or not sentence is entailed by
KB
• A sound inference procedure generates only
entailed sentences
• Modus ponens is sound A, A  B
B
• Abduction is not sound B, A  B

A
• Logic gone bad
Definitions
• A complete inference procedure can generate all entailed sentences
from the knowledge base.
• The meaning of a sentence is a mapping onto the world (a model).
• This mapping is an interpretation (interpretation of Lisp code).
• A sentence is valid (necessarily true, tautology) iff true under all
possible interpretations.
– A V -A
• A could be:
– Stench at [1,1]
– Today is Monday
– 2+3=5
• These statements are not valid.
– A ^ -A
– AVB
• The last statement is satisfiable, meaning there exists at least one
interpretation that makes the statement true. The previous statement
is unsatisfiable.
Logics
• Logics are formal languages for representing information
such that conclusions can be drawn
• Logics are characterized by their “primitives” commitments
– Ontological commitment: What exists? Facts? Objects? Time?
Beliefs?
– Epistemological commitment:
Ontological What areEpistemological
the states of knowledge?
Language
Commitment Commitment
Propositional logic facts true/false/unknown
First-order logic facts, objects, relations true/false/unknown
facts, objects, relations,
Temporal logic true/false/unknown
times
Probability theory facts value in [0, 1]
Fuzzy logic degree of truth known interval value
Examples
• Propositional logic
– Simple logic
– Symbols represent entire facts
– Boolean connectives (&, v, ->, <=>, ~)
– Propositions (symbols, facts) are either TRUE or
FALSE
• First-order logic
– Extend propositional logic to include
variables, quantifiers, functions, objects
Propositional Logic
• Proposition symbols P, Q, etc., are sentences
• The true/false value of propositions and combinations of
propositions can be calculated using a truth table
• If P and S are sentences, then so are –P, P^Q, PvQ, P->Q, P<->Q
• An interpretation I consists of an assignment of truth values to all
proposition symbols I(S)
– An interpretation is a logician's word for what is often called a “possible
world”
– Given 3 proposition symbols P, Q, and R, there are 8 interpretations
– Given n proposition symbols, there are 2 n interpretations
• To determine the truth of a complex statement for I, we can

– Substitute I's truth value for every symbol
– Use truth tables to reduce the statement to a single truth value
– End result is a single truth value, either True or False
Propositional Logic
• For propositional logic, a row in the truth table is one
interpretation
• A logic is monotonic as long as entailed sentences are
preserved as more knowledge is added
Rules of Inference for Propositional Logic
• Modus ponens A, A  B • Double-negation A

B A
All men are mortal (Man -> Mortal) elimination
Socrates is a man (Man)
-----------------------------------------------
• Unit resolution AvB , B
Socrates is mortal (Mortal) A
A, B Today is Tuesday or Thursday
• And introduction Today is not Thursday
A^ B ---------------------------------------
• Or introduction Today is Tuesday
• Resolution
A
AvBvCvDv... AvB , BvC  A  B, B   C
• And elimination
AvC  A  C
A^ B ^ C ^...^ Z Today is Tuesday or Thursday
Today is not Thursday or tomorrow is Friday
A ----------------------------------------------------------
Today is Tuesday or tomorrow is Friday
Normal Forms
• Other approaches to inference use syntactic operations on
sentences, often expressed in standardized forms
• Conjunctive Normal Form (CNF)
conjunction of disjunctions of literals (conjunction of clauses)
For example, (A v –B) ^ (B v –C v –D)
• Disjunctive Normal Form (DNF)
disjunction of conjunctions of literals (disjunction of terms)
For example, (A ^ B) v (A ^ -C) v (A ^ -D) v (-B ^ -C) v (-B ^ -D)
• Horn Form (restricted)
conjunction of Horn clauses (clauses with <= 1 positive literal)
For example, (A v –B) ^ (B v –C v –D)
Often written as a set of implications:
B -> A and (C ^ D) -> B
Proof methods
• Model checking
– Truth table enumeration (sound and complete for
propositional logic)
• Show that all interpretations in which the left hand side of the rule is
true, the right hand side is also true
– Application of inference rules

• Sound generation of new sentences from old
Proof = a sequence of inference rule applications
Can use inference rules as operators in a standard search algorithm
Wumpus World KB
• Vocabulary • “Pits cause breezes in
– Let Pi,j be true if there is adjacent squres”
a pit in [i,j] – B1,1 <-> P1,2 v P2,1
– Let Bi,j be true if there is – B2,1 <-> P1,1 v P2,2 v P3,1
a breeze in [i,j]
• Sentences
– -P1,1
– -B1,1
– B2,1
An Agent for the Wumpus World
• Imagine we are at a stage in the game where we have had
some experience
– What is in our knowledge base?
– What can we deduce
about the world?
• Example: Finding the wumpus
• If we are in [1,1] and know
– -S11
– S12
– S21
– -S11 -> -W11 & -W12 & -W21
– S12 -> W11 v W12 v W13 v W22
– S21 -> W11 v W21 v W31 v W22
• What can we conclude?
Limitations of Propositional Logic
• Propositional logic cannot express general-purpose
knowledge succinctly
• We need 32 sentences to describe the relationship between
wumpi and stenches
• We would need another 32 sentences for pits and breezes
• We would need at least 64 sentences to describe the effects
of actions
• How would we express the fact that there is only one
wumpus?
• Difficult to identify specific individuals (Mary, among 3)
• Generalizations, patterns, regularities difficult to represent
(all triangles have 3 sides)
First-Order Predicate Calculus
• Propositional Logic uses only propositions (symbols
representing facts), only possible values are True and
False
• First-Order Logic includes:
– Objects: peoples, numbers, places, ideas (atoms)
– Relations: relationships between objects (predicates, T/F
value)
• Example: father(fred, mary)
• Properties: properties of atoms (predicates, T/F value)
Example: red(ball)
– Functions: father-of(mary), next(3), (any value in range)
• Constant: function with no parameters, MARY
FOPC Models
Example
• Express “Socrates is a man” in
• Propositional logic
– MANSOCRATES - single proposition representing
entire idea
• First-Order Predicate Calculus
– Man(SOCRATES) - predicate representing property
of constant SOCRATES
FOPC Syntax
• Constant symbols (Capitalized, Functions with
no arguments)
Interpretation must map to exactly one object
in the world
• Predicates (can take arguments, True/False)
Interpretation maps to relationship or
property T/F value
• Function (can take arguments)
Maps to exactly one object in the world
Definitions
• Term
Anything that identifies an object
Function(args)
Constant - function with 0 args
• Atomic sentence
Predicate with term arguments
Enemies(WilyCoyote, RoadRunner)
Married(FatherOf(Alex), MotherOf(Alex))
• Literals
atomic sentences and negated atomic sentences
• Connectives
(&), (v), (->), (<=>), (~)
if connected by , conjunction (components are conjuncts)
if connected by , disjunction (components are disjuncts)
• Quantifiers 
Universal Quantifier 
Universal Quantifiers
• How do we express “All unicorns speak English” in Propositional Logic?
• We would need to specify a proposition for each unicorn
•  is used to express facts and relationships that we know to be true for
all members of a group (objects in the world)
• Avariable is used in the place of an object
x Unicorn(x) -> SpeakEnglish(x)
The domain of x is the world 
The scope of x is the statement following (sometimes in [])
• Same as specifying
– Unicorn(Uni1) -> SpeakEnglish(Uni1) &
– ...
– Unicorn(Table1) -> Table(Table1) &
– ...
• One statement for each object in the world
• We will leave variables lower case (sometimes ?x)
Notice that x ranges over all objects, not just unicorns.
Existential Quantifier
• This makes a statement about some object (not named)
•  x [Bunny(x) ^ EatsCarrots(x)]
• This means there exists some object in the world (at least
one) for which the statement is true. Same as disjunction
over all objects in the world.
– (Bunny(Bun1) & EatsCarrots(Bun1)) v
– ...
– (Bunny(Table1) & EatsCarrots(Table1)) v
– ...

• What about x Unicorn(x) -> SpeakEnglish(x)?
• Means implication applies to at least one object in the
universe
DeMorgan Rules
• x P  xP
• xP  x P
•  x P  xP
•
 xP  x P
• Example:
xLovesWate rmelon( x)  x LovesWater melon( x)

Other Properties
• (X->Y) <-> -XvY
– Can prove with truth table
• Not true:
– (X->Y) <-> (Y->X)
– This is a type of inference that is not sound
(abduction)
Examples
• All men are mortal
Examples
–  x [Man(x) -> Mortal(x)]
Examples
• Socrates is a man
Examples
– Man(Socrates)
Examples
– Man(Socrates)
• Socrates is mortal
– Mortal(Socrates)
Examples
– Man(Socrates)
• All purple mushrooms are poisonous
Examples
– Man(Socrates)
– x [(Purple(x) ^ Mushroom(x)) -> Poisonous(x)]

Examples
– Man(Socrates)
–  x [(Purple(x) ^ Mushroom(x)) -> Poisonous(x)]
• A mushroom is poisonous only if it is purple
Examples
– Man(Socrates)
–  x [(Purple(x) ^ Mushroom(x)) -> Poisonous(x)]
• A mushroom is poisonous only if it is purple
Examples
– Man(Socrates)
• Allpurple mushrooms are poisonous
• A mushroom
 is poisonous only if it is purple
– x [(Mushroom(x) ^ Poisonous(x)) -> Purple(x)]
Examples
– Man(Socrates)
• A mushroom
• No purple mushroom is poisonous
Examples
– Man(Socrates)
• A mushroom
• No purple
 mushroom is poisonous
– -( x [Purple(x) ^ Mushroom(x) ^ Poisonous(x)])
Examples
• There is exactly one mushroom
Examples
xMushroom( x)  (y ( NEQ ( x, y )   Mushroom( y )))]
– Because “exactly one” is difficult to express we can
use ! Todenote exactly one of a type of object
• Every city has a dog catcher who has been
bitten by every dog in town
Examples
xMushroom( x)  (y ( NEQ ( x, y )   Mushroom( y )))]
– Because “exactly one” is difficult to express we can
use ! Todenote exactly one of a type of object
• Every city has a dog catcher who has been
bitten by every dog in town
– Use City(c), DogCatcher(c), Bit(d,x), Lives(x,c)
a, b[City (a )  cDogCatcher (c)  ( Dog (b)  Lives(b, a )  Bit (b, c))]
Examples
• No human enjoys golf
Examples
x[ Human( x)   Enjoys( x, Golf )
• Some professor that is not a historian writes
programs
Examples
programs
x[Pr ofessor ( x)   Historian ( x)  Writes ( x, Pr ograms )]
• Every boy owns a dog
Examples
programs
x[Pr ofessor ( x)   Historian ( x)  Writes ( x, Pr ograms )]
• Every
xboy
y[ Boyowns
( x)  a dog( x, y )]
Owns
yx[ Boy ( x)  Owns ( x, y )]
– Do these mean the same thing?

– Brothers are siblings
– “Sibling” is reflexive and symmetric
– One’s mother is one’s female parent
Higher-Order Logic
• FOPC quantifies over objects in the universe.
• Higher-order logic quantifies over relations and
functions as well as objects.
– All functions with a single argument return a value of
1 
• x, y [Equal(x(y), 1)]
– Two objects are equal iff all properties applied to
 
them are equivalent
• x, y [(x=y) <-> ( p [p(x) <-> p(y)])]
– Note that we use “=“ as a shorthand for equal,
meaning they are in fact the same object
Additional Operators
• Existential Elimination
–  v [..v..]
– Substitute k for v anywhere in sentence, where k is a
constant (term with no arguments) and does not already
appear in the sentence (Skolemization)
• Existential Introduction
– If [..g..] true (where g is ground term)

– then v [..v..] true (v is substituted for g)
• Universal

Elimination
– x [..x..]
– Substitute M for x throughout entire sentence, where M is
a constant and does not already appear in the sentence
Example Proof
Known: Prove: Lulu is older than Fifi
1. If x is a parent of y, then x is (Older(Lulu, Fifi))
older than y 4. Parent(Lulu, Fifi)
– x,y [Parent(x,y) -> Older(x,y)] – 2,3, Universal Elimination,
2. If x is the mother of y, then x is Modus Ponens
a parent of y 5. Older(Lulu, Fifi)
–  x,y [Mother(x,y) -> Parent(x,y)] – 1,4, Universal Elimination,
3. Lulu is the mother of Fifi Modus Ponens
– Mother(Lulu, Fifi) – We “bind” the variable to a
constant
Example Proof
The law says that it is a crime for an American to
sell weapons to hostile nations.
1) FAx,y,z[(American(x)&Weapon(y)&Nation(z)&
Hostile(z)&Sells(x,z,y)) -> Criminal(x)]
Example Proof
The law says that it is a crime for an American to
sell weapons to hostile nations. The country
Nono, an enemy of America, has some missiles,
and all of its missiles were sold to it by Colonel
West, who is an American.
1) FAx,y,z[(American(x)&Weapon(y)&Nation(z)&
Hostile(z)&Sells(x,z,y)) -> Criminal(x)]
2) EX x [Owns(Nono,x) & Missile(x)]
Example Proof
The law says that it is a crime for an American to sell weapons
to hostile nations. The country Nono, an enemy of America,
has some missiles, and all of its missiles were sold to it by
Colonel West, who is an American.
1) FAx,y,z[(American(x)&Weapon(y)&Nation(z)&Hostile(z)&
Sells(x,z,y)) -> Criminal(x)]
2) EX x [Owns(Nono,x) & Missile(x)]
3) FA x [Owns(Nono,x) & Missile(x)) -> Sells(West, Nono,x)]
4) FA x [Missile(x) -> Weapon(x)]
5) FA x [Enemy(x,America) -> Hostile(x)]
6) American(West)
7) Nation(Nono)
8) Enemy(Nono, America)
9) Nation(America)
Prove: West is a criminal.
Prove: West is a Criminal
1) FAx,y,z[(American(x)& 10) Owns(Nono,M1) & Missile(M1)
Weapon(y)&Nation(z)&
Hostile(z)& Sells(x,z,y)) -> – 2 & Existential Elimination
Criminal(x)]
2) EX x [Owns(Nono,x) &
Missile(x)]
3) FA x [Owns(Nono,x) &
Missile(x)) -> Sells(West,
Nono,x)]
4) FA x [Missile(x) ->
Weapon(x)]
5) FA x [Enemy(x,America) ->
Hostile(x)]
6) American(West)
7) Nation(Nono)
9) Nation(America)
Hostile(z)& Sells(x,z,y)) -> 11) Owns(Nono, M1)
Criminal(x)] – 10 & And Elimination
Missile(x)]
Nono,x)]
Weapon(x)]
Hostile(x)]
6) American(West)
7) Nation(Nono)
9) Nation(America)
Criminal(x)]
12) Missile(M1)
Missile(x)] – 10 & And Elimination
Nono,x)]
Weapon(x)]
Hostile(x)]
6) American(West)
7) Nation(Nono)
9) Nation(America)
Criminal(x)]
12) Missile(M1)
Missile(x)] 13) Missile(M1) -> Weapon(M1)
– 4 & Elimination
Universal Universal Elimination
Nono,x)]
FORALL v []
Weapon(x)] If true for universal variable v,
5) FA x [Enemy(x,America) -> then true for a ground term
Hostile(x)] (term with no variables)
6) American(West)
7) Nation(Nono)
9) Nation(America)
Criminal(x)]
12) Missile(M1)
Missile(x)) -> Sells(West, 14) Weapon(M1)
Nono,x)] – 12, 13, Modus Ponens
Weapon(x)]
Hostile(x)]
6) American(West)
7) Nation(Nono)
9) Nation(America)
Criminal(x)]
12) Missile(M1)
4)
Nono,x)]
FA x [Missile(x) ->
15) Owns(Nono,M1) & Missile(M1) ->
Weapon(x)] Sells(West,Nono,M1)
5) FA x [Enemy(x,America) -> –
Hostile(x)]
3 & Universal Elimination
6) American(West)
7) Nation(Nono)
9) Nation(America)
Criminal(x)]
12) Missile(M1)
4)
Nono,x)]
FA x [Missile(x) ->
15) Owns(Nono,M1) & Missile(M1) ->
Hostile(x)] 16) Sells(West,Nono,M1)
6) American(West) – 10, 15, Modus Ponens
7) Nation(Nono)
9) Nation(America)
Criminal(x)] 12) Missile(M1)
14) Weapon(M1)
Nono,x)] 15) Owns(Nono,M1) & Missile(M1) ->
5) FA x [Enemy(x,America) -> 16) Sells(West,Nono,M1)
Hostile(x)]
6) American(West) 17) American(West) & Weapon(M1) &
7) Nation(Nono) Nation(Nono) & Hostile(Nono) &
9) Nation(America) Sells(West,Nono,M1) ->
Criminal(West)
– 1, Universal Elimination (x West) (y
Hostile(z)& Sells(x,z,y)) ->
11) Owns(Nono, M1)
Missile(x)]
13) Missile(M1) -> Weapon(M1)
3) FA x [Owns(Nono,x) & 14) Weapon(M1)
Missile(x)) -> Sells(West, 15) Owns(Nono,M1) & Missile(M1) ->
Nono,x)]
4) FA x [Missile(x) -> Sells(West,Nono,M1)
Weapon(x)] 16) Sells(West,Nono,M1)
Hostile(x)] 17) American(West) & Weapon(M1) &
6) American(West) Nation(Nono) & Hostile(Nono) &
7) Nation(Nono)
8) Enemy(Nono, America) Sells(West,Nono,M1) ->
9) Nation(America) Criminal(West)
18) Enemy(Nono,America) ->
Hostile(Nono)
11) Owns(Nono, M1)
2) EX x [Owns(Nono,x) & 13) Missile(M1) -> Weapon(M1)
Missile(x)]
3) FA x [Owns(Nono,x) & 14) Weapon(M1)
Missile(x)) -> Sells(West, 15) Owns(Nono,M1) & Missile(M1) ->
Nono,x)]
4) FA x [Missile(x) -> Sells(West,Nono,M1)
Weapon(x)] 16) Sells(West,Nono,M1)
Hostile(x)]
17) American(West) & Weapon(M1) &
6) American(West) Nation(Nono) & Hostile(Nono) &
7) Nation(Nono)
Sells(West,Nono,M1) -> Criminal(West)
9) Nation(America) 18) Enemy(Nono,America) ->
Hostile(Nono)
19) Hostile(Nono)
Weapon(y)&Nation(z)& 11) Owns(Nono, M1)
12) Missile(M1)
Criminal(x)]
2) EX x [Owns(Nono,x) & 13) Missile(M1) -> Weapon(M1)
Missile(x)] 14) Weapon(M1)
3) FA x [Owns(Nono,x) & 15) Owns(Nono,M1) & Missile(M1) ->
Sells(West,Nono,M1)
Nono,x)]
4) FA x [Missile(x) -> 16) Sells(West,Nono,M1)
Weapon(x)] 17) American(West) & Weapon(M1) &
5) FA x [Enemy(x,America) -> Nation(Nono) & Hostile(Nono) &
Hostile(x)]
Sells(West,Nono,M1) -> Criminal(West)
6) American(West)
7) Nation(Nono) 18) Enemy(Nono,America) -> Hostile(Nono)
8) Enemy(Nono, America) 19) Hostile(Nono)
9) Nation(America) 20) American(West) & Weapon(M1) &
Nation(Nono) & Hostile(Nono) &
Sells(West,Nono,M1)
Weapon(y)&Nation(z)& 11) Owns(Nono, M1)
Hostile(z)& Sells(x,z,y)) -> 12) Missile(M1)
Criminal(x)]
13) Missile(M1) -> Weapon(M1)
Missile(x)] 14) Weapon(M1)
3) FA x [Owns(Nono,x) & 15) Owns(Nono,M1) & Missile(M1) ->
Missile(x)) -> Sells(West, Sells(West,Nono,M1)
Nono,x)] 16) Sells(West,Nono,M1)
Weapon(x)]
5) FA x [Enemy(x,America) -> Nation(Nono) & Hostile(Nono) &
Hostile(x)] Sells(West,Nono,M1) -> Criminal(West)
6) American(West) 18) Enemy(Nono,America) -> Hostile(Nono)
7) Nation(Nono) 19) Hostile(Nono)
9) Nation(America)
Nation(Nono) & Hostile(Nono) &
Sells(West,Nono,M1)
21) Criminal(West)
FOPC and the Wumpus World
• Perception rules
–  b,g,t Percept([Smell,b,g],t) -> Smelled(t)
– Here we are indicating a Percept occurring at time t
–  s,b,t Percept([s,b,Glitter],t) -> AtGold(t)
• We can use FOPC to write rules for selecting actions:
– Reflex agent:  t AtGold(t) -> Action(Grab, t)
– Reflex agent with internal state:
 t AtGold(t) & -Holding(Gold,t) ->
Action(Grab, t)
– Holding(Gold,t) cannot be observed, so keeping track of
change is essential
Deducing Hidden Properties
• Properties of locations:
• Squares are breezy near a pit
– Diagnostic rule: infer cause from effect
•  y Breezy(y) ->  x Pit(x) & Adjacent(x,y)
– Causal rule: infer effect from cause
•  x,y Pit(x) & Adjacent(x,y) -> Breezy(y)
• Neither of these is complete
• For example, causal rule doesn’t say whether
squares far away from pits can be breezy
• Definition for Breezy predicate
– Breezy(y) <-> [  Pit(x) & Adjacent(x,y)]
Inference As Search
• Operators are inference rules
• States are sets of sentences
• Goal test checks state to see if it contains query sentence
• AI, UE, MP a common inference pattern, but generate a
huge branching factor
• We need a single, more powerful inference rule
Generalized Modus Ponens
• If we have a rule
– p1(x) & p2(x) & p3(x,y) & p4(y) & p5(x,y) -> q(x,y)
• Each p involves universal / existential quantifiers
• Assume each antecedent appears in KB
– p1(WSU)
– p2(WSU)
– p3(WSU, Washington)
– p4(Washington)
– p5(WSU, Washington)
• If we find a way to “match” the variables
• Then we can infer q(WSU, Washington)
GMP Example
• Rule: Missile(x) & Owns(Nono, x) ->
Sells(West, Nono,x)
• KB contains
– Missile(M1)
– Owns(Nono,M1)
• To apply, GMP, make sure instantiations of x
are the same
• Variable matching process is called unification
Keeping Track Of Change
• Facts hold in situations, rather than forever
– Example, Holding(Gold,Now) rather than Holding(Gold)
• Situation calculus is one way to represent change in FOPC
– Adds a situation argument to each time-dependent predicate
– Example, Now in Holding(Gold,Now) denotes a situation
• Situations are connected by the Result function
– Result(a,s) is the situation that results from applying action a in s
Describing Actions
• Effect axiom: describe changes due to action
–  s AtGold(s) -> Holding(Gold, Result(Grab, s))
• Frame axiom--describe non-changes due to action
–  s HaveArrow(s) -> HaveArrow(Result(Grab, s))
• Frame problem: find an elegant way to handle non-change

(a) Representation--avoid frame axioms
(b) Inference--avoid repeated ``copy-overs'' to keep track of state
• Qualification problem : true descriptions of real actions require
endless caveats - what if gold is slippery or nailed down or …
• Ramification problem : real actions have many secondary
consequences - what about the dust on the gold, wear and tear
on gloves, …
Describing Actions
• Successor-state axioms solve the representational
frame problem
• Each axiom is about a predicate (not an action per
se)
– P true afterwords <->
• [an action made P true
• v P true already and no action made P false]
• For holding
 the gold
– a,s Holding(Gold, Result(a,s)) <->
((a = Grab & AtGold(s)) v (Holding(gold,s) & a !=
Release))
Generating Plans
• Initial condition in KB
– At(Agent, [1,1], S0)
– At(Gold, [1,2], S0)
• Query
– Ask(KB,  s Holding(Gold,s))
– In what situation will I be holding the gold?
• Answer: {s/Result(Grab, Result(Forward, S0))}
– Go forward and then grab the gold
– This assumes that the agent is interested in plans
starting at S0 and that S0 is the only situation
described in the KB
Generating Plans: A Better Way
• Represent plans as action sequences [a1, a2, .., an}
• PlanResult(p,s) is the result of execute p (an action
sequence) in s
• Then query Ask(KB, p Holding(Gold,PlanResult(p, S0)) has

solution {p/[Forward, Grab]}
• Definition of PlanResult in terms of Result:
• s PlanResult([], s) = s

• a,p,s PlanResult([a|p], s) = PlanResult(p, Result(a,s))

– Planning systems are special-purpose reasoners designed
to do this type of inference more efficiently than a general-
purpose reasoner

L4 Knowledge Representation

Uploaded by

Copyright:

Available Formats

L4 Knowledge Representation

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

L4 Knowledge Representation

Uploaded by

Copyright:

Available Formats

CSE 3201

Artificial Intelligence and Machine Learning

– Capture the knowledge needed to formalize the

– Apply a search technique to solve problem

– Execute the problem solution

• A good description, developed within the conventions

• How to represent facts / beliefs

• Syntax: Form used to represent x + 2 >= y is a sentence

• Situation after detecting nothing in [1,1], moving right,

KB = wumpus world rules + observations

KB = wumpus world rules + observations

KB = wumpus world rules + observations

alpha2 = “[2,2] is safe”, KB |= alpha2

• Abduction is not sound B, A  B

• To determine the truth of a complex statement for I, we can

• Modus ponens A, A  B • Double-negation A

– Application of inference rules

xLovesWate rmelon( x)  x LovesWater melon( x)

– Do these mean the same thing?

• Frame problem: find an elegant way to handle non-change

You might also like

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.