MODULE 4

informed () search , Heuristic

function,
logical agents.
Using problem specific
knowledge to aid searching
 Without incorporating
Search
knowledge into searching,
one can have no bias (i.e. a everywhere
preference) on the search !!
space.

 Without a bias, one is forced

to look everywhere to find the
answer. Hence, the
complexity of uninformed
search is intractable.

2
 Using problem specific
knowledge to aid searching
 With knowledge, one can search the state space as if he was
given “hints” when exploring a maze.
 Heuristic information in search = Hints
 Leads to dramatic speed up in efficiency.

B C D E
Search
F G H I J
only in
this K L M N
subtree!!
O 3
 More formally, why heuristic
functions work?
 In any search problem where there are at most b choices
at each node and a depth of d at the goal node, a naive
search algorithm would have to, in the worst case, search
around O(bd) nodes before finding a solution
(Exponential Time Complexity).

 Heuristics improve the efficiency of search algorithms

by reducing the effective branching factor from b to
(ideally) a low constant b* such that
 1 =< b* < b

4
Informed Search
Strategies

Best First Search

 An implementation of Best
First Search
function BEST-FIRST-SEARCH (problem, eval-
fn)
returns a solution sequence, or failure

queuing-fn = a function that sorts nodes by eval-fn

return GENERIC-SEARCH (problem,queuing-

fn)

6
Informed Search
Strategies

Greedy Search
eval-fn: f(n) = h(n)
8
9
 Greedy Search
Start State Heuristic:
A 75 h(n)
118
A 366
140 B
C B 374
111
E C 329
D 80 99 D 244
E 253
G F
F 178
97
G 193
H 211 H 98
101 I 0
I
Goal f(n) = h (n) = straight-line distance
heuristic
10
 Greedy Search
Start State Heuristic:
A 75 h(n)
118
A 366
140 B
C B 374
111
E C 329
D 80 99 D 244
E 253
G F
F 178
97
G 193
H 211 H 98
101 I 0
I
Goal f(n) = h (n) = straight-line distance
heuristic
11
 Greedy Search
Start State Heuristic:
A 75 h(n)
118
A 366
140 B
C B 374
111
E C 329
D 80 99 D 244
E 253
G F
F 178
97
G 193
H 211 H 98
101 I 0
I
Goal f(n) = h (n) = straight-line distance
heuristic
12
 Greedy Search
Start State Heuristic:
A 75 h(n)
118
A 366
140 B
C B 374
111
E C 329
D 80 99 D 244
E 253
G F
F 178
97
G 193
H 211 H 98
101 I 0
I
Goal f(n) = h (n) = straight-line distance
heuristic
13
 Greedy Search
Start State Heuristic:
A 75 h(n)
118
A 366
140 B
C B 374
111
E C 329
D 80 99 D 244
E 253
G F
F 178
97
G 193
H 211 H 98
101 I 0
I
Goal f(n) = h (n) = straight-line distance
heuristic
14
 Greedy Search
Start State Heuristic:
A 75 h(n)
118
A 366
140 B
C B 374
111
E C 329
D 80 99 D 244
E 253
G F
F 178
97
G 193
H 211 H 98
101 I 0
I
Goal f(n) = h (n) = straight-line distance
heuristic
15
 Greedy Search
Start State Heuristic:
A 75 h(n)
118
A 366
140 B
C B 374
111
E C 329
D 80 99 D 244
E 253
G F
F 178
97
G 193
H 211 H 98
101 I 0
I
Goal f(n) = h (n) = straight-line distance
heuristic
16
 Greedy Search
Start State Heuristic:
A 75 h(n)
118
A 366
140 B
C B 374
111
E C 329
D 80 99 D 244
E 253
G F
F 178
97
G 193
H 211 H 98
101 I 0
I
Goal f(n) = h (n) = straight-line distance
heuristic
17
 Greedy Search
Start State Heuristic:
A 75 h(n)
118
A 366
140 B
C B 374
111
E C 329
D 80 99 D 244
E 253
G F
F 178
97
G 193
H 211 H 98
101 I 0
I
Goal f(n) = h (n) = straight-line distance
heuristic
18
 Greedy Search
Start State Heuristic:
A 75 h(n)
118
A 366
140 B
C B 374
111
E C 329
D 80 99 D 244
E 253
G F
F 178
97
G 193
H 211 H 98
101 I 0
I
Goal f(n) = h (n) = straight-line distance
heuristic
19
Greedy Search: Tree
Search
Start
A

20
Greedy Search: Tree
Search
Start
A 75
118

[329] C 140 [374] B

[253] E

21
Greedy Search: Tree
Search
Start
A 75
118

[329] C 140 [374] B

[253] E
80 99

[193] [178]
G F
[366] A

22
Greedy Search: Tree
Search
Start
A 75
118

[329] C 140 [374] B

[253] E
80 99

[193] [178]
G F
[366] A
211

[253] E I [0]

Goal

23
 Greedy Search: Tree
Search
Start
A 75
118

[329] C 140 [374] B

[253] E
80 99

[193] [178]
G F
[366] A
211

[253] E I [0]

Goal
Path cost(A-E-F-I) = 253 + 178 + 0 = 431
24
dist(A-E-F-I) = 140 + 99 + 211 = 450
 Greedy Search: Optimal ?
Start State Heuristic:
A 75 h(n)
118
A 366
140 B
C B 374
111
E C 329
D 80 99 D 244
E 253
G F
F 178
97
G 193
H 211 H 98
101 I 0
I
Goal f(n) = h (n) = straight-line distance
heuristic
25
 Greedy Search:
Complete ?
Start State Heuristic:
A 75 h(n)
118
A 366
140 B
C B 374
111
E ** C 250
D 80 99 D 244
E 253
G F
F 178
97
G 193
H 211 H 98
101 I 0
I
Goal f(n) = h (n) = straight-line distance
heuristic
26
Greedy Search: Tree
Search
Start
A

27
Greedy Search: Tree
Search
Start
A 75
118

[250] C 140 [374] B

[253] E

28
 Greedy Search: Tree
Search
Start
A 75
118

[250] C 140 [374] B

[253] E
111
[244] D

29
 Greedy Search: Tree
Search
Start
A 75
118

[250] C 140 [374] B

[253] E
111
[244] D

Infinite Branch !
[250] C

30
 Greedy Search: Tree
Search
Start
A 75
118

[250] C 140 [374] B

[253] E
111
[244] D

Infinite Branch !
[250] C

[244] D

31
 Greedy Search: Tree
Search
Start
A 75
118

[250] C 140 [374] B

[253] E
111
[244] D

Infinite Branch !
[250] C

[244] D

32
 Greedy Search: Time and
Space Complexity ?
Start
A
118 75 • Greedy search is not
C 140 B optimal.
111
E
• Greedy search is
D 80 99
incomplete without
G F systematic checking of
97 repeated states.
H 211
• In the worst case, the
101
I Time and Space Complexity
Goal of Greedy Search are both
33
O(b )
m
Informed Search
Strategies

A* Search

eval-fn: f(n)=g(n)+h(n)
A* (A Star)
 Greedy Search minimizes a heuristic h(n) which is an
estimated cost from a node n to the goal state. Greedy
Search is efficient but it is not optimal nor complete.

 Uniform Cost Search minimizes the cost g(n) from the

initial state to n. UCS is optimal and complete but not
efficient.

 New Strategy: Combine Greedy Search and UCS to get an

efficient algorithm which is complete and optimal.

35
A* (A Star)
 A* uses a heuristic function which
combines g(n) and h(n): f(n) = g(n) + h(n)

 g(n) is the exact cost to reach node n from

the initial state.

 h(n) is an estimation of the remaining cost

to reach the goal.
36
A* (A Star)

g(n)

f(n) = g(n) n
+h(n)

h(n)

37
 A* Search
Start State Heuristic:
A 75 h(n)
118
A 366
140 B
C B 374
111
E C 329
D 80 99 D 244
E 253
G F
F 178
97
G 193
H 211 H 98
101 I 0
I
Goal f(n) = g(n) + h (n)
g(n): is the exact cost to reach node n from the initial 38
A* Search: Tree Search
A Start

39
A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
]

40
A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
] 80 99

[413 G F [417
] ]

41
A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
] 80 99

[413 G F [417
] ]
97
[415 H
]

42
A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
] 80 99

[413 G F [417
] ]
97
[415 H
]
101
Goal I [418]

43
A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
] 80 99

[413 G F [417
] ]
97
[415 H I [450
] ]
101
Goal I [418]

44
A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
] 80 99

[413 G F [417
] ]
97
[415 H I [450
] ]
101
Goal I [418]

45
A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
] 80 99

[413 G F [417
] ]
97
[415 H I [450
] ]
101
Goal I [418]

46
A* with f() not Admissible

h() overestimates the cost to

reach the goal state
 A* Search: h not
admissible !
Start State Heuristic:
A 75 h(n)
118
A 366
140 B
C B 374
111
E C 329
D 80 99 D 244
E 253
G F
F 178
97
G 193
H 211 H 138
101 I 0
I
Goal f(n) = g(n) + h (n) – (H-I) Overestimated
g(n): is the exact cost to reach node n from the initial 48
A* Search: Tree Search
A Start

49
A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
]

50
A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
] 80 99

[413 G F [417
] ]

51
A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
] 80 99

[413 G F [417
] ]
97
[455 H
]

52
A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
] 80 99

[413 G F [417
] ]
97
[455 H Goal I [450
] ]

53
 A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
] 80 99

[473 [413 G F [417

D
] ] ]
97
[455 H Goal I [450
] ]

54
 A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
] 80 99

[473 [413 G F [417

D
] ] ]
97
[455 H Goal I [450
] ]

55
 A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
] 80 99

[473 [413 G F [417

D
] ] ]
97
[455 H Goal I [450
] ]

56
 A* Search: Tree Search
A Start

118 75
140

[447 C E [393 B [449

] ]
] 80 99

[473 [413 G F [417

D
] ] ]
97
[455 H Goal I [450
] ]

A* not optimal !!!

57
A* Algorithm

A* with systematic checking

for repeated states …
A* Algorithm
1. Search queue Q is empty.
2. Place the start state s in Q with f value h(s).
3. If Q is empty, return failure.
4. Take node n from Q with lowest f value.
(Keep Q sorted by f values and pick the first element).
5. If n is a goal node, stop and return solution.
6. Generate successors of node n.
7. For each successor n’ of n do:
a) Compute f(n’) = g(n) + cost(n,n’) + h(n’).
b) If n’ is new (never generated before), add n’ to Q.
c) If node n’ is already in Q with a higher f value,
replace it with current f(n’) and place it in sorted
order in Q.
End for
8. Go back to step 3. 59
 A* Search: Analysis
Start •A* is complete except if there
118
A 75 is an infinity of nodes with f <
B f(G).
C 140
111 •A* is optimal if heuristic h is
E
D 80 99 admissible.
•Time complexity depends on
G F
the quality of heuristic but is
97
still exponential.
H 211
•For space complexity, A*
101
keeps all nodes in memory. A*
I
Goal has worst case O(bd) space
complexity, but an iterative 60
A* Search
 Complete

Yes, unless there are infinitely many nodes with f <= f(G)
 Time

Exponential

The better the heuristic, the better the time

Best case h is perfect, O(d)

Worst case h = 0, O(bd) same as BFS
 Space

Keeps all nodes in memory and save in case of repetition

This is O(bd) or worse

A* usually runs out of space before it runs out of time
 Optimal

Yes, cannot expand fi+1 unless fi is finished

AI: Chapter 4: Informed

Search and Exploration 61
Iterative Deepening
A*:IDA*


Use f(N) = g(N) + h(N) with
admissible and consistent h

 Each iteration is depth-first with

cutoff on the value of f of
expanded nodes
62
Consistent Heuristic
 The admissible heuristic h is consistent
(or satisfies the monotone restriction) if
for every node N and every successor N’
of N: N
c(N,N’)

N’ h(N)
h(N)  c(N,N’) + h(N’)
h(N’)

(triangular inequality)
 A consistent heuristic is admissible.
63
Recursive breadth first
Search (RBSF)
 F(n)=g(n)+h(n)
 Requires linear space
 Uses f-limit variable to keep track of f-value to
keep track of best alternative path.

AI: Chapter 4: Informed

Search and Exploration 64
Recursive breadth first
Search (RBSF)
 F(n)=g(n)+h(n)
 Requires linear space
 Uses f-limit variable to keep track of f-value to
keep track of best alternative path.

AI: Chapter 4: Informed

Search and Exploration 65
Recursive breadth first
Search (RBSF)
 Complete

Yes, if b is finite
 Time

Difficult to charaterise
 Space

linear
 Optimal

Yes, if h(n) is admissible

AI: Chapter 4: Informed

Search and Exploration 66
 Heuristic Functions
 A heuristic function is a function f(n) that gives an estimation on
the “cost” of getting from node n to the goal state – so that the
node with the least cost among all possible choices can be
selected for expansion first.

 Three approaches to defining f:


f measures the value of the current state (its “goodness”)


f measures the estimated cost of getting to the goal from the current
state:

f(n) = h(n) where h(n) = an estimate of the cost to get
from n to a goal


f measures the estimated cost of getting to the goal state from the
current state and the cost of the existing path to it. Often, in this
case, we decompose f:

f(n) = g(n) + h(n) where g(n) = the cost to get to n (from
initial state) 67
When to Use Search
Techniques
 The search space is small, and
 There are no other available techniques,
or
 It is not worth the effort to develop a
more efficient technique

 The search space is large, and

 There is no other available techniques,
and
 There exist “good” heuristics
68
Heuristic Functions
 To use A* a heuristic function must be
used that never overestimates the
number of steps to the goal

 h1=the number of misplaced tiles

 h2=the sum of the Manhattan distances

of the tiles from their goal positions
AI: Chapter 4: Informed
Search and Exploration 69
Heuristic Functions
 h1 = 7
 h2 = 4+0+3+3+1+0+2+1 = 14

AI: Chapter 4: Informed

Search and Exploration 70
Dominance
 If h2(n) > h1(n) for all n (both
admissible) then h2(n) dominates
h1(n) and is better for the search

 Take a look at figure 4.8!

AI: Chapter 4: Informed

Search and Exploration 71
Dominance
 If h2(n) > h1(n) for all n (both
admissible) then h2(n) dominates
h1(n) and is better for the search

 Take a look at figure 4.8!

AI: Chapter 4: Informed

Search and Exploration 72
Relaxed Problems
 A Relaxed Problem is a problem with
fewer restrictions on the actions
 The cost of an optimal solution to a
relaxed problem is an admissible heuristic
for the original problem

 Key point: The optimal solution of a

relaxed problem is no greater than the
optimal solution of the real problem
AI: Chapter 4: Informed
Search and Exploration 73
Relaxed Problems
 Example: 8-puzzle
 Consider only getting tiles 1, 2, 3, and
4 into place

 If the rules are relaxed such that a tile

can move anywhere then h1(n) gives
the shortest solution
 If the rules are relaxed such that a tile
can move to any adjacent square then
h2(n) gives the shortest solution
AI: Chapter 4: Informed
Search and Exploration 74
 Conclusions
 Frustration with uninformed search led to the idea
of using domain specific knowledge in a search so
that one can intelligently explore only the relevant
part of the search space that has a good chance of
containing the goal state. These new techniques
are called informed (heuristic) search strategies.

 Even though heuristics improve the performance

of informed search algorithms, they are still time
consuming especially for large size instances.
75
Logical Agents
 Humans can know “things” and “reason”

Representation: How are the things stored?

Reasoning: How is the knowledge used?

To solve a problem…

To generate more knowledge…

 Knowledge and reasoning are important to

artificial agents because they enable successful
behaviors difficult to achieve otherwise

Useful in partially observable environments

 Can benefit from knowledge in very general

forms, combining and recombining information
AI: Chapter 7: Logical Agents 76
Knowledge-Based Agents
 Central component of a Knowledge-
Based Agent is a Knowledge-Base
 A set of sentences in a formal language

Sentences are expressed using a knowledge
representation language

 Two generic functions:

 TELL - add new sentences (facts) to the KB

“Tell it what it needs to know”
 ASK - query what is known from the KB

“Ask what to do next”

AI: Chapter 7: Logical Agents 77

 Knowledge bases

 Knowledge base = set of sentences in a formal language

 Declarative approach to building an agent (or other system):


Tell it what it needs to know

 Then it can Ask itself what to do - answers should follow from

the KB

 Agents can be viewed at the knowledge level

i.e., what they know, regardless of how implemented

Knowledge-Based Agents

AI: Chapter 7: Logical Agents 79

Wumpus World

AI: Chapter 7: Logical Agents 80

Wumpus World

AI: Chapter 7: Logical Agents 81

Wumpus World

AI: Chapter 7: Logical Agents 82

Wumpus World
 Performance Measure

Gold +1000, Death – 1000

Step -1, Use arrow -10

 Environment

Square adjacent to the Wumpus are
smelly

Squares adjacent to the pit are breezy

Glitter iff gold is in the same square

Shooting kills Wumpus if you are
facing it

Shooting uses up the only arrow

Grabbing picks up the gold if in the
same square

Releasing drops the gold in the same
square

 Actuators

Left turn, right turn, forward, grab,
release, shoot

 Sensors

Breeze, glitter, and smell

 See page 197-8 for more details!

83
Wumpus World
 Characterization of Wumpus World

Observable

partial, only local perception

Deterministic

Yes, outcomes are specified

Static

Yes, Wumpus and pits do not move

Discrete

Yes

Single Agent

Yes

AI: Chapter 7: Logical Agents 84

Wumpus World

Wumpus World

85
Wumpus World

Wumpus World

86
Logic
 Knowledge bases  Example:
consist of sentences x + 2 >= y is a sentence
in a formal language
 Syntax x2 + y > is not a sentence

Sentences are well
formed x + 2 >= y is true iff x + 2
 Semantics is no less than y

The “meaning” of the
sentence x + 2 >= y is true in a

The truth of each world where x = 7, y=1
sentence with
respect to each x + 2 >= y is false in
possible world
world where x = 0, y =6
(model)

AI: Chapter 7: Logical Agents 87

Logic
 Entailment means that one thing follows
logically from another
 |= 

  |=  iff in every model in which  is true,

 is also true

 if  is true, then  must be true

 the truth of  is “contained” in the truth of 

AI: Chapter 7: Logical Agents 88
Logic
 Example:
 A KB containing

“Cleveland won”

“Dallas won”

Entails…

“Either Cleveland won or Dallas won”

 Example:
x + y = 4 entails 4 = x + y
AI: Chapter 7: Logical Agents 89
Logic

90
 Propositional logic: Syntax
 Propositional logic is the simplest logic – illustrates basic
ideas
 Atomic sentences = single proposition symbols

E.g., P, Q, R

Special cases: True = always true, False = always false

 Complex sentences:

If S is a sentence, S is a sentence (negation)

If S1 and S2 are sentences, S1  S2 is a sentence (conjunction)

If S1 and S2 are sentences, S1  S2 is a sentence (disjunction)

If S1 and S2 are sentences, S1  S2 is a sentence (implication)

If S1 and S2 are sentences, S1  S2 is a sentence (biconditional)
 Wumpus world sentences
Let Pi,j be true if there is a pit in [i, j].
Let Bi,j be true if there is a breeze in [i, j].
start:  P1,1
 B1,1
B2,1

 "Pits cause breezes in adjacent squares"

B1,1  (P1,2  P2,1)
B2,1  (P1,1  P2,2  P3,1)

 KB can be expressed as the conjunction of all of these

sentences

 Note that these sentences are rather long-winded!

 E.g., breeze “rule” must be stated explicitly for each square
 First-order logic will allow us to define more general patterns.
Truth tables for
connectives
 Truth tables for
connectives
Evaluation Demo -

mplication is always true

hen the premise is false

hy? P=>Q means “if P is true then I am claiming that

otherwise no claim”
nly way for this to be false is if P is true and Q is false
 Wumpus models

 KB = all possible wumpus-worlds

consistent with the observations and the
“physics” of the Wumpus world.
Listing of possible worlds for the Wumpus
KB
α1 = ”square [1,2] is safe”.
KB = detect nothing in [1,1], detect breeze in [2,1]
Wumpus World Sentences
 Let Pi,j be True if  “Pits cause breezes
there is a pit in [i,j] in adjacent squares”
 Let Bi,j be True if
there is a breeze B1,1  (P1,2  P2,1)
in [i,j]
B2,1  (P1,1  P2,1  P3,1)
 ¬P1,1
 A square is breezy if
 ¬ B1,1
and only if there is
 B2,1 an adjacent pit
February 20, 2006 AI: Chapter 7: Logical Agents 97
A Simple Knowledge Base

February 20, 2006 AI: Chapter 7: Logical Agents 98

 A Simple Knowledge Base


R1: ¬P1,1  KB consists of

R2: B1,1  (P1,2  P2,1) sentences R1 thru R5

R3: B2,1 (P1,1  P2,2  P3,1)

R4: ¬ B1,1
 R1  R2  R3  R4  R5

R5: B2,1

February 20, 2006 AI: Chapter 7: Logical Agents 99

 A Simple Knowledge Base

 Every known inference algorithm for propositional logic has a worst-case

complexity that is exponential in the size of the input. (co-NP complete)

February 20, 2006 AI: Chapter 7: Logical Agents 100

Equivalence, Validity,
Satisfiability

February 20, 2006 AI: Chapter 7: Logical Agents 101

Equivalence, Validity,
Satisfiability
 A sentence if valid if it is true in all models
 e.g. True, A  ¬A, A  A, (A  (A  B)  B
 Validity is connected to inference via the Deduction
Theorem
 KB |-  iff (KB  ) is valid
 A sentence is satisfiable if it is True in some model
 e.g. A  B, C
 A sentence is unstatisfiable if it is True in no models
 e.g. A  ¬A
 Satisfiability is connected to inference via the
following
 KB |=  iff (KB  ¬) is unsatisfiable
 proof by contradiction

February 20, 2006 AI: Chapter 7: Logical Agents 102

Reasoning Patterns
 Inference Rules

Patterns of inference that can be applied to derive chains
of conclusions that lead to the desired goal.

 Modus Ponens

Given: S1  S2 and S1, derive S2

 And-Elimination

Given: S1  S2, derive S1

Given: S1  S2, derive S2

 DeMorgan’s Law

Given: ( A  B) derive A  B

Given: ( A  B) derive A  B

February 20, 2006 AI: Chapter 7: Logical Agents 103

Reasoning Patterns
 And Elimination  Modus Ponens
     

 

 From a
 Whenever sentences
of the form    and
conjunction, any of  are given, then
the conjuncts can sentence  can be
be inferred inferred
 (WumpusAhead  WumpusAlive)
 Shoot and (WumpusAhead 
 (WumpusAhead  WumpusAlive), Shoot can be
WumpusAlive), WumpusAlive inferred
can be inferred

February 20, 2006 AI: Chapter 7: Logical Agents 104

Example Proof By
Deduction
 Knowledge
S1: B22  ( P21  P23  P12  P32 ) rule
S2: B22 observation
 Inferences
S3: (B22  (P21  P23  P12  P32 ))
((P21  P23  P12  P32 )  B22) [S1,bi elim]
S4: ((P21  P23  P12  P32 )  B22) [S3, and elim]
S5: (B22  ( P21  P23  P12  P32 )) [contrapos]
S6: (P21  P23  P12  P32 ) [S2,S6, MP]
S7: P21  P23  P12  P32 [S6, DeMorg]

February 20, 2006 AI: Chapter 7: Logical Agents 105

Evaluation of Deductive
Inference
 Sound
 Yes, because the inference rules themselves are
sound. (This can be proven using a truth table
argument).
 Complete
 If we allow all possible inference rules, we’re
searching in an infinite space, hence not complete
 If we limit inference rules, we run the risk of
leaving out the necessary one…
 Monotonic
 If we have a proof, adding information to the DB
will not invalidate the proof

February 20, 2006 AI: Chapter 7: Logical Agents 106

Resolution
 Resolution allows a complete inference
mechanism (search-based) using only one
rule of inference
 Resolution rule:
 Given: P1  P2  P3 … Pn, and P1  Q1 … Qm
 Conclude: P2  P3 … Pn  Q1 … Qm
Complementary literals P1 and P1 “cancel out”
 Why it works:
 Consider 2 cases: P1 is true, and P1 is false

February 20, 2006 AI: Chapter 7: Logical Agents 107

Resolution in Wumpus
World
 There is a pit at 2,1 or 2,3 or 1,2 or
3,2
 P21  P23  P12  P32
 There is no pit at 2,1
 P21
 Therefore (by resolution) the pit must
be at 2,3 or 1,2 or 3,2
 P23  P12  P32
February 20, 2006 AI: Chapter 7: Logical Agents 108
Proof using Resolution
 To prove a fact P, repeatedly apply resolution until
either:

No new clauses can be added, (KB does not entail P)

The empty clause is derived (KB does entail P)

 This is proof by contradiction: if we prove that KB  P

derives a contradiction (empty clause) and we know KB
is true, then P must be false, so P must be true!

 To apply resolution mechanically, facts need to be in

Conjunctive Normal Form (CNF)

 To carry out the proof, need a search mechanism that

will enumerate all possible resolutions.

February 20, 2006 AI: Chapter 7: Logical Agents 109

CNF Example
1. B22  ( P21  P23  P12  P32 )

2. Eliminate  , replacing with two implications

(B22  ( P21  P23  P12  P32 ))  ((P21  P23  P12  P32 )  B22)

3. Replace implication (A  B) by A  B
(B22  ( P21  P23  P12  P32 ))  ((P21  P23  P12  P32 )  B22)

4. Move  “inwards” (unnecessary parens removed)

(B22  P21  P23  P12  P32 )  ( (P21  P23  P12  P32 )  B22)

4. Distributive Law
(B22  P21  P23  P12  P32 )  (P21  B22)  (P23  B22)  (P12  B22)  (P32 
B22)

(Final result has 5 clauses)

February 20, 2006 AI: Chapter 7: Logical Agents 110
Resolution Example
 Given B22 and P21 and P23 and P32 ,
prove P12
 (B22  P21  P23  P12  P32 ) ; P12
 (B22  P21  P23  P32 ) ; P21
 (B22  P23  P32 ) ; P23
 (B22  P32 ) ; P32
 (B22) ; B22
 [empty clause]
February 20, 2006 AI: Chapter 7: Logical Agents 111
Evaluation of Resolution
 Resolution is sound
 Because the resolution rule is true in all cases
 Resolution is complete
 Provided a complete search method is used
to find the proof, if a proof can be found it will
 Note: you must know what you’re trying to
prove in order to prove it!
 Resolution is exponential
 The number of clauses that we must search
grows exponentially…

February 20, 2006 AI: Chapter 7: Logical Agents 112

Horn Clauses
 A Horn Clause is a CNF clause with exactly one
positive literal
 The positive literal is called the head
 The negative literals are called the body
 Prolog: head:- body1, body2, body3 …
 English: “To prove the head, prove body1, …”
 Implication: If (body1, body2 …) then head
 Horn Clauses form the basis of forward and
backward chaining
 The Prolog language is based on Horn Clauses
 Deciding entailment with Horn Clauses is linear
in the size of the knowledge base
February 20, 2006 AI: Chapter 7: Logical Agents 113
Reasoning with Horn
Clauses
 Forward Chaining
 For each new piece of data, generate all new
facts, until the desired fact is generated
 Data-directed reasoning
 Backward Chaining
 To prove the goal, find a clause that contains
the goal as its head, and prove the body
recursively
 (Backtrack when you chose the wrong
clause)
 Goal-directed reasoning
February 20, 2006 AI: Chapter 7: Logical Agents 114
 Forward Chaining
 Fire any rule whose premises are satisfied in the KB
 Add its conclusion to the KB until the query is found

February 20, 2006 AI: Chapter 7: Logical Agents 115

 Forward Chaining
 AND-OR Graph

multiple links joined by an arc indicate conjunction – every link must be proved

multiple links without an arc indicate disjunction – any link can be proved

February 20, 2006 AI: Chapter 7: Logical Agents 116

Forward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 117

Forward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 118

Forward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 119

Forward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 120

Forward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 121

Forward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 122

Forward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 123

Forward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 124

Backward Chaining
 Idea: work backwards from the query q:
 To prove q by BC,

Check if q is known already, or

Prove by BC all premises of some rule concluding q

 Avoid loops
 Check if new subgoal is already on the goal stack

 Avoid repeated work: check if new subgoal

 Has already been proved true, or
 Has already failed

February 20, 2006 AI: Chapter 7: Logical Agents 125

Backward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 126

Backward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 127

Backward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 128

Backward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 129

Backward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 130

Backward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 131

Backward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 132

Backward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 133

Backward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 134

Backward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 135

Backward Chaining

February 20, 2006 AI: Chapter 7: Logical Agents 136

Forward Chaining vs.
Backward Chaining
 Forward Chaining is data driven

Automatic, unconscious processing

E.g. object recognition, routine decisions

May do lots of work that is irrelevant to the goal

 Backward Chaining is goal driven


Appropriate for problem solving

E.g. “Where are my keys?”, “How do I start the
car?”

 The complexity of BC can be much less

than linear in size of the KB
February 20, 2006 AI: Chapter 7: Logical Agents 137