Artificial Intelligence

Fundamentals
Problem Solving
Goal Trees
Rule-Based Expert Systems
 The Problem-Reduction method
• Convert difficult goals into one or more easier-
to-achieve subgoals
• Each subgoal may be divided still more finely
into one or more lower-level subgoals
• Recognize goals -> convert them into
appropriate subgoals
• Problem reduction -> goal reduction
 Example 1
• Resolve the following indefinite integral

• Can you do it in your head?

• A program that can do that, it’s an intelligent program?
• In 1963, Slagle James, "A Heuristic Program that Solves
Symbolic Integration Problems in Freshman Calculus.“,
presents SAINT (Symbolic Automation INTegrator)
 Standards forms
• If an integral is into a “standard form” the goal is
immediately achieved by substitution
• SAINT uses 26 standard forms
 Simple (Safe) transformation
1. Deal with minus sign

2. Take the constant out

3. Decompose the sum

4. Polynomial division
 Simple transformation
• a safe transformation – a transformation
which, when applicable, is always or almost
always appropriate
• for a goal, a transformation is appropriate if
it’s the correct next step to bring that goal
nearer to achievement
• SAINT uses 8 simple transformations
 The framework
SUCCESS
Apply all simple Done ?
Look in table
transformation

GOAL

Safe transformation

SUB-GOAL

• This version of framework doesn’t tell us anything

about what happens for transformation no. 3
 The framework - update
SUCCESS
Apply all simple Done ?
Look in table
transformation

GOAL

Safe transformation

SUB-GOAL
AND NODE
 Problem reduction

• Apply all simple transformations

1 2
 Heuristic transformations
• Heuristic method - method that often work;
isn’t guaranty that will always works
• Is not an algorithm – it’s an attempt
 Heuristic transformation
A. Transformation from trigonometric form into other trigonometric
form

B. Transformation from trigonometric form into polynomial form

C. Transformation from polynomial form into trigonometric form

Problem reduction - continue

C
A

GOAL

Safe transformation

SUB-GOAL GOAL TREE

AND NODE

AND/OR TREE
OR NODE

Which one is easier ?

• Measure the depth of function composition
• The winner is tg(x)

B 4

b 1

b
 Problem reduction - continue
C
b

• Before doing this transformation, SAINT compute the

function composition (which is 3) and return to a
previous choice which have 2

• After doing some transformation, the function

composition raises to 4, and because of that, SAINT
returns to this transformation
 The framework - update
SUCCESS
Apply all simple Done ?
Look in table
transformation

Apply heuristic
Find problem
transformation
 Questions
• How good the integration program is it?
– 32K memory
– Resolves 54 from 56 of the hardest problem (lacks 2
transformations)
– Depth of the tree – 7 levels
– Average depth – approx. 3
– Branches unused – approx. 1
• What kind of knowledge is involved in this?
– Knowledge about transformations
– How goal tree works
– Knowledge about standard forms
– Knowledge about domain
 Questions
• How is the knowledge represented?
– Table expressions
– LISP relations
– Goal tree information is embedded into
procedures
• How much knowledge is required?
– Table of integrals – standard form – 26
– Simple transformations – 8
– Heuristics transformations - 12
 Blocks world
• Demo
• Written by Terry Winograd – professor at Stanford

PUT-ON

GET SPACE MAKE SPACE

GRASP CLEAR TOP

GET RID OF

MOVE

UNGRASP
 Example
B1 B3
B2 B4 PUT-ON B2 B4

GET SPACE GRASP B2 MOVE B2 UNGRASP B1

MAKE SPACE CLEAR TOP B2

WHY HOW
GET RID OF B3 GET RID OF B1

PUT ON B1 Table
GOAL TREE
 How and Why questions
• Deal with “how” questions:
– Identify the goal involved in the goal tree
– If the goal is an AND node report all immediate
subgoals
– If the goal is an OR node report the immediate
subgoal that was achieved
• Deal with “why” questions:
– Identify the goal involved in the goal tree
– Report the immediate supergoal
• The complexity of the behavior is largely a
consequence not of the complexity of the program, but
the complexity of a problem (the environment)

Ant road

C(behavior) = max(C(program), C(environment))

 Rule-based expert systems
• Emulate the decision-making ability of a
human expert
• All the knowledge in a form of simple rules: “If
…..then…”
 Example – Animals from Zoo
R1
HAS HAIR MAMMAL

HAS CLAWS
R2 CARNIVORE R4
SHARP TEETH
CHEETAH
FORWARD POINTING EYE
R3
EATS MEAT

SPOTS

SPOTS

FORWARD CHAINING RULE-BASED EXPERT SYSTEM AND NODE

 Rule based expert systems
• Is not an expert in the real meaning of the
word because it didn’t have the experience of
previous cases
• Only follows some already defined rules
• The systems can answers to the questions
about the behavior
– Why are you interested in the animal claws?
– Because I will try to see if it’s a carnivore.
 Backward chaining
HAS HAIR MAMMAL

HAS CLAWS
CARNIVORE CHEETAH
SHARP TEETH

FORWARD POINTING EYE

SPOTS

SPOTS

BACKWARD CHAINING RULE-BASED EXPERT SYSTEM

 Expert systems
• Are deduction systems – working with facts to
produce new facts
• One of the first expert system – MYCIN
– Rule example: If x’s type is primary bacteremia
and x’s suspected portal is gastrointestinal and the
site of the culture of x is sterile Then there is
evidence that x is bacteroides
 Principles of knowledge engineering
• Deal with specific cases – learn knowledge
they otherwise they missed
• Ask questions about the things that appear to
be the same but actually they are handled
differently -> new words in my domain
• Build a system and see when it crack ->
missing rules
 Questions
• Is an expert system real smart ?
• He doesn’t know about the knowledge
involves into an expert system
• Rule based systems doesn’t have anything to
do with common sense
 Sample exam problem
• Due to constant pressure from the AIF staff, J.
K. Rowling decides to write an 8th Harry
Potter book. But, she's suffering from a bad
case of writer's block and decides to use a
rule-based system to help her with the plot for
Harry Potter and the Deadhorse Principle.
She's given you a set of rules and assertions
and would like your help with developing key
plot points.
• ?x, ?y – variables waiting to be bound
• after IF we have some antecedents that must
be true in order to match the rule
• after THEN we have consequences that will be
added into DB
• Check assertions before using a rule
 Backward chaining
• When working on a hypothesis, the backward chainer tries to find a
matching assertion in the list of assertions.
– If we must demonstrate that “Seamus snogs Millicent” then we done.
Because it’s in assertion list then we prove it.
• If no matching assertion is found, the backward chainer tries to find
a rule with a matching consequent.
– If we must demonstrate that “Seamus is a protagonist” then based on
P1 we can demonstrate that someone is a protagonist if is living in GT,
and we have that Seamus lives in GT, so again we done.
• In case none are found, then the backward chainer assumes the
hypothesis is false.
• The backward chainer never alters the list of assertions, so it can
derive the same result multiple times.
• Rules are tried in the order they appear.
• Antecedents are tried in the order they appear
• The goal tree is traverse in depth-first order
 Millicent becomes Hermione's friend
P4

Millicent studies a lot Millicent is a protagonist

 Millicent becomes Hermione's friend
P4

Millicent studies a lot Millicent is a protagonist

P3
Millicent is
ambitious

Millicent is a Millicent is a
protagonist villain
 Millicent becomes Hermione's friend
P4

Millicent studies a lot Millicent is a protagonist

P3
Millicent is
ambitious

Millicent is a Millicent is a
protagonist villain

P1

Millicent lives
in Gryffindor
Tower
 Millicent becomes Hermione's friend
P4

Millicent studies a lot Millicent is a protagonist

P3
Millicent is
ambitious

Millicent is a Millicent is a
protagonist villain

P1 P2

Millicent lives Millicent lives

in Gryffindor in Slytherin
Tower dungeon
A0
 Millicent becomes Hermione's friend
P4

Millicent studies a lot Millicent is a protagonist

P3
Millicent is
ambitious
A1
Millicent is a Millicent is a
protagonist villain

P1 P2

Millicent lives Millicent lives

in Gryffindor in Slytherin
Tower dungeon
A0
 Millicent becomes Hermione's friend
P4

Millicent studies a lot Millicent is a protagonist

P3
P1
Millicent is
ambitious Millicent lives
in Gryffindor
A1
Millicent is a Millicent is a Tower
protagonist villain

P1 P2

Millicent lives Millicent lives

in Gryffindor in Slytherin
Tower dungeon
A0
 Millicent becomes Hermione's friend
P4

Millicent studies a lot Millicent is a protagonist

P3
P1
Millicent is
ambitious Millicent lives
in Gryffindor
A1
Millicent is a Millicent is a Tower
protagonist villain

P1 P2

Millicent lives Millicent lives

in Gryffindor in Slytherin
Tower dungeon
A0
 A2 part
• Now, determine the minimum number of additional
assertions required for Millicent to become Hermione's
friend and list those assertions. Include no assertion
that matches the consequent of a rule.

Millicent lives in Gryffindor Tower

• Your solution to Part A2 creates an uncommon
situation. What is that uncommon situation and if J. K.
considers the situation to be a problem, what should
she do to the list of assertions to solve the problem?

remove the A0
 Forward chaining
• Assume rule-ordering conflict resolution
• New assertions are added to the bottom of the
list of assertions.
• If a particular rule matches assertions in the list
of assertions in more than one way, the matches
are considered in the order corresponding to the
top-to-bottom order of the matched assertions.
Thus, if a particular rule has an antecedent that
matches both A1 and A2, the match with A1 is
considered first.
• Run forward chaining on the rules and
assertions. For the first two iterations, fill out
the first two rows in the table below, noting
the rules whose antecedents match the data,
the rule that fires, and the new assertions that
are added by the rule. For the remainder,
supply only the fired rules and new assertions.
 Matched Fired New Assertions Added to List of Assertions

P1,P2,P5 P1 Seamus is a protagonist

P1,P2,P5 P2 Millicent is a villain
P1,P2,P3,P5 P3 Millicent studies a lot
P1,P2,P3,P5 P5 Seamus has a bad term
 Related resources
• http://aitopics.org/sites/default/files/classic/Feigenbaum_Feldman/Computers_And_T
hought-Part_1_SAINT.pdf
• http://ocw.mit.edu/courses/electrical-engineering-and-computer-science/6-034-
artificial-intelligence-fall-2010/exams/MIT6_034F10_quiz1_2009.pdf

Readings
• Artificial Intelligence (3rd Edition), Patrick Winston, pp. 53-60