29
1 Artificial Intelligence Artificial Intelligence Mostafa M. Aref University of Bridgeport Department of Computer Science & Engineering

KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

  • Upload
    others

  • View
    4

  • Download
    0

Embed Size (px)

Citation preview

Page 1: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

1

Artificial IntelligenceArtificial Intelligence

Mostafa M. Aref

University of BridgeportDepartment of Computer Science &

Engineering

Page 2: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

2

What is Artificial What is Artificial IntelligenceIntelligence

AI is the study of how to make computers do things which, at the moment, people do better.AI problems– Game Playing Theorem proving– Common sense reasoning (GPS)– Perception– Natural Language Understanding– Expert Systems

The Underlying Assumption– A physical symbol system

entities: symbols, tokens, expression (symbol structure)Collection of processes: create, modification, reproduction and destruction.

– HypothesisThe physical symbol system has the necessary and sufficient means for general intelligent action.

Page 3: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

3

Artificial IntelligenceArtificial IntelligenceTechniquesTechniques

Manipulate SymbolsIntelligence requires knowledge– Voluminous– hard to characterize accurately– constantly changing– organized in a way that it will be used

knowledge should be represented such that– captures generalization– understood by people who provide it– easily to modified– be used even it is not totally accurate or

complete– help to narrow the range of possiblities

Examples: 8 3 4– Tic-Tac-Toe blank, X, O 1 5 9– 0,1,2 - 2,3,5 (18 or 50) - magic square 6 7 2

Search - Knowledge - Abstraction

Page 4: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

4

Level of ModelLevel of Model

same way as people ⇔ whatever easiest wayWhy do we model human performance?– Test psychological theories of human

performance– Enable computers to understand human reasoning– Enable people to understand computer reasoning– Exploit what knowledge we can get from people– Cognitive science

psychologistsLinguistsComputer Scientist

Criteria for Success– Turing Test– Playing Chess– Dendral: analyzes organic compounds to

determined their structures– XCON: configures computer systems

Page 5: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

5

Problems, Problem Spaces Problems, Problem Spaces and Searchand Search

To build a system to solve a particular problem, we need to:– Define the problem precisly

initial situation(s)final situation(s)

– Analyze the problem– Isolate & represent the needed task

knowledge– Choose the best problem-solving

technique(s)Defining the problem as a state spacePlay chess– Rule for each possible position (10120)

No one can supply these rulesNo program can handle all these rules

– Write a rule for legal moves in general– State space of play chess

Page 6: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

6

Problems, Problem Spaces Problems, Problem Spaces and Searchand Search

Advantages– It allows for formal definition of a problem– converts a given situation into desired situation

using a set of permissible operations– defines the problem as:

techniques (a rule defining step)search, very important for hard problems

Example: A water jug problem– Two jugs 4-gallon and 3-gallon jug, how to get

exactly 2 gallons in 4-gallon jug?(0,0) Start State

(4,0) (0,3)(4,3) (1,3) (3,0)

(3,3)(4,2)

(0,2)(2,0)

Goal State

Page 7: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

7

Problem Spaces and Problem Spaces and SearchSearch

Specific rules: no search - no problem solving– solve the problem a head of time

General rules– Search - Applicable to different situation

Special rules: good some times - bad some timesOperationalization: – Informal description -> formal one– Hard and difficult– To get the formal description

define a state space (possible without enumeration all of them)Specify initial state(s) and goal state(s)Specify a set of rules (operations)

– Unstated assumptions– How general should be the rules– how much of work will be precomputed and

represented in the rule– control strategy to search the problem space from

the initial state to the goal state – use algorithmic approach if needed during search– search when no direct method is known

Page 8: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

8

AI Programming AI Programming LanguagesLanguages

Lisp (list processing)– atoms, list, functional programming language– a abc (ab abbc 3 test) (this is a test)– (car (abc sad fgh)) --> abc– (cdr (abc sad fgh)) --> (sad fgh)– (cons abb (abc sad fgh)) --> (abb abc sad fgh)) – (cond ((eq x 2) (+ x 4))

((> x 2) (- x 3))(t (* x 5)) )

Prolog (programming in logic)– rules, backward search

person(marcus).ruler(caesar).tryassassinate(marcus,caesar).hate(X,caesar) :- person(X), notloyalto(X,caesar).notloyalto(X,Y) :- person(X), ruler(Y),

tryassassinate(X,Y).

Page 9: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

9

Production SystemsProduction SystemsProduction Systems– Clips (C Lang. Integrated Production System)

Control Strategies– It causes motion It is Systematic

Breadth-First Search– Not trapped in a blind alley– It is guaranteed to find an existing solution– If multiple solution, minimum solution is found

Depth-First Search– Requires less memory– Less search specially if there are many solutions

ControlControlStructureStructure

WorkingWorkingMemoryMemoryKnowledgeKnowledge

BaseBase

AgendaAgenda

Page 10: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

10

Heuristic SearchHeuristic Search– Mobility systematically– no longer guaranteed the best answer, yet will almost

always finds a very good answer– points to interesting directions– may miss points of interest to particular case– Ex1: nearest neighbor (TSP)– Ex2: f(x,y) ----> f(x,x)

* squareset union identify functionkill suicide

Why we use heuristic?– People are not optimizer but rather satisficers– no very good in the worst case that rarely arise– lead to deep understanding of the problem

heuristics are introduced in search as:– rules by themselves– evaluation function - the more accurate, more direct

solution– Trade off between cost of evaluation of heuristic function

and saving in search time

Page 11: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

11

Problem CharacteristicsProblem CharacteristicsAI The study of techniques for solving exponentially hard problem in polynomial time by exploiting knowledge about the problem domainProblem Characteristics– Decomposable problem

integrationblocks world problem

AC BA B C

– Solution steps can be ignored or undoneProve mathematical theorem (ignorable)8-puzzle (can be undone) (recoverable)

2 8 3 1 2 31 6 4 8 47 5 7 6 5

chess playing (irrecoverable ---> recoverable)

Page 12: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

12

Problem CharacteristicsProblem CharacteristicsDifferent formulation lead to different c/cThe complexity of the control structure depends on the problem c/c

– Predictable problem8-puzzle - certain outcomecard game - uncertain outcomecontrolling a robot arm - legal assistant uncertain outcome + irrecoverable --> hard problem

– Absolute or relative good solutionany-path best-pathTSP

– A state or a path solutionNLU interested in the final interpretation

– The bank president ate a dish of salad with the forkwater jug how to get to the solutionstore the path with each state

– The role of knowledgeplaying chess medical diagnosis

– Interactive solutionCAD/CAM/CAITheorem provingMedical diagnosis

Page 13: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

13

Problem ClassificationProblem Classification

– Classification (e.g. Diagnosis, selection)– Design - propose and refine

Production System Characteristics– Monotonic PS: Rule application never prevents

the other rules application– Partially Commutative PS: Any permutation of

a set of rule transforms state X to state YMonotonic Non- Monotonic

Partially Commutative Changes occur but can

Commutative Ignorable problems be reversed

Theorem proving Order is not criticalNo backtracking Robot, 8-puzzle

blocks world

Not Partially Irreversible changes

Commutative occurChemical synthesis

Page 14: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

14

Design of Search ProgramDesign of Search Program– Forward versus backward reasoning (direction)– Matching– Knowledge representation (frame problem)– Search trees versus search graphs

Additional problem– Missionaries and cannibals

(farmer, grass, fox and goat)– The tower of Hanoi– Monkey and bananas – Cryptarithmetic

SEND DONALD CROSS+ MORE + GERALD + ROADSMONEY ROBERT DANGER

Summary– Define the problem, spesify problem space, the

operator, start state, goal state– problem characteristics– knowledge representation– Problem solving techniques

Page 15: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

15

Knowledge RepresentationKnowledge Representation

Representation and Mapping– Facts: truths that we want to represent

(Knowledge level)– Representation: to be able to manipulate

(Symbol Level)

Facts Internal ResoningRepresentation

Understanding Generation

LanguageRepresentation

Example: Ahmed is a student => student(Ahmed)All students have classes => ∀x: student(x) --> hasclass(x)Ahmed has class <= hasclass(Ahmed)

The mapping is a many-to-many relationEvery student has a class– Mutilated Checkerboard

Page 16: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

16

Properties ofProperties ofKnowledge RepresentationKnowledge RepresentationRepresentation Adequacy– The ability to represent the required knowledge

Inferential Adequacy– The ability to manipulate the knowledge

represented to produce new knowledge corresponding to that inferred from the original

Inferential Efficiency– The ability to direct the inferential mechanisms

into the most productive directions by storing appropriate guides

Acquisition Efficiency– The ability to acquire new knowledge using

automatic methods wherever possible rather than reliance on human intervention

Page 17: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

17

Knowledge RepresentationKnowledge RepresentationApproachesApproaches

Simple Relational KnowledgeStudent Level GPA graduationAhmed Al-bader FR 3.25 2001Said Al-sheikh JR 2.87 1999Fahd Al-Baar SR 3.86 1998– Provide very weak inferential capabilities

Procedural Knowledge– what to do when– low in inferential adequacy and acquisition

efficiency

Inferential Knowledge– Resolution

Inheritable Knowledge– isa class inclusion– instance class membership

Page 18: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

18

Issues in Knowledge Issues in Knowledge RepresentationRepresentation

Important AttributesRelationships among AttributesInversesAn Isa Hierarchy of AttributesTechniques for Reasoning about ValuesSingle-Valued AttributesChoosing the Granularity of RepresentationRepresenting set of ObjectsFinding the Right Structures as Needed– Selecting an Initial Structure

Index the structureconsider major conceptsLocate one major clue

– Revising the Choice When necessaryMake excuse

The Frame Problem

Page 19: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

19

Predicate LogicPredicate Logic

Propositional & Predicate Logic– man(marcus) pompeian(marcus)– ruler(caesar) tryassassinate(marcus, caesar)– pompeian(X) --> roman(X)– roman(X)-->loyalto(X, caesar)Vhate (X, caesar)– man(X)^ruler(Y)^tryassassinate(X,Y) -->

~ loyalto(X,Y)prove hate(marcus,caeser)Resolution & unification – Many sentences are ambiguous– simple representation preclude reasoning– missing some information– instance and isa relations

Question and answeringNatural Deduction– judge(X) ^ ~crooked(X) --> educated(X)– ~ judge(X) V crooked(X) V educated(X)

Page 20: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

20

ResolutionResolution– Resolution is not guaranteed to halt if given a

non-theorem to prove– Semantic information are lost in converting

from predicate to clause form– Difficult to interact with users– Difficulty to represent incomplete infromation:

very, often, assume, more, belief

Using RulesUsing RulesProcedural versus DeclarativeLogic Programming

logic prologvariables quant. explicitly implicitlyAND - OR ^ V , or

p --> q q :- p– Prolog: rules only, fixed control strategy

reason backward, depth first search– Clips: different KR, multiple control strategy,

reason forward

Page 21: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

21

ReasoningReasoningForward versus Backward– more start states or goal states– branching factors– if justification is needed– type of event: arrival of new fact (forward)

query & response (backward)– Examples:

Driving unfamiliar place <----> home– same branching factor

Symbolic integration (forward)Theorem Proving (backward)

– bi-direction searchMatching– Indexing– Matching with variables (Rete-Match)• Rules do not change the state description radically• Structural similarity in rules• Persistence of variable binding consistency

• son(x,y) ^ son(y,z) ==> grandson(x,z)• son(ahmed, ali) and son(omar, saad)

Page 22: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

22

MatchingMatching– Complex and Approximate Matching

Speech understandingELIZA: Psychology Therapist

Person: Men are all alike.ELIZA: In what way?Person: They’re always bugging us about something or other.ELIZA: Can think of a specific example?Person: Well, my boyfriend made me come here.ELIZA: Your boyfriend made you come here.Person: He says I’m depressed much of the time.ELIZA: I am sorry to hear you are depressed.Person: It’s true. I am unhappy.ELIZA: Do you think coming here will help you not be unhappy.Person: I need some help, that much seems certain.ELIZA: What would it mean to you if you gt some help?Person: Perhaps I could learn to get along with my mother.ELIZA: Tell me more about your family– Some ELIZA rules:(X me Y) ==> (X you Y)(I remember X) ==> (Why do remember X just now)(My {family-member} is Y) ==> (Who else in your family is Y)(X {family-member} Y) ==> (Tell me more about your family)

Page 23: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

23

Conflict ResolutionConflict Resolution

Based on Rules– physical order -- give priority– reject general rules

preconditions set is a subset of another rule preconditionshave free variable

Based on objects– priority assigned to obejcts– most recent elements (Short term memory)

Based on states– evaluate the result of each rule

Control Knowledge– search control knowledge

which states are more preferable which rule to apply in a given situationorder to pursue subgoalssequence of rules to apply

– meta-knowledge

Page 24: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

24

Reasoning under Reasoning under uncertaintyuncertainty

Monotonicity of reasoning system– Complete (all facts are present)– Consistent– Only way to change is to add facts

Non-monotonic issues– inferences based on lack of knowledge– ~P it is not known P– defeasible (may be defeated)– updating KB and changing the proofs

Statistical ReasoningStatistical ReasoningProbability and Bayes’ TheoremP(Hi/E) = P(E/Hi) . P(Hi )

∑kn=1 P(E/Hn) . P(Hn)

Fuzzy Logic

tallvery tall

height height

Page 25: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

25

Slot and Filler StructuresSlot and Filler StructuresSemantic Nets– A set of nodes (concepts) connected by a set of

labeled arcs (relations)Mammal

has-partPerson Nose

uniform color instance teamblue Pee-wee-reese Brooklyn

– intersection search– non-binary predicates

John gave the book to Marygive book

agent instance instanceJohn Ev7 Bk23

beneficiary objectMary

– Making some important distinctionssome arcs define new entities (height)other describe relationships among existing entities (greater than, value)

John Billvalue height height

72 H1 H2Greater-than

– Partitioned semantic nets– The Evolution into frames

Page 26: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

26

FramesFrames– A frame is collection of attributes and associated values

that describes some entity– Frames as sets & instances

Kinds of attributes– about the class itself– about the class member

Regular class and meta-classMultiple inheritance

– Slots as full-fledged objectsAttribute properties

– attribute class -- attribute type or value– shared values for all subclasses– default values -- single or multi value– rules for inheriting values– rules for computing values– an inverse attribute

attributes of attribute (frame)

Conceptual Dependency (Schank 1973-1975)provides both the structures and primitivesset of primitives and set of building blocksAdvantages: fewer inference rules

– initial structures have holes to be filledDisadvantages: need more primitives to represent more knowledge

Page 27: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

27

ScriptsScriptsA script– Describes a stereotyped sequence of events.– Consists of a set of frames with default values– Points out between different events:

go restaurant, order, pay, leave– Fleet scripts (should not be activated)

Susan passed her favorite restaurant on her way to the museum

– Builds coherent interpretation from a collection of observation:

John went, ordered, paid, left– Did john eat dinner?

Why did waitress bring Susan a menu?– Because Susan asked to– So that Susan could decided what to eat

Focuses attention on unusual events– Waited for a long time, he got mad and left

CYC (D. Lenat 1990)Large knowledge-based for common sence projectrepresentation for events, objects, attributesissues of large scaleMotivation: brittleness, content focus not form, shared knowledgehand coded (10 million facts)

Page 28: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

28

OntologyOntologySpecification of concepts to be used for expressing knowledge– What are categories?– How they are related to each other– Collection (superset/subset)/Individual (has

parts)Person - Nation - NoseAhmed - Qatar - Ahmed’s Nosesubstance (retains its properties)

– Intangible(no mass)/tangible(mass)/Compositesets - numbers - eventsperson’s body - an orange

– Intrinsic (color)/Extrinsic– Event (walking)/Process (walking2min)– slots -Time (intervals) – agents (subset of composite)

collectionindividualbeliefs

– Space - causality - structures– persistence of object through time

Page 29: KHABEER As A Machine Translation Toolaref/504/ai1.pdf · 5. Problems, Problem Spaces and Search. z. To build a system to solve a particular problem, we need to: – Define the problem

29

Knowledge Representation Knowledge Representation SummarySummary

Syntactic-Semantic Spectrum of Rep.– Predicate Logic -- Production Rules– Nonmonotonic Systems -- Statistical Reasoning– Semantic nets -- Frames– Conceptual Dependency -- Scripts– Objects

Syntactic Representation– problem solving methods– adequate for any problem– slow for hard problem

Semantic Representation– Stronger– More Effectively

KR systems– not just to hold knowledge– but also to provide inference procedures

Role of knowledge– define the search space and the criteria for determining a

solution to a problem (essential knowledge)– improve the efficiency of a reasoning procedure

(heuristic knowledge)