Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin –...

30
Artificial Artificial Intelligence Intelligence Dr. Paul Wagner Dr. Paul Wagner Department of Computer Department of Computer Science Science University of Wisconsin – Eau University of Wisconsin – Eau Claire Claire

Transcript of Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin –...

Page 1: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

Artificial IntelligenceArtificial Intelligence

Dr. Paul WagnerDr. Paul Wagner

Department of Computer ScienceDepartment of Computer Science

University of Wisconsin – Eau University of Wisconsin – Eau ClaireClaire

Page 2: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

MessagesMessages

Artificial Intelligence (AI) is an Artificial Intelligence (AI) is an interesting sub-field of computer interesting sub-field of computer science that provides many science that provides many contributions to the overall fieldcontributions to the overall field

CS 420, as the AI course at CS 420, as the AI course at UWEC, is a good opportunity to UWEC, is a good opportunity to begin to explore these issuesbegin to explore these issues

Page 3: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

OutlineOutline

OverviewOverview AI TopicsAI Topics

– Knowledge representationKnowledge representation– Problem solving and search space manipulationProblem solving and search space manipulation– PlanningPlanning– LearningLearning– CommunicatingCommunicating– UncertaintyUncertainty– Intelligent agentsIntelligent agents– RoboticsRobotics

AI LanguagesAI Languages MICS Robot Contest VideoMICS Robot Contest Video

Page 4: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

Overview of Artificial Overview of Artificial IntelligenceIntelligence Definitions – four Definitions – four

major major combinationscombinations– Based on Based on thinkingthinking

or or actingacting– Based on activityBased on activity

like humanslike humans or or performed in performed in rational wayrational way

Systems Systems that that think like think like humanshumans

Systems Systems that that think think rationallrationallyy

Systems Systems that act that act like like humanshumans

Systems Systems that act that act rationallrationallyy

Page 5: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

AI DefinitionsAI Definitions

Acting HumanlyActing Humanly– Turing Test – computer passes test if Turing Test – computer passes test if

a human interrogator asking written a human interrogator asking written questions can distinguish written questions can distinguish written answers from computer or humananswers from computer or human

– Computer needs:Computer needs: Natural language processingNatural language processing Knowledge representationKnowledge representation Automated reasoningAutomated reasoning Machine learningMachine learning

Page 6: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

AI Definitions (2)AI Definitions (2)

– Total Turing Test – includes video Total Turing Test – includes video component (to test subject’s component (to test subject’s perceptual abilities) and opportunity perceptual abilities) and opportunity to pass physical objects to subjectto pass physical objects to subject

– Computer also needs:Computer also needs: Computer visionComputer vision RoboticsRobotics

Page 7: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

AI Definitions (3)AI Definitions (3)

Thinking HumanlyThinking Humanly– Cognitive Modeling approach to AICognitive Modeling approach to AI– Involves crossover between Involves crossover between

computer science and psychology – computer science and psychology – cognitive sciencecognitive science

– Areas of interestAreas of interest Cognitive modelsCognitive models Neural networksNeural networks

Page 8: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

AI Definitions (4)AI Definitions (4)

Thinking RationallyThinking Rationally– ““Laws of thought” approach to AILaws of thought” approach to AI– Goal: solve any problem based on logical Goal: solve any problem based on logical

manipulationmanipulation– ProblemsProblems

Difficult to represent certain types of knowledge Difficult to represent certain types of knowledge (e.g. common sense, informal knowledge)(e.g. common sense, informal knowledge)

Difference between solving problems in Difference between solving problems in principle and in practiceprinciple and in practice

– E.g. computational limitsE.g. computational limits

Page 9: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

AI Definitions (4)AI Definitions (4)

Acting RationallyActing Rationally– ““Design a rational agent” approach Design a rational agent” approach

to AIto AI– Advantages over logic approachAdvantages over logic approach

Logic is only one tool or many that can Logic is only one tool or many that can be used to design rational agentbe used to design rational agent

Scientific advances can provide more Scientific advances can provide more tools for developing better agentstools for developing better agents

Page 10: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

Knowledge Knowledge RepresentationRepresentation How to represent information?How to represent information? Generally, we use some sort of tree, Generally, we use some sort of tree,

grid or networkgrid or network OptionsOptions

– OO programming languages: OO programming languages: classes/objectsclasses/objects

– Relational database system: Relational database system: tables/rows/columnstables/rows/columns

ProblemProblem– The world is more varied, with many types The world is more varied, with many types

of things to representof things to represent

Page 11: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

Knowledge Knowledge Representation (2)Representation (2) Abstract ObjectsAbstract Objects

– SetsSets– SentencesSentences– MeasurementsMeasurements

TimesTimes WeightsWeights

Generalized EventsGeneralized Events– IntervalsIntervals– PlacesPlaces– Physical ObjectsPhysical Objects– ProcessesProcesses

Page 12: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

Knowledge Knowledge Representation (3)Representation (3) Some things are very difficult to representSome things are very difficult to represent

– Common senseCommon sense See See http://www.cyc.com/http://www.cyc.com/

– Combinations of multiple typesCombinations of multiple types Issues of:Issues of:

– TypeType– ScaleScale– GranularityGranularity– CombinationCombination

Other QuestionsOther Questions– How to distinguish knowledge and belief?How to distinguish knowledge and belief?– What is the best way to reason with this What is the best way to reason with this

information?information?

Page 13: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

Problem Solving and Problem Solving and Search Space Search Space ManipulationManipulation Many Algorithmic Approaches to Many Algorithmic Approaches to

Problem SolvingProblem Solving– Depth-First SearchDepth-First Search– Breadth-First SearchBreadth-First Search

VariationsVariations– Depth-Limited SearchDepth-Limited Search– Iterative Deepening Depth-First Iterative Deepening Depth-First

SearchSearch– Bi-directional SearchBi-directional Search

Page 14: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

Problem Solving and Problem Solving and Search Space Search Space Manipulation (2)Manipulation (2) Smarter SearchSmarter Search

– Greedy best-first searchGreedy best-first search– A* search (combine costs of path so A* search (combine costs of path so

far plus path from current node to far plus path from current node to goal)goal)

– Memory-bounded heuristic searchMemory-bounded heuristic search Heuristic – means of estimating a Heuristic – means of estimating a

measurement such as cost of searchmeasurement such as cost of search

Page 15: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

Problem Solving and Problem Solving and Search Space Search Space Manipulation (3)Manipulation (3) IssuesIssues

– Avoiding repeated searchAvoiding repeated search– Searching with partial informationSearching with partial information

Page 16: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

Problem Solving and Problem Solving and Search Space Search Space Manipulation (4)Manipulation (4) Adversarial SearchAdversarial Search

– E.g. games and game treesE.g. games and game trees– Minimax algorithmMinimax algorithm– Alpha-Beta pruningAlpha-Beta pruning

Page 17: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

Problem Solving and Problem Solving and Search Space Search Space Manipulation (5)Manipulation (5) Applications of Problem SolvingApplications of Problem Solving

– Expert SystemsExpert Systems Approximating the functionality of an Approximating the functionality of an

absent human expertabsent human expert

– RoboticsRobotics Encountering unexpected obstaclesEncountering unexpected obstacles

Page 18: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

PlanningPlanning

Many types of Many types of problemsproblems– ““Blocks world”Blocks world”– Getting yourself from Getting yourself from

Eau Claire to the Eau Claire to the AAAI conference in AAAI conference in BostonBoston

– Changing a flat tireChanging a flat tire– Completing all of Completing all of

your projects at the your projects at the end of the semesterend of the semester

– Developing a large Developing a large software applicationsoftware application

Page 19: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

Planning (2)Planning (2)

ApproachesApproaches– State-based searchState-based search– Partial-order planningPartial-order planning– Planning graphsPlanning graphs

IssuesIssues– TimeTime– SchedulingScheduling– ResourcesResources

Page 20: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

LearningLearning

Definition - Building on current Definition - Building on current knowledge by using experience to knowledge by using experience to improve a systemimprove a system

Various approachesVarious approaches– Supervised/unsupervised/reinforcementSupervised/unsupervised/reinforcement

Forms of learning algorithmsForms of learning algorithms– Inductive logicInductive logic

Example: given a set of point, approximate a lineExample: given a set of point, approximate a line

– Decision tree (set of questions, act differently Decision tree (set of questions, act differently depending on answer)depending on answer)

Page 21: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

Learning (2)Learning (2)

IssuesIssues– Computational Learning TheoryComputational Learning Theory

Intersection of theoretical CS, AI, Intersection of theoretical CS, AI, statistics statistics

– How many examples do you need?How many examples do you need?

Page 22: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

CommunicatingCommunicating

Major issue - Natural language Major issue - Natural language processingprocessing– Many issuesMany issues

SyntaxSyntax SemanticsSemantics ContextContext

– StepsSteps PerceptionPerception ParsingParsing AnalysisAnalysis DisambiguationDisambiguation IncorporationIncorporation

Page 23: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

UncertaintyUncertainty

Much knowledge is not absoluteMuch knowledge is not absolute– Boundary between knowledge and belief is grayBoundary between knowledge and belief is gray

Techniques for dealing with uncertaintyTechniques for dealing with uncertainty– Probabilistic reasoningProbabilistic reasoning– Probabilistic reasoning over timeProbabilistic reasoning over time– Fuzzy sets / fuzzy logicFuzzy sets / fuzzy logic– Simple decision-making (evaluating utility)Simple decision-making (evaluating utility)– Complex decision-making (taking ability to Complex decision-making (taking ability to

reevaluate into account)reevaluate into account) ApplicationsApplications

– Expert systemsExpert systems

Page 24: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

Intelligent AgentsIntelligent Agents

Everything we’ve talked about Everything we’ve talked about can be viewed in terms of can be viewed in terms of embedding intelligence within an embedding intelligence within an agentagent– Software systemSoftware system– Machine with embedded softwareMachine with embedded software– RobotRobot

Page 25: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

Intelligent Agents (2)Intelligent Agents (2)

Issues for agentsIssues for agents– Limitations on memoryLimitations on memory– Perceiving its environmentPerceiving its environment– Working with other agentsWorking with other agents– Affecting its environment (through Affecting its environment (through

actuators)actuators) ProcessesProcesses

– Simple – based on rulesSimple – based on rules– Complex – based on multiple pieces of logic, Complex – based on multiple pieces of logic,

dealing with uncertaintydealing with uncertainty

Page 26: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

RoboticsRobotics

Field encompassing elements of computer Field encompassing elements of computer science/AI, engineering, physical systemsscience/AI, engineering, physical systems

IssuesIssues– Many that we’ve discussed, plus:Many that we’ve discussed, plus:– PerceptionPerception– ActuationActuation

Recent successesRecent successes– Worker bots (e.g. floor cleaners)Worker bots (e.g. floor cleaners)– Intelligent navigation (DARPA vehicle contest)Intelligent navigation (DARPA vehicle contest)

Test environmentsTest environments– Lego MindstormsLego Mindstorms– Other robot packages or custom systemsOther robot packages or custom systems

Page 27: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

AI LanguagesAI Languages

Scheme / LISPScheme / LISP– FunctionalFunctional– Simple knowledge representation (list)Simple knowledge representation (list)– Easy to apply functionality to represented elementsEasy to apply functionality to represented elements

PrologProlog– Logic-basedLogic-based– Facts and rules easily representedFacts and rules easily represented– Built-in search engineBuilt-in search engine

Specialized languagesSpecialized languages– Rule languages (e.g. CLIPS)Rule languages (e.g. CLIPS)– Planning languages (e.g. STRIPS)Planning languages (e.g. STRIPS)

Page 28: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

CS 420CS 420

Spring semester, about every Spring semester, about every other yearother year

Will be offered Spring 2007Will be offered Spring 2007 Prerequisite: CS 330 (to get Prerequisite: CS 330 (to get

Scheme and Prolog background)Scheme and Prolog background) TopicsTopics

– All of the above!All of the above!

Page 29: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

CS 420 (2)CS 420 (2)

Possible ProjectsPossible Projects– Neural network to simulate decision Neural network to simulate decision

making, natural language processingmaking, natural language processing– Software development planning Software development planning

through cooperating intelligent agentsthrough cooperating intelligent agents– Expert system for deciding which Expert system for deciding which

courses to take to complete a CS majorcourses to take to complete a CS major– Sumo robots?Sumo robots?

Page 30: Artificial Intelligence Dr. Paul Wagner Department of Computer Science University of Wisconsin – Eau Claire.

MICS Robot Contest MICS Robot Contest VideoVideo http://http://

video.google.com/videoplay?docivideo.google.com/videoplay?docidd=7851913746457357108&hl=en=7851913746457357108&hl=en