20
 C    o    p    y    r i    g h    t R  . W    e b    e    r INFO 629 Concepts in Artifcial Intelligence Expert Systems  Fall 2004 Proessor! "r# $osina %e&er

629ppt2

Embed Size (px)

DESCRIPTION

knowledge base system

Citation preview

  • 5/17/2018 629ppt2

    1/20

    Copyrigh

    tR.W

    eb

    er

    INFO 629 Concepts in Artifcial Intelligence

    Expert Systems

    Fall 2004

    Proessor! "r# $osina %e&er

  • 5/17/2018 629ppt2

    2/20

    Copyrigh

    tR.W

    eb

    er

    'ig(lig(ts Concept

    Methodology Knowledge and reasoning

    Knowledge representation

    Forward, backward chaining ES and AI tasks

    Maintenance

    Knowledge acquisition

    i!ited, bounded do!ains

    "se o# shells

    Ad$antages%disad$antages o# ES

  • 5/17/2018 629ppt2

    3/20

    Copyrigh

    tR.W

    eb

    er

    Expert Systems

    Co!puter syste!s that can per#or!e&pert tasks'(general, $ague)

    A !ethodology that !anipulatese&plicit knowledge with an in#erence

    engine to per#or! AI tasks'

  • 5/17/2018 629ppt2

    4/20

    Copyrigh

    tR.W

    eb

    er

    t(e concept

    knowledgebase

    (e.g.,framesand methods)

    knowledgebase

    (e.g.,framesand methods)

    e&pertproble!

    e&pertproble!

    inferenceengine

    (agenda)

    inferenceengine

    (agenda)e&pert

    solution

    e&pert

    solution

    )no*le+ge

    reasoning

  • 5/17/2018 629ppt2

    5/20

    Copyrigh

    tR.W

    eb

    er

    e&pertsolution

    e&pertsolution

    *he co!plete !ethodology

    knowledgebase

    (e.g.,framesand methods)

    knowledgebase

    (e.g.,framesand methods)

    explanation

    explanation

    general

    knowledge

    general

    knowledge

    userInt

    erfac

    e

    userInterfac

    e

    e&pertproble!

    e&pertproble!inference

    engine(agenda)

    inferenceengine

    (agenda)

    working memory(short-term mem/information)

    working memory(short-term mem/information)

    Knowledge acuisition

    Knowledge acuisition

  • 5/17/2018 629ppt2

    6/20

    Copyrigh

    tR.W

    eb

    er

    Expert Systems

    Knowledge

    and

    reasoning

  • 5/17/2018 629ppt2

    7/20

    Copyrigh

    tR.W

    eb

    er

    ,no*le+ge representation ormalisms

    (production) rules

    #ra!es (concepts, ob+ects,#acts)

    belie# networks

    !ethods

    ob+ectoriented se!antic nets

    logic

  • 5/17/2018 629ppt2

    8/20

    Copyrigh

    tR.W

    eb

    er

    In#erence Engines

    Forward chaining-Analysis, !ultiple outco!es

    .ackward chaining-Atte!pt to test li!ited nu!ber o#hypotheses

  • 5/17/2018 629ppt2

    9/20

    Copyrigh

    tR.W

    eb

    er

    -aintenance

    Maintenance #ocus on knowledge Co!ple&ity o# interrelations a!ong

    rules

    /i0cult to auto!ate !aintenance

  • 5/17/2018 629ppt2

    10/20

    Copyrigh

    tR.W

    eb

    er

    ,no*le+ge ac./isition

    Fro! se$eral hu!an e&perts- "nstructured inter$iews

    - Structured inter$iews

    - Methods learned #ro! psychology Auto!ated through !achine learning

    !ethods

  • 5/17/2018 629ppt2

    11/20

    Copyrigh

    tR.W

    eb

    er

    "omains

    i!ited, bounded do!ains

  • 5/17/2018 629ppt2

    12/20

    Copyrigh

    tR.W

    eb

    er

    ES S(ells

    Easy prototyping to test ideas KA11A 1C

    CI1S

    E&a!ples in KA11A 1C

  • 5/17/2018 629ppt2

    13/20

    Copyrigh

    tR.W

    eb

    er

    ES an+ AI tas)s

    Fro!2

    /urkin, 3'(4556)'

    E&pertSyste!s2design andde$elop!ent'1rentice7all,Inc', 8ew3ersey'

  • 5/17/2018 629ppt2

    14/20

    Copyrigh

    tR.W

    eb

    er

    a+antages 1i

    1er!anence o# knowledge E&pert syste!s do not #orget orretire or quit, but hu!an e&perts !ay

    .readth 9ne ES can (and should) entail knowledge learned#ro! an unli!ited nu!ber o# hu!an e&perts'

    :eproducibility Many copies o# an e&pert syste! can be

    !ade, but training new hu!an e&perts is ti!econsu!ing ande&pensi$e'

    *i!eliness Fraud and%or errors can be pre$ented' In#or!ationis a$ailable sooner #or decision !aking

    Entry barriers E&pert syste!s can help a ;r! create entry

    barriers #or potential co!petitors

  • 5/17/2018 629ppt2

    15/20

    Copyrigh

    tR.W

    eb

    er

    a+antages 1ii

    E0ciency can increase throughput and decreasepersonnel costs

    Although e&pert syste!s !ay be e&pensi$e to build and!aintain, they are ine&pensi$e to operate

    I# there is a !a

  • 5/17/2018 629ppt2

    16/20

    Copyrigh

    tR.W

    eb

    er

    a+antages 1iii

    /ocu!entation An e&pert syste! can pro$ide per!anent

    docu!entation o# the decision process Increased a$ailability2 the !ass production o# e&pertise

    Co!pleteness An e&pert syste! can re$iew all thetransactions, a hu!an e&pert can only re$iew a sa!ple? an ESsolution will always be co!plete and deter!inistic

    Consistency @ith e&pert syste!s si!ilar transactions handledin the sa!e way' 7u!ans are inuenced by recency e>ects andpri!acy e>ects (early in#or!ation do!inates the +udg!ent)'

  • 5/17/2018 629ppt2

    17/20

    Copyrigh

    tR.W

    eb

    er

    a+antages 1i

    /i>erentiation In so!e cases, an e&pert syste! candi>erentiate a product or can be related to the #ocus o# the ;r!

    :educed danger2 ES can be used in any en$iron!ent

    :eliability2 ES will keep working properly regardless o# o#e&ternal conditions that !ay cause stress to hu!ans

    E&planation2 ES can trace back their reasoning pro$iding

    +usti;cation, increasing the con;dence that the correct decisionwas !ade

    Indirect ad$antage is that the de$elop!ent o# an ES requiresthat knowledge and processes are $eri;ed #or correctness,co!pleteness, and consistency'

  • 5/17/2018 629ppt2

    18/20

    Copyrigh

    tR.W

    eb

    er

    +isa+antages

    Co!!on sense In addition to a great deal o# technicalknowledge, hu!an e&perts ha$e co!!on sense' *o progra!co!!on sense in an ES, you !ust acquire and represent rules,which is not #easible'

    Creati$ity 7u!an e&perts can respond creati$ely to unusualsituations, e&pert syste!s cannot'

    earning 7u!an e&perts auto!atically adapt to changing

    en$iron!ents? e&pert syste!s !ust be e&plicitly updated' Co!ple&ity and interrelations o# rules grow e&ponentially as

    !ore rules are added'

    Sensory E&perience 7u!an e&perts ha$e a$ailable to the! awide range o# sensory e&perience? e&pert syste!s are currently

    dependent on sy!bolic input'

  • 5/17/2018 629ppt2

    19/20

    Copyrigh

    tR.W

    eb

    er

    +isa+antages 1ii

    /egradation E&pert syste!s are not good at recogni

  • 5/17/2018 629ppt2

    20/20

    Copyrigh

    tR.W

    eb

    er

    Necessary gro/n+s or comp/ter/n+erstan+ing

    Ability to represent knowledge andreason with it'

    1ercei$e equi$alences and analogiesbetween two di>erent representations

    o# the sa!e entity%situation' earning and reorgani