4190.408 . 2015Spring ExpertSystems

knowledge,!represented! ... unusual!"space!cadetkeyboard".! Bio Intelligence! 4190.408 Ar#ficial! ... behavior! SMH.PAL,!IntelligentClinical!...

0 downloads 140 Views 3MB Size
4190.408  

 

 

 

 

 

 

 

Expert  Systems  

 2015-­‐Spring  

Byoung-­‐Tak  Zhang   TA:  Hyo-­‐Sun  Chun   School  of  Computer  Science  and  Engineering   Seoul  NaHonal  University  

Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Bethune  Cookman  University,  CIS  332]  

Expert  Systems   •  Computer  soLware  that:   –  Emulates  human  expert   –  Deals  with  small,  well  defined  domains  of  experHse   –  Is  able  to  solve  real-­‐world  problems   –  Is  able  to  act  as  a  cost-­‐effecHve  consultant   –  Can  explains  reasoning  behind  any  soluHons  it  finds   –  Should  be  able  to  learn  from  experience.  

Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Wikipedia]  

Expert  Systems   •  A  computer  system  that   emulates  the  decision-­‐making   ability  of  a  human  expert  

–  Designed  to  solve  complex   problems  by  reasoning  about   knowledge,  represented   primarily  as  if–then  rules  rather   than  through  convenHonal   procedural  code  

•  An  example  of  a  knowledge-­‐ based  system  

–  The  first  commercial  systems  to   use  a  knowledge-­‐based   architecture   Bio Intelligence  

A  Symbolics  Lisp  Machine:  An  Early   Pla]orm  for  Expert  Systems.  Note  the   unusual  "space  cadet  keyboard".  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Wikipedia]  

•  Divided  into  two  sub-­‐ systems:     –  Knowledge  base   •  The  knowledge  base  represents   facts  and  rules.    

–  Inference  engine   •  The  inference  engine  applies  the   rules  to  the  known  facts  to   deduce  new  facts.     •  Inference  engines  can  also   include  explanaHon  and   debugging  capabiliHes.   Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Bethune  Cookman  University,  CIS  332]  

Components  of  an  Expert  System   Expert  System   Knowledge   Base     User     Interface   Inference   Engine     User   Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Wikipedia]  

Knowledge  Base   •  A  technology  used  to  store  informaHon  used   by  a  computer  system   •  Represents  facts  about  the  world   •  In  early  expert  systems,  flat  asserHons  about   variables   •  In  later  expert  systems,  more  structure  and   uHlized  concepts  from  object-­‐oriented   programming   Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Wikipedia]  

Inference  Engine   •  An  automated  reasoning  system     –  Evaluates  the  current  state  of  the  knowledge-­‐base   –  applies  relevant  rules   –  asserts  new  knowledge  into  the  knowledge  base  

•  CapabiliHes  for  explanaHon   –  can  explain  to  a  user  the  chain  of  reasoning   –  forward  chaining:  data-­‐driven   –  backward  chaining:  goal-­‐driven   Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Wikipedia]  

Various  types  of  inference  engines   •  Truth  Maintenance   –  Truth  maintenance  systems  record  the  dependencies  in  a  knowledge-­‐base  so   that  when  facts  are  altered  dependent  knowledge  can  be  altered  accordingly.    

•  Hypothe#cal  Reasoning     –  In  hypotheHcal  reasoning,  the  knowledge  base  can  be  divided  up  into  many   possible  views.     –  This  allows  the  inference  engine  to  explore  mulHple  possibiliHes  in  parallel.    

•  Fuzzy  Logic   –  One  of  the  first  extensions  of  simply  using  rules  to  represent  knowledge  was   also  to  associate  a  probability  with  each  rule.    

•  Ontology  Classifica#on   –  With  the  addiHon  of  object  classes  to  the  knowledge  base  a  new  type  of   reasoning  was  possible.     –  Rather  than  reason  simply  about  the  values  of  the  objects  the  system  could   also  reason  about  the  structure  of  the  objects  as  well.     Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Bethune  Cookman  University,  CIS  332]  

Problem  Domain  vs.  Knowledge  Domain     •  An  expert’s  knowledge  is  specific  to  one   problem  domain  –  medicine,  finance,  science,   engineering,  etc.   •  The  expert’s  knowledge  about  solving  specific   problems  is  called  the  knowledge  domain.   •  The  problem  domain  is  always  a  superset  of   the  knowledge  domain.  

Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Bethune  Cookman  University,  CIS  332]  

Knowledge  Engineering   •  The  process  of  building  an  expert  system:   –  The  knowledge  engineer  establishes  a  dialog  with  the  human  expert  to   elicit  (obtain)  knowledge.   –  The  knowledge  engineer  codes  the  knowledge  explicitly  in  the   knowledge  base.   –  The  expert  evaluates  the  expert  system  and  gives  a  criHque  to  the   knowledge  engineer.  

Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Bethune  Cookman  University,  CIS  332]  

Development  of  an  Expert  System  

Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Bethune  Cookman  University,  CIS  332]  

The  Role  of  AI   •  An  algorithm  is  an  ideal  soluHon  guaranteed   to  yield  a  soluHon  in  a  finite  amount  of  Hme.   •  When  an  algorithm  is  not  available  or  is   insufficient,  we  rely  on  arHficial  intelligence.   •  Expert  system  relies  on  inference  (conclusion)   –  we  accept  a  “reasonable  soluHon.”  

Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Bethune  Cookman  University,  CIS  332]  

Uncertainty   •  Both  human  experts  and  expert  systems  must   be  able  to  deal  with  uncertainty.   •  It  is  easier  to  program  expert  systems  with   shallow  knowledge  than  with  deep   knowledge.   •  Shallow  knowledge  –  based  on  empirical  and   heurisHc  knowledge.   •  Deep  knowledge  –  based  on  basic  structure,   funcHon,  and  behavior  of  objects.   Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Bethune  Cookman  University,  CIS  332]  

Advantages  of  Expert  Systems   Increased  availability   Reduced  cost   Reduced  danger   Performance   MulHple  experHse   Increased  reliability   ExplanaHon   Fast  response   Steady,  unemoHonal,  and   complete  responses  at  all  Hmes   •  Intelligent  database   •  •  •  •  •  •  •  •  • 

Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[University  of  Nomngham,  G64FAI]  

Problems  with  Expert  System   •  Limited  domain     •  Systems  are  not  always  up  to  date,  and  don’t   learn     •  No  “common  sense”     •  Experts  needed  to  setup  and  maintain  system     •  Who  is  responsible  if  the  advice  is  wrong?  

Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Wikipedia]  

Applica#ons   Category  

Problem  Addressed  

Examples  

InterpretaHon  

Inferring  situaHon  descripHons  from  sensor  data  

Hearsay  (Speech  RecogniHon),   PROSPECTOR  

PredicHon  

Inferring  likely  consequences  of  given  situaHons  

Preterm  Birth  Risk  Assessment  

Diagnosis  

Inferring  system  malfuncHons  from  observables  

CADUCEUS,  MYCIN,  PUFF,  Mistral,   Eydenet,  Kaleidos  

Design  

Configuring  objects  under  constraints  

Dendral,  Mortgage  Loan  Advisor,  R1   (Dec  Vax  ConfiguraHon)  

Planning  

Designing  acHons  

Mission  Planning  for  Autonomous   Underwater  Vehicle  

Monitoring  

Comparing  observaHons  to  plan  vulnerabiliHes  

REACTOR  

Debugging  

Providing  incremental  soluHons  for  complex   problems  

SAINT,  MATHLAB,  MACSYMA  

Repair  

ExecuHng  a  plan  to  administer  a  prescribed   remedy  

Toxic  Spill  Crisis  Management  

InstrucHon  

Diagnosing,  assessing,  and  repairing  student   behavior  

SMH.PAL,  Intelligent  Clinical   Training,STEAMER  

Control  

InterpreHng,   predicHng,  repairing,  and   4190.408    Ar#ficial   Real  Time   Process   Control,  Space   Bio Intelligence   (2015-­‐Spring)   Intelligences   ystem  behaviors   monitoring   Shurle  Mission  Control  

Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[University  of  Nomngham,  G64FAI]  

Expert  System  Tools   •  Algorithmic  languages  

–  (such  as  'C',  Pascal,  Basic)    

•  Symbolic  languages   –  (such  as  Prolog,  LISP)    

Bio Intelligence  

•  Development  Environments   –  (such  as  Art,  KEE,  LOOPS)  —     •  Expert  System  Shells   –  (such  as  Crystal,  XpertRule,   Leonardo,  Xi-­‐Plus)  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[University  of  Nomngham,  G64FAI]  

Algorithmic  Languages   •  Flexible  and  powerful     •  They  can  be  used  to  tailor  a  system  exactly  to   an  applicaHon  —   •   Lacking  in  knowledge  engineering  framework.  

Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[University  of  Nomngham,  G64FAI]  

Symbolic  Languages   •  Computer  languages  for  logic  programming   must  have  structures  for  storing  and  retrieving   known  and  deduced  facts  from  a  fact  base  or   knowledge  base,  and  they  must  have   funcHons  or  procedures  for  deducing  new   facts.  —     •  LISP  —     •  Prolog   Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[University  of  Nomngham,  G64FAI]  

Development  Environments   •  Expert  system  programming  environments  are   special  packages  of  pre-­‐wriren  code.  They  are   "power  tools"  for  building  knowledge-­‐based   systems.     •  They  provide  a  set  of  building  blocks  that   cater  for  the  programmers  needs,  and  hence   are  known  as  ‘tool  kits’.     •  In  most  cases,  the  price  to  pay  is  the  loss  of   flexibility,  however,  in  using  a  higher-­‐level   tool.   Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[University  of  Nomngham,  G64FAI]  

Expert  System  Shells   •  Shells  are  tools  for  building  expert  systems  that  provide  knowledge   representaHon  faciliHes  and  inferencing  mechanisms.  —     •  The  programmer  must  gain  detailed  knowledge  about  a  parHcular  domain   from  an  expert  and  informaHon  source.  —     •  May  be  thought  of  as  an  expert  system  with  all  the  domain  specific   knowledge  removed  and  a  facility  for  entering  a  new  knowledge-­‐base   provided.  

Bio Intelligence  

4190.408    Ar#ficial  Intelligence  (2015-­‐Spring)  

[Decision Support Systems and Intelligent Systems, E. Turban and J. E. Aronson]

History of Expert Systems 1. Early to Mid-1960s – 

■  ■ 

One attempt: the General-purpose Problem Solver (GPS)

General-purpose Problem Solver (GPS) A procedure developed by Newell and Simon [1973] from their Logic Theory Machine – 

Attempted to create an "intelligent" computer • 

–  – 

general problem-solving methods applicable across domains

Predecessor to ES Not successful, but a good start

5 23

[Decision Support Systems and Intelligent Systems, E. Turban and J. E. Aronson]

2. Mid-1960s: Special-purpose ES programs –  – 

■ 

DENDRAL MYCIN

Researchers recognized that the problem-solving mechanism is only a small part of a complete, intelligent computer system –  – 

–  – 

General problem solvers cannot be used to build high performance ES Human problem solvers are good only if they operate in a very narrow domain Expert systems must be constantly updated with new information The complexity of problems requires a considerable amount of knowledge about the problem area

6 24

[Decision Support Systems and Intelligent Systems, E. Turban and J. E. Aronson]

3. Mid 1970s –  –  – 

Several Real Expert Systems Emerge Recognition of the Central Role of Knowledge AI Scientists Develop •  • 

■ 

Comprehensive knowledge representation theories General-purpose, decision-making procedures and inferences

Limited Success Because –  – 

Knowledge is Too Broad and Diverse Efforts to Solve Fairly General KnowledgeBased Problems were Premature 7 25

[Decision Support Systems and Intelligent Systems, E. Turban and J. E. Aronson]

BUT ■ 

Several knowledge representations worked

Key Insight ■ 

The power of an ES is derived from the specific knowledge it possesses, not from the particular formalisms and inference schemes it employs

8 26

[Decision Support Systems and Intelligent Systems, E. Turban and J. E. Aronson]

4. Early 1980s ■ 

ES Technology Starts to go Commercial –  –  – 

■ 

Programming Tools and Shells Appear –  –  –  – 

■ 

XCON XSEL CATS-1 EMYCIN EXPERT META-DENDRAL EURISKO

About 1/3 of These Systems Are Very Successful and Are Still in Use

9 27

[Decision Support Systems and Intelligent Systems, E. Turban and J. E. Aronson]

Latest ES Developments ■  ■  ■  ■  ■ 

Many tools to expedite the construction of ES at a reduced cost Dissemination of ES in thousands of organizations Extensive integration of ES with other CBIS Increased use of expert systems in many tasks Use of ES technology to expedite IS construction (ES Shell) 10 28