Expert system : Examples

1
Expert system Examples
2
Classic systems

• Pre-1980s
• Systems showed how to
• capture heuristic knowledge and store it
• make a software that could mimic advice
  dispensation like expert human do.
• Techniques that were implemented were used in
  many subsequent systems,
• Many expert system shells were developed.

3
Expert systems

• MACSYMA
• advised the user on how to solve complex maths
  problems.
• DENDRAL
• advised the user on how to interpret the output
  from a mass spectrograph
• MYCIN
• PROSPECTOR
• R1/XCON

4
Others

• CENTAUR
• INTERNIST
• PUFF
• CASNET
• DELTA - locomotive engineering
• Drilling Advisor - oilfield prospecting
• ExperTax - tax minimisation advice
• XSEL - computer sales

All medical expert systems
5
Task Classification

• Various tasks could be performed
• A layout presented by Hayes-Roth colleagues in
  1983 is presented here

6
Diagnosis

• finding faults in a system, or diseases in a
  living system
• MYCIN - diagnosed blood infection. Shortliffe,
  1976.

7
Interpretation

• The analysis of data, to determine their meaning
• PROSPECTOR - interpreted geological data as
  potential evidence for mineral deposits. Duda,
  Hart, et al 1976.

8
Monitoring

• The continuous interpretation of signals from a
  system for avoiding dangerous situations
• NAVEX - monitored radar data and estimated the
  velocity and position of the space shuttle.
  Marsh, 1984

9
(No Transcript)
10
Design

• To ensure production of specifications,
  satisfying particular requirements
• R1/XCON - configured VAX computer systems on the
  basis of customers' needs. McDermott, 1980.

11
Planning

• Production of a sequence of actions that will
  achieve a particular goal.
• MOLGEN - planned chemical processes whose purpose
  was to analyse and synthesise DNA. Stefik, 1981.

12
(No Transcript)
13
Instruction Intelligent Tutoring Systems

• Teaching a student a body of knowledge, varying
  the teaching according to assessments
• SOPHIE - instructed the student on the repair of
  an electronic power-pack. Brown, Burton de
  Kleer, 1982.

14
Prediction

• Forecasting future events, using a model based on
  past events.
• PLANT - predicted the damage to be expected when
  a corn crop was invaded by black cutworm.
  Boulanger, 1983.

15
(No Transcript)
16
Debugging repair

• Generating, administering remedies for system
  faults.
• COOKER ADVISER - provides repair advice with
  respect to canned soup sterilising machines.
  Texas Instruments, 1986.

17
(No Transcript)
18
Controls

• Governing the behaviour of a system by
  anticipating problems, planning solutions, and
  monitoring actions.
• VENTILATOR MANAGEMENT ASSISTANT - scrutinised the
  data from hospital breathing-support machines,
  and provided accounts of the patients'
  conditions. Fagan, 1978.

19
MYCIN Diagnosis System

• Domain diagnose blood infections of the sort
  that might be contracted in hospital
• Developed by Edward Shortliffe and colleagues,
  1972 to late 1970s.

20
MYCIN

• Purpose to assist a physician, who was not an
  expert in the field of antibiotics, with the
  diagnosis treatment of blood disorders (and in
  particular to establish whether the patient was
  suffering from a serious infection like
  meningitis).
• Input symptoms test results
• Output a diagnosis, accompanied by a degree of
  certainty, recommended therapy

21
MYCIN

• Knowledge representation production rules
• Inference engine Mixed chaining, but principally
  backward chaining from a top goal
• Dealing with uncertainty By calculating
  certainty factors.

22
MYCIN

• A Complete system that did complex task.
• Performed better than medical students and
  non-specialist doctors.
• Performed equally good to blood infection
  specialist doctors
• MYCIN was a laboratory demonstration only
• Was never commercialised, was never marketed, or
  installed in a hospital and used for routine work.

23
MYCIN

• Description of the system.
• Written in LISP.
• MYCIN was a mixed-chaining production system.

24
MYCIN

• The sequence of operations was that the system
  asked
• Questions to get general details about the
  patient (name, age, sex, race, clinical test
  results already known).
• Questions designed to find a possible, general
  diagnosis (e.g. the patient has probably got some
  form of meningitis), by forward chaining.

25
MYCIN

• Questions designed to test this theory, and
  establish specific details, by backward chaining.
  This was where most of the reasoning was done.
• Questions designed to produce a recommended
  treatment, again by forward chaining.

26
MYCIN

• MYCIN could explain its reasoning in a rather
  simple way
• when asked "Why do you think that is the
  diagnosis?, MYCIN listed the rules it had
  applied, in reverse order, with CFs.
• When asked "Why do you want to know that?", MYCIN
  described the rule it was trying to execute, and
  what value it was trying to find.

27
MYCIN

• A Consultation with MYCIN
• _at_ mycin
• MYCIN 26-March
• Instructions? (Y or N)
• N
• 26-March 1042
• -----PATIENT-1-----
• Patient's name
• Yousaf

28
Consultation with MYCIN

• Age?
• 34
• Sex?
• Why
• i.e. Why is it important to determine the sex of
  PATIENT-1?
• The sex of the patient is one of the four
  standard parameters that identify a patient. This
  may be useful later.

29
Consultation with MYCIN

• Sex?
• Male
• Race?
• Sindhi
• Are there any cultures for Yousaf that may be
  related to the present illness, and from which
  organisms have been grown successfully in the
  lab?
• No

30
Consultation with MYCIN

• Are there any pending cultures for Yousaf?
• Yes
• -----CULTURE-1-----
• From what site was the specimen for CULTURE-1
  taken?
• Cerebro Spinal Fluid
• Please give the date and time when the pending
  CSF culture was obtained.

31
Consultation with MYCIN

• 26-March-700
• Was a smear examined at the time the specimen
  was obtained?
• Yes
• Were there any cultures seen on the smear of the
  specimen?
• No
• Are there any other pending cultures?

32
Consultation with MYCIN

• No
• Has Yousaf recently had symptoms of persistent
  headache or other abnormal neurological symptoms
  (dizziness, lethargy, etc)?
• Yes
• Has Yousaf recently had objective evidence of
  abnormal neurological signs (nuchal rigidity,
  coma, seizures, etc) documented by a physician?

33
A Consultation with MYCIN

• Yes
• The CSF cultures will be considered to be
  associated with MENINGITIS.
• Please give the date on which clinical evidence
  (symptoms, signs, or laboratory tests) of the
  meningitis first appeared.
• 25 March 1000
• Has Yousaf had neurosurgery?

34
A Consultation with MYCIN

• Why
• i.e. WHY is it important to determine whether
  PATIENT-1 has had neurosurgery?
• This will aid in determining the organisms (other
  than those seen on cultures or smears) that might
  be causing the infection.

35
A Consultation with MYCIN

• It has already been established that
• 1 the culture that requires therapy is
  meningitis, and
• 2 organisms were not seen on the stain of the
  pending csf culture (CULTURE-1), and
• 3 the type of infection is bacterial

36
A Consultation with MYCIN

• Therefore, if
• 4 Yousaf has had neurosurgery, and
• 5 the time since the neurosurgical operation is
  less than two months, and
• 6 the type of neurosurgery is not
  ventricular-urethal shunt
• then
• there is evidence that the organisms that might
  be causing the infection are staphylococcus-coag-p
  os (0.75), staphylococcus-coag-neg (.5), e.coli
  (.4), klebsiella (0.3)
• Author Yu RULE511

37
A Consultation with MYCIN

• Has Yousaf had neurosurgery?
• How 3
• i.e. HOW was it established that the type of
  infection is bacterial?

38
A Consultation with MYCIN

• The following rules concluded about the
  likelihood that the type of infection is
  bacterial
• Rule Yes No
• 500 (.4)
• 501 (.17)
• 502 (.1)
• 524 (.42)
• 504 (.28)
• Thus it has been established that the type of
  infection might be bacterial.

39
A Consultation with MYCIN

• Has Yousaf had neurosurgery?
• No
• Is Yousaf a burn patient?
• Yes

40
A Consultation with MYCIN

• I have found evidence (RULE507) that treatment
  should cover for Diplococcus pneumoniae causing
  meningitis.
• I have found evidence (RULE545) that treatment
  should cover for e.coli causing meningitis.
• I have found evidence (RULE545 RULE578) that
  treatment should cover for Pseudomonas aeruginosa
  causing meningitis.

41
A Consultation with MYCIN

• Does Yousaf have focal neurological signs?
• No
• Determining which dosages are desirable for use
  against the e.coli...
• Determining which drugs are desirable for use
  ...

42
A Consultation with MYCIN

• INFECTION-1 is MENINGITIS
• ltITEM-1gt DIPLOCOCCUS-PNEUMONIAE (from clinical
  evidence only)
• ltITEM-2gt E.COLI (from clinical evidence only)
• ltITEM-3gt PSEUDOMONAS AERUGINOSA (from clinical
  evidence only)

43
A Consultation with MYCIN

• REC-1 My preferred therapy recommendation is as
  follows
• In order to cover for items lt1 2 3gt
• Give the following in combination
• 1) AMPICILLIN
• Dose 3.85g(30.8ml) q4h IV
• 2) GENTAMICIN
• Dose 131mg(3.2ml) q8h IV

44
A Consultation with MYCIN

• Comments monitor serum concentrations.
• Since high concentrations of penicillins can
  inactivate aminoglycosides, do not mix these
  antibiotics in the same IV bottle.
• Do you wish to see the next choice therapy?
• No

45
CROP ADVISOR

• Developed by ICI (in 1989) to advise cereal grain
  farmers on appropriate fertilisers and pesticides
  for their farms.
• The choice of chemical, amount, and time of
  application depends on such factors as crop to be
  grown, previous cropping, soil condition, acidity
  of soil, and weather.
• Farmers can access the system via the internet.

46
CROP ADVISOR

• Given relevant data, the system produces various
  financial return projections for different
  application rates of different chemicals.
• The system uses statistical reasoning to come to
  these conclusions.
• If the question asked is outside the system's
  expertise, it refers the caller to a human
  expert.

47
CROP ADVISOR

• The chief advantages of this system have been
• that employees at ICI have been relieved of the
  need to provide lengthy telephone advice
  sessions,
• and the quality of the advice has become much
  more uniform, which has increased confidence in
  the company's products.

48
R1/XCON

• Knowledge domain Configuring VAX computers, to
  customers' specifications.
• Written by John McDermott and colleagues, 1978 -
  1981
• Input Required characteristics of the computer
  system.
• Output Specification for the computer system.

49
R1/XCON

• Knowledge representation Production rules.
• Inference engine Forward chaining the output
  specification was assembled in working memory.
• Dealing with uncertainty No mechanism for this
  the system simply assembled one answer, assumed
  to be good enough to do the job.

50
R1/XCON

• Significance
• A rather simple forward-chaining rule-based
  expert system, which performed well, solved a
  difficult manufacturing problem, and proved to be
  enormously profitable.

51
R1/XCON

• Digital Equipment Corporation's problem was that
  they were marketing the best-selling Vax-11
  series of computers, and the department
  responsible for configuration was failing to keep
  up with customer demand.
• Each computer was the result of a consultation
  between a sales executive and the customer,
  designed to discover the customer's requirements,
  after which a configuration was drawn up, from
  which the system was built.
• Each configuration was taking 25 minutes, and
  orders were arriving at a rate of 10,000 a year.
• High error rate in the configurations was
  recorded.

52
R1/XCON

• DEC tried a conventional program to solve this
  problem, with no success, then asked McDermott to
  write an AI system.
• McDermott wrote R1/XCON.
• By 1986, it had processed 80,000 orders, and
  achieved 95-98 accuracy.
• It was reckoned to be saving DEC 25M a year.

53
R1/XCON

• However, R1/XCON suffered from the shortcomings
  of simple production-rule-based systems.
• When the nature of the task changed, fresh rules
  were simply added at the end of the rulebase.
• Soon, the rulebase was very large, unreliable and
  incomprehensible.
• Expensive rewriting was needed to restore the
  operation of the system.

54
OPTIMUM-AIV

• OPTIMUM-AIV is a planner used by the European
  Space Agency (1994) to help in the assembly,
  integration, and verification of spacecraft.
• It generates plans, and monitors their execution.

55
OPTIMUM-AIV

• it has a knowledgebase that describes the causal
  links that describe that in what particular order
  the assembly must be undertaken.
• Also, if a plan fails and has to be repaired, the
  system can make intelligent decisions about the
  alternative plans that will work and will not.

56
OPTIMUM-AIV

• It can engage in hierarchical planning - this
  involves producing a top-level plan with very
  little detail, and then turning this into
  increasingly more detailed lower-level plans.
• It can reason about complex conditions, time, and
  resources (such as budget constraints).