Part II Methods of AI

1
Part IIMethods of AI

• Chapter 3
• Knowledge and Reasoning

2
Part II Methods of AI
Chapter 3 Knowledge Representation and Reasoning
3.1 Summary of Logic and Reasoning
3.2 Reasoning Deduction Systems
3.3 Rulebased Reasoning
3.4 Knowledge Representation General Issues
3.5 Knowledge Representation Semantic Nets
3.6 Knowledge Representation Description Logics
3
3.3 Reasoning
XPS Rule based Systems
4
(No Transcript)
5
Whats the idea ?
6
Why XPS?
XPS
multiply
the knowledge of human experts
preserve
7
Structure of an XPS
Expert
User
General (static) knowledge
Knowledge acquisition
Case-base (dynamical) knowledge
Explanation
Interviewer
Problem solving component
Intermediary resultssolution hypotheses
8
MYCIN in Consultation Mode

• System What is the patient's name?
• User George Washington
• System What is the patient's age?
• User 252 years
• System Is the patient male or female?
• User Male
• System What are the symptoms?
• User Headache dizziness upset stomach high fever
• System When did these symptoms first appear?
• User December 13,1799
• System I conclude that the patient has a viral
  infection , my recommendation is that he should
  take two aspirin, drink plenty of fluid, get lots
  of rest and call me in the morning.
• Note George Washington died the next day of a
  viral infection no antibiotics

9
TEIRESAS-MYCIN Explanation Mode (I)

• Mycin has been informed that a patient has
  contracted meningitis while
• in hospital. Doctors wish to interrogate Mycin to
  check this diagnosis.
• Mycin then asks the doctors this question
• Mycin Is Meningitis a hospital acquired illness?
• Doctor why?
• Mycin 2.0 this will aid in determining the
  orgamisms that might be causing the infection
• It has already established
• 2.1 the infection that requires therapy is
  meningitis
• 2.2 organisms were not seen on the stain of
  this pending csf culture
• 2.3 the type of infection is bacterial
• 2.4 neurosurgery is not one of the relevant
  items of patient 100
• 2.5 the age of patient 100 is greater than 25
• therefore if 2.6 the infection was acquired
  while the patient was in hospital then there is
  evidence that the organisms that might be causing
  the infection are E coli(0.75) staphylococcus-coag
  -pos(0.3) pseudomonas-aeruginosa(0.3)
  klebsiella-pneumoniae(0.5)

10
TEIRESIAS-MYCIN Explanation Mode (II)

• If the doctors wanted to know how something was
  concluded they might ask
• Doctor How 2.3
• The system then reponds with
• Mycin The following rules concluded about the
  likelihood that the type of the infection is
  bacterial
• 2.7 RULE148 (0.19)
• 2.8 RULE500 (0.51)
• 2.9 RULE501 (0.95)
• 3.0 RULE502 (0.97)
• 3.1 RULE526 (0.98)
• 3.2 RULE504 (0.97)
• 3.3 RULE524 (0.98)
• The system takes the doctors through the
  production system AND/OR
• tree to explain its conclusions. The numbers in
  brackets are called
• certainty factors because little can be
  absolutely certain. The certainty
• factors range from -1 to 1 where -1 means
  absolutely not and 1
• absolutely so and 0 undecided either way.

11
Toulmins Argumentation Scheme
therefore
Qualifier, inference result
Facts

if not
cause
Inference rule
Exception rule
because of
support
COGNITIVE SCIENCE
12
Toulmins argumentation scheme An example
therefore
The offeredused car is old
Probably, the offered used car is cheap

if not
cause
Used cars are cheapmost of the time
The offered used caris a collectors item
because of
Used things loose their valuewhen time goes by,
because they break down more often etc.
13
Basic principles of XPS (1)
Every production rule has two parts
A
B
Assumption Antecedence Evidence If-Part Left hand
side (LHS) Condition
Conclusion Consequence Hypotheses Then-Part Right
hand side (RHS) Action
Productions are evaluated over a (data) pool (not
data base!) that is named working memory (WM)
or at applications in cognitive psychology
denoted as short-term memory (STM)
14
Basic principles of XPS (2)
There are two modes of evaluating production
rules
Backward chaining
Forward chaining
A
B
A
B
Data controlled inference Antecedence-oriented
inference Bottom-up inference If-added
methods LHS-controlled chaining
Goal controlled inference Consequence-oriented
inference Top-down inference If-needed
methods RHS-controlled chaining
15
Production Rule Systems
Facts
(( car_no DÜW-AW 205) motor_status
on) oil_control on) air_pressure 0,1 bar) ...)
Rules
(1) IF (motor_status on) AND
(oil_control on) THEN WRITE(Stop
motor) AND SET (motor_status off)
(2) IF (car_no x) AND (air_pressure y)
AND (LESS y 1.5) THEN WRITE( x
has a flat tire)
16
General Structure of Production Rules
Simple rules
IF B1 ? B2 ? ? Bn THEN DO A1 ? A2 ? ?
Am ELSEDO C1
Example
IF the site of the culture is throat AND the
organism is streptococcus THEN there is strong
evidence that the subtype is not of group-D
17
(No Transcript)
18
Architecture of a Production System
data base
rules
C1 C2 ? A1 C3 ? A2 C1 C3 ? A3 C4 ? A4 C5 ?
A5
C5 C1 C3
Rule interpreter
recognition
action
conflict set
match
productionrules againstdata base
C3 ? A2 C1 C3 ? A3 C5 ? A5
C3 ? A2
evaluate A2
19
Recall first intro lecture!
Evaluation
positive

• Medical Competence

negative

• Knowledge Engineering

• No Deep Modelling

• No Self-assessment

20
Examples taken from Mycin (1)
Rules encoding relationship between identity of
infecting organisms and possible therapies
21
Examples taken from Mycin (2)
Rules encoding exclusion relationship between
therapies and other diagnoses
22
Examples taken from Mycin (3)
Rules to establish the possible identity of the
infecting organism
23
Examples taken from Mycin (4)
Rules supporting established possibilities
24
Examples taken from Mycin (5)
Rules to establish the alternative possible
identities of the infecting organism
25
Examples taken from Mycin (6)
Rules describing actions to take and to select
the best therapy
26
Translation of the Rules

• Example

IF the stain of the organism is GRAMNEG AND the
morphology of the organism is ROD AND the
aerobicity of the organism is AEROBIC THENDO there
is strongly suggestive evidence (0.8) that the
class of the organism is enterobacteriaceae
Translated into
(AND (Same CNTXT STAIN GRAMNEG) (Same CNTXT
MORPH ROD) (Same CNTXT AIR AEROBIC))
Premise
Action
( Conclude CNTXT class ENTEROBACT. TALLY 0.8)
27
Rules with Certainty Factors

• General Form

IF C1(w1) ? C2(w2) ? ? Cn(wn) THEN
DO A(W)
Example
IF the organism is gram-pos AND the organism
grows in chains AND the morphology is
spherical THEN by 70 evidence the organism is
streptococcus
28
Calculation of the Total Evidence
General Formula Evidence Preconditions _at_
Evidence of the Rule Total Evidence

• Example (Streptococcus) with MIN-Calculation
• Evidence of Rule 70
• GRAM-POS 100
• CHAINS 60
• SPHERE 70

min 60
Total Evidence 6070 42
29
Calculation of the Total Evidence
General Formula Evidence Preconditions _at_
Evidence of the Rule Total Evidence

• Example (Streptococcus) with MULT-Calculation
• Evidence of Rule 70
• GRAM-POS 100
• CHAINS 60
• SPHERE 70

Mult 1007060 42
Total Evidence 4270 29,4
30
Calculation of the Total Evidence

Evidence Preconditions _at_ Evidence of the Rule
Total Evidence

• Evidence Amplification
• given W1 the evidence of rule 1
• W2 the evidence of rule 2
• then calculate W as the new total
  evidence with

Example W1 42 as before W2 50 for
STREPTOCOCCUS
W 0.42 (1 - 0.42)0.5 71
31
Rules with Certainty Factors

• Positive and negative evidence
• separate calculations
• EVGes EVPos EVNeg

• Context-dependent Calculation of Total Evidence

Problem MYCIN-mechanism often too uniform
Solution Evidence Classes
given W1, W2 as before
W FR (W1 W2)
where FR is a function that is different for
different classes of rules
32
CONTEXT Classification of Rules in MYCIN

• CUL rules rules for some culture
• CURCUL rules rules for current culture
• CURORG rules rules for the current organism
• DRG rules rules for drugs
• OP rules rules for operations
• PAT rules rules concerning patients
• PDRG rules rules for already administered drugs
• PROGR rules rules for previously isolated
  organisms
• ORG rules rules for every organism

33
Rules with Certainty Factors

• Final Remark
• Evidence Theory, Probability Theory and
  Certainty Factors have become an independent
  (AI-) research field on its own, and all the
  previous calculations are just ad hoc solutions
  that worked empirically in the past .
• See Special Lecture on Uncertainty and
  Probability

34
Structured Rules (mapping relations)

Condition
Action
Default
Context
Example (causal relations)
IF (COND serious diarrhea AND longer than two
days) (CTXT malabsorprion) (DFLT no bicarbonat
therapy) THEN medium metabolic acidosis with a
normal anion ELSE light metabolic acidosis with
normal anions
35
Knowledge Sources in MED-II

• Focus of attention
• Dynamic chaining

36
Processing the Rules Processing the Conditions
Part

• Pattern matching
• e.g., (John takes ?drug)
• Unification
• (bilateral matching as in resolution principle)
• Association
• e.g., (F12 W1) ? (Z19 24) where F12 and W1 have
  some intuitive meaning (e.g. serious
  diarrhea)
• Predicates
• e.g., (before T1 T2) or
• e.g., some LISP-function triggering further
  rules

37
Processing the Rules A Rule Interpreter

• Rule Interpreter the abstract machine

DATA ? DATA BASIS UNTIL data fulfils termination
DO BEGIN Choose rule R applicable on
DATA DATA ? R (DATA) END
Problem Choose
Control Structure !
38
An empirical Solution Processing the Rules in
MYCIN
39
More ad hocery and Engineering The
Findout Mechanism in MYCIN
40
A More General Solution

• General Procedure
• Process the rules
• Add item to working memory
• Remove item from working memory

Tasks

• Matching ? naive match (unification)
  problem unifications in one cycle
• Conflict resolution
• Act (e.g. print, add, delete etc.)

41
The RETE Algorithm
SOAR
OPS 5
ACT-R
rule database
working memory(WM)
r elements
w elements
p1 ? p2 ? ? pn ? act1 ? act2 ...
ACTIONS add, delete elements from WM, print etc.
42
Processing Steps

• Compile rules into net (inclusive constraints)
• Insert working memory (WM) entries
• Perform actions and modify net
• Shared representation of same LHS (elimination of
  rules)
• Example delete(A) ? bigger impact (propagated
  through net)

43
Example
D
AD
add E
E(2)
A
B
AB
C
add D
C(5)
B(2)
D(2)
A(1)
B(3)
A(2)
E
delete A
B(4)
? B(x)
? C(y)
? add D(x)
A(x)
? B(y)
? D(x)
? add E(x)
A(x)
? B(x)
? E(y)
? delete A(x)
A(x)
44
Examples for resolution strategies
Conflict Resolution for competing rules

• No duplication
• Prefer rules referring to recent WM elements
• Prefer rules that are more specific
• Prefer actions with higher priority

A General Programming Language for Rule based
Systems
OPS-5
45
OPS-5 Basic Syntax and its Elements
1) Structure of the Database
2) Structure of the Production Rules
46
Monkey World XPS

• (p HoldsObject-Ceiling
• (goal status active
• type holds
• objid ltO1gt)
• (physical-object id ltO1gt
• weight light
• at ltpgt
• on ceiling)
• (physical-object id ladder
• at ltpgt
• on floor)
• (monkey on ladder holds NIL)
• -(physical-object on ltO1gt)
• --gt
• (write (crlf) Grab ltO1gt (crlf))
• (modify 2 on NIL)
• (modify 4 holds ltO1gt)
• (modify 1 status satisfied) )

47
FERPLAN Production Rule
FERPLAN XPS Assembly Planing for workpiece
productions (Fertigungsplanung)
48
More Conditions
49
Actions
50
The Working Memory of OPS5

• The Working Menory (Database) consists of pairs
• (lttime-taggt, ltwm-elementgt)
• lttime-taggt
• Creation or last modification time of the
  element
• The larger the time-tag value, the more recent
  the element is
• Can be used for conflict resolution
• ltwm-elementgt as before.

51
Control Strategies
52
More Control strategies

• Exhaustive Forward Search
• Exhaustive Backward Search
• Generate-and-Test
• Partitioning into Phases
• Establish-Refine
• Hypothesize-and-Test
• Constraint Propagation
• Opportunistic Scheduling (via an Agenda)
• Knowledge Levels

53
Establish-Refine Strategy
54
(No Transcript)
55
(No Transcript)
56
(No Transcript)
57
Summary XPS

• Power of XPSs results from great deal of
  domain-specific knowledge not from a single
  powerful technique
• The required knowledge is about a particular area
  and well defined
• Usually build with aid from experts, who are
  willing to spend effort in transferring their
  expertise
• Knowledge transfer is incremental
• Amount of knowledge depends on task (40 to
  thousands)
• Control structure depends system characteristics
• Non-domain specific parts can be extracted and
  reused to build XPS for other domains

58
Major Problems in XPS

• Brittleness
• XPS have no general knowledge to fall back on
• Lack of Meta-Knowledge
• XPS cannot reason about their own behaviour
• Knowledge Acquisition
• Despite tools to assist knowledge acquisition,
  this is the major bottleneck to apply XPS
  techniques in other domains.
• Validation
• Mesuring of system performance difficult no
  means to quantify use of knowledge correctness
  proofs are impossible