Intro to Computation and AI

1
Intro to Computation and AI

• Dr. Jill Fain Lehman
• School of Computer Science
• Lecture 6 December 4, 1997

2
Review

• Last lecture looked at AI uses of basic graph
  structures and algorithms
• Inheritance hierarchies and inferential reasoning
• Constraint satisfaction and propagation
• In addition to relying on basic CS structures and
  algorithms, AI has also created its own
  formalisms e.g. rule-based systems.

3
Rule-based Systems

• Formalism critical in 1980s for moving AI out of
  the lab and into practical industry applications.
• Basis of Expert Systems Expert System Shells.
• Also basis of 2 long-standing cognitive
  architectures ACT and Soar.
• 2 parts knowledge structure and processing
  mechanism (of course).

4
Knowledge Structure

• Rules encode the knowledge of the domain required
  to solve problems or tasks in the domain.
• Conceptually, a rule is made up of 2 parts
• The if portion (aka left-hand side, conditions
  or preconditions)
• The then portion (aka right-hand side, actions,
  effects).
• There are many variations on this theme, e.g.
  including a confidence measure with a rule to
  help accommodate uncertainty.

5
Example MYCIN (Shortliffe)
An expert system for guiding bacterial infection
therapy.
Premise (and (same cntxt gram grampos)
(same contxt morph coccus)
(same cntxt conform
clumps)) Action (conclude cntxt ident
staphylococcus tally 0.7)
If (1) the stain of the organism is
gram-positive, and (2) the morphology of the
organism is coccus, and (3) the growth
conformation of the organism is clumps,l Then
there is suggestive evidence (0.7) that the
identity of the organism is staphylococcus.
6
Example R1 (McDermott)
If the most current active context is
distributing massbus devices, there is a
single-port drive that hasnt been assigned to a
massbus, there are no unassigned dual-port
disk drives, the number of devices that
each massbus should support is known, there
is a massbus that has been assigned at least 1
drive that should support additional disk
drives, the type of cable needed to connect
the disk drive to the previous device on
the massbus is known Then assign the disk drive
to the massbus.
7
Processing Mechanism

• Conceptually a rule interpreter that matches the
  if portion of the rule against a state
  description and then changes the state
  description based on the then portion.
• Exactly what goes in the state and what goes in
  the rule and what goes in the interpreter is the
  basis of the differences between particular rule
  systems.

8
Simple Example

• Suppose state had 2 objects blocka and blockb
  and blocka was on the table and blockb was on top
  of blocka.

State (isa block blocka) (isa block blockb)
(ontop blockb blocka) (ontable
blocka) Goal (ontable blockb) Operator-move-o
ff If (state (isa block ltxgt) (ontop ltxgt
ltygt))(goal (ontable ltxgt)) Then (add
(ontable ltxgt)) (remove (ontop ltxgt ltygt)) What
does the interpreter do???
9
Expressiveness of Rule Language

• Need to have variables that can be bound to
  particular instances during problem solving.
• Usually need logical operations (not
  disjunction) in addition to implicit
  conjunction).
• Need to encode directions to the interpreter
  (e.g. add remove).
• May need metalanguage (ability to talk about
  rules themselves) (this may also be hidden in the
  interpreter).

10
Behavior of Interpreter

• Enforce consistency of bindings across clauses.
• Must match conditions efficiently
• Large numbers of rules cannot be searched
  linearly need some form of indexing
• trade-off betw. expressiveness of language/ease
  of match
• Needs conflict resolution strategy when more than
  one rule can fire.
• E.g. fire all vs. fire exactly one (based on
  rules, objects, or actions?)
• May require additional capabilities to, e.g.,
  combine certainty factors.

11
Some Issues in Using Rule-based Systems

• Knowledge acquisition how to get the expertise
  out of experts, how to represent it efficiently,
  how to maintain consistency, how to check for
  completeness.
• Iterative approach, lots of time spent on tools.
• Explanation how to know what the system is
  doing, how to assure experts that reasons for
  behavior are appropriate.
• Alternatives to NL?
• Ethics