-*- Mode:Text; Fonts:(CPTFONT CPTFONTB) -*-

;1;;CHAINING (FORWARD/BACKWARD)*

"Chaining" refers to the dynamic sequence of rules that become "active"
during the execution of a 1rule-based expert system*.  Rules become
active as the conditions they require become satisfied, or true.  

"Forward chaining" is sometimes called "data-driven".  A
forward-chaining expert system starts with an initial set of data, and
executes every rule that applies to the data to draw additional
conclusions ("knowledge").  If any of the executed rules provide
additional knowledge, more rules can then be activated, and so on, until
no further information can be derived (no more rules can be satisfied).

"Backward chaining" is sometimes called "goal-driven".  Given an initial
goal, in the form of a conclusion that the user wishes to prove or
derive, and (sometimes) an initial set of data, a backward-chaining
expert system tries to establish a sequence of rules that will bring the
system closer to the goal.

For example, given the following set of rules in a simple expert system:

  RULE-01 : IF AGE IS GREATER THAN 65 THEN <PERSON> IS 'RETIRED'

  RULE-02 : IF <PERSON> IS 'RETIRED'
              THEN <PERSON> IS 'ELIGIBLE-FOR-MEDICARE' 

If a FORWARD-CHAINING system starts with initial information that the
<PERSON> in question is over 65 years of age, this system would execute
RULE-01 and conclude that the <PERSON> is retired.  Subsequently,
RULE-02 would be checked and executed, and the system would conclude
that the <PERSON> is eligible for Medicare.

A BACKWARD-CHAINING system might be asked by the user whether a
particular person is eligible for Medicare.  The system would note that
RULE-02 is one way to "prove" that the person might be eligible.  If the
data does not include the fact that the person is retired, RULE-02
cannot be satisfied, but the system notes that RULE-01 might provide the
missing piece of information.  Now the system is waiting to satisfy both
rules. 1 *It might ask the user, "Is the person's age greater than 65?"
If the user says "yes", the system could then execute RULE-01 and then
RULE-02 (just as in forward-chaining), and satisfy its initial goal, to
prove whether the person is eligible for Medicare.

Note that some expert systems combine both techniques as appropriate.

;1;;DOMAIN*

The 1DOMAIN* of an 1expert system* is the field of knowledge or human
expertise that the system applies to.  For example, an expert system
used to advise an insurance adjustor might be said to "operate in the
domain of insurance claims knowledge".  An expert system can be written
in such a way that its domain can be expanded as needed: for example, a
medical expert system that "knows about" cancer may be enhanced with
knowledge about immune-deficiency diseases, thus expanding its domain. 
 
;1;;GARBAGE COLLECTION*

The computer programs used for Artificial Intelligence applications,
often written in languages such as LISP and Prolog, may have to operate
on very large amounts of data.  They may construct very large databases
of knowledge and control information.  Historically this has been a
problem that limited the size and scope of AI research and applications,
since these programs did not execute efficiently on conventional
computers with fixed or limited memory.

To overcome these limitations, many implementations of LISP (and some
Prologs) offer a feature known as "garbage collection", or GC.  This
feature provides a way to use large portions of memory dynamically
without worrying about reclaiming the storage for later use.  A garbage
collector usually runs on the computer in the background, looking for
memory that is no longer in use, and returning it to the central pool
where it can be used later for another purpose.

1;;;EXPERT SYSTEMS*

1. The class of computer programs which provide (or illustrate, or
emulate) human-like expertise or knowledge.  Some prefer the term
"advisory systems".

2. A 2rule-based* computer program, one1 *whose algorithm does not follow a
fixed programmed sequence, but rather a processing sequence determined
by the (approximately natural-language) rules input to the system.
These rules comprise its1 knowledge base*.

Not all expert systems are rule-based, and not all rule-based systems
are necessarily expert systems.

Related terms include: inference engine,1 *domain, chaining.

;1;;HEURISTICS*

;1;;INFERENCE ENGINE*

;1;;INFERENCE PARADIGM*

;1;;KNOWLEDGE REPRESENTATION*

;1;;KNOWLEDGE VALIDATION*

;1;;META-KNOWLEDGE*

;1;;REAL-TIME*

;1;;TRUTH MAINTENANCE*

;1;;STATIC PROBLEM