• semantically rich domains (in contrast to cryptarithmetic problems)

• condition ===> action architecture (production systems)

• constant competition:

elaboration of knowledge

compression into more economical forms by theories

• knowledge creation + knowledge integration + knowledge dissemination

• question: what is a "domain?
• "poor":

- knowledge in cryptarithmetic problems: numbers, how to add and subtract, facts about parity

- puzzles (missionaries and cannibals)

• "rich":

- driving a taxi in a big city

- medicine

- law

- using UNIX, Micorsoft Word (high functionality applications)

• questions: what is

- chess

- programming

• Well-Defined Problems:

- the essential conditions of the problem are stated

- their solutions are the same for all problem solvers

- examples: school problems, mutilated checker board, implementing given algorithms

• Ill-Defined (or Wicked) Problems: problem solvers take an active role what the problem is

- fill gaps in the problem definition

- jump into the problem

- use information gained while trying to solve the problem

- examples: architects, engineers, lawyers, legislators, software designers, writers, teachers, ....

• processes -- access and retrieval:

- new chunk into LTM: 5-10 seconds

- to retrieve stored information: a few hundred milliseconds

• structure:

- nodes

- liberally cross-referenced

- elaborate index (recognition capability)

• LTM: a 2nd environment to the environment sensed (e.g., a physician's mental library of medical knowledge and the patient's body)

• claim: true expert behavior requires 50 000 chunks

• even the most talented people require approx. a decade to reach top professional proficiency

• if there is more information than can be acquired in a decade:

- specialization

- external sources of reference

• advantage: simplicity and uniformity of their structure

• productions of two kinds:

- test on the contents of short term memory (STM); e.g.: "If your goal is to enter the house, open the door" (goal driven)

- test on the contents of the world; e.g.: "If the door is locked, use your key" (data driven)

• UNDERSTAND (--> to construct internal representations; Simon and Hayes)

• ISAAC (--> to understand physics; Gordon Novak)

• Learning from Examples with adaptive production systems

• Discovery processes

AM (--> problem solving without a goal; Doug Lenat)

BACON (--> discovering invariants in bodies of numerical data)

9X + 17 = 6X + 23

3X + 17 = 23 (subtract 6X from both sides)

3X = 6 (subtract 17 from both sides)

X = 2

production system:

- if expression has the form <variable> = <real number> ---> halt

- if expression has variable term on right side ---> subtract variable term from both sides and simplify

- if expression has numerical term on left side ---> subtract numerical term from both sides and simplify

- if variable term has coefficient other than unity ---> divide both sides by coefficient

Information is in the World

D. Norman: "The Psychology of Everyday Things", 1988, the coin example, p 57

• parameters of the inner environment

limitations of short term memory

power of the visual system

• general control and search-guiding mechanisms (domain-independent)

depth-first and breath first searches

means-end analysis

generate and test

• learning and discovery mechanisms (permit the system to adapt with gradually increasing effectiveness to a particular environment)

50 000 chunks

learning from examples

discovery