"Everything should be made as simple as possible, but no simpler." [Albert Einstein]

Basic AI Hypothesis: [Newell and Simon, 1976]
The Physical Symbol System Hypothesis:
A physical symbol system has the necessary and sufficient means for general intelligent action.

Definition of AI: The study of techniques for solving exponentially hard problems in polynomial time by exploiting knowledge about the problem domain. [pg. 37 Rich, 1983] (Jackson:pg 15)

What does one expect from a machine to act intelligently? to be an expert?

---

One wants a system to "do it all".
Early attempts: (60's) find general methods to solve broad classes of problems

GPS:
General Problem Solver [Newell & Simon,1963]
Means-ends: understand what (end) needs to be done and use whatever available method to get there (goal driven) (states/operators)

Problems: (general methods) the more classes of problems a single program could handle, the more poorly it handled individual cases.
(more in Ch.2)

(note: OO Simula '65)

*methodology:

• write individual modules
• use union

Principles:

• explicit naming
• modularity
• graceful degradation
• least commitment
• exploit natural constraints

heuristics: rules of thumb
simplifications that effectively limit search for solutions

AI growth: from expert problem-solving to lower level. I.e., it was easier to do harder problems

misleading
• - perhaps weren't capturing expertnesseg. air traffic control

• - some problems lend themselves to certain strategies

***AI questions: important!

(representation)
• how should knowledge be
  • represented
  • organized
  (control)

• so that it can be
  • accessed
  • used

The power of an expert system derives from the relevant knowledge it possesses, not from the particular "generalized" formalisms and inference schemes it employs (page 28 and Chandra quote)

Note that there is knowledge in representation and knowledge in control

Book: "To make a program intelligent, provide it with lots of high-quality, specific knowledge about some problem area."

(conference: Feigenbaum/Chandra (like Shakespear bunch of words))

What specific knowledge? (wrt problem area?)

knowledge of:

• domain (e.g. medicine (application area))
• strategy (of the problem-solver)
• task (e.g. classification)

"The most important point about a representation is that it makes certain information explicit" [Marr]

Knowledge Engineering: determining the knowledge of procedures, strategies, rules of thumb (etc.) and organizing them for use

Levels of analysis -- Newell and Knowledge Level

knowledge level	what	(functional)
symbol level (program)	how	(structural (and processes))
computer systems	how	(hardware)

Characteristics of an Expert System: (Jackson: page 3)

• It simulates human reasoning about a problem domain (matter of opinion: psychological or physical (neural nets, genetic algorithms)

  compare to mathematical modeling

  model-based reasoning (naive physics)

• It performs reasoning over representations of human knowledge

  knowledge base & inference engine

• It solves problems by heuristic or approximate methods

  (make NP look like P) ...note approximate

How differ from general AI? (See Jackson:pg. 3)

Terms:

• Domain Expert
• Knowledge Engineer
• Expert-system-building Tool
• User

Overview and Focus:

• Background in AI
• Theoretical Foundations
  • Representation and Control
  • Automated Reasoning (logic)
  • Representing Uncertainty
• Knowledge Representations
• Practical Problem Solving
• Software Tools and Architectures

Research Topics in Expert Systems

• Knowledge Acquisition (bottleneck)(Jackson:pg 4)
• Knowledge Representation (Jackson:pg 6,7 terms)
• The application of knowledge
  • planning
  • control Jackson page 8 (metaknowledge)
• Explanation (consider architecture) (Jackson: page 9, types)

Good News/ Bad News (Waterman, page 12, 14)

Read Chapters 1 and 2 - very well written