Expert Systems: series of rules that imply outcomes
- Human experts
  - Assumption: for many experts and many domains problems can be 
    represented as production rules.
- Knowledge Acquisition
  - Problem: Even if we can acquire the rules, they may change.
- Knowledge Maintenance
- Explanation of the problem solution
  - Explain to the user the solution, the reasons, and the knowledge that 
    was used.


Knowledge Acquisition
- Expertise is of a given form (production rules)
- Knowledge Elicitation
  - Give several experts a problem to solve and ask them how they solve it.
  - What are the step?
  - What knowledge made you apply these steps?
- Knowledge Comparison
- Knowledge Validation
  - Try it out in the world, see if it works

Expert Systems succeeded only partially because...
- Acquisition bottleneck
  - hard to elicit knowledge from experts
  - knowledge is tacit (not accessible by us)
    - especially control knowledge because it is unarticulated.
  - some knowledge is subjective
  - knowledge is of different forms
    - Neural Network: knowledge is distributed among many nodes/layers 
      (you can't point at one particular rule).
    - logical assertions (predicates)
    - semantic networks (objects and relations)
- Knowledge Maintenance
  - add one rule and it creates many conflicts with previous system.
- Explanation
  - in the form of a "trace" (list of rules that activated in the system 
    during query).  This provided too much detail.  Which leads to ...

Second theory of explanation of problem solving
- not a language of rules
- instead, "tasks and methods"
  - focus on goals/tasks with Domain Knowledge and Control Knowledge
  - Clancy McDermott

Now Expert Systems are any system that encapsulates knowledge from domain 
experts.

----------------------------------
Production System as a Cognitive Model
- SOAR \_ exhibit same power law
- ACT* / 

evaluation --> give examples
performance: AI thoery compared with Human subjects 
- problem, solution, trace

Power law = as the # of repetitions (n) of the same task increase, the 
time (t) it takes to complete the task decreases with the relationship:
log(t) ~ (log(n))^-2