Download 4 Instructor presentation How can problem

Survey
yes no Was this document useful for you?
   Thank you for your participation!

* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project

Document related concepts

Collaborative information seeking wikipedia , lookup

Human–computer interaction wikipedia , lookup

Incomplete Nature wikipedia , lookup

Enactivism wikipedia , lookup

Personal knowledge base wikipedia , lookup

Existential risk from artificial general intelligence wikipedia , lookup

Multi-armed bandit wikipedia , lookup

Philosophy of artificial intelligence wikipedia , lookup

Hard problem of consciousness wikipedia , lookup

AI winter wikipedia , lookup

Ecological interface design wikipedia , lookup

Embodied cognitive science wikipedia , lookup

Computer Go wikipedia , lookup

History of artificial intelligence wikipedia , lookup

Knowledge representation and reasoning wikipedia , lookup

Transcript
1. 4 Instructor presentation
1. How can problem-solving be defined as search through a problem space?
2. What is a “search tree.”
3. What is “combinatorial explosion” and how does heuristic search help deal
with it?
4. Is chess-playing a well-structured or an ill-structured problem? Why?
2. Cawsey, Alison (1998), The Essence of Artificial Intelligence – Chap. 4 (pp. 68-95)
Chap. 4 Review Questions
5. How does breadth-first search work? How does depth-first search work?
6. When is breadth-first search advantageous? When is depth-first search
advantageous?
7. What is heuristic search? Is it exhaustive?
8. How could an evaluation function know that one node is “closer” to the
solution and assign it a better score?
9. How is the cost function in A* search different from the evaluation function in
Best First search?
10. In the water jug problem, the representation of the problem and the algorithms
used to solve it are programmed. The program apparently runs from start to finish
in the same way every time. Does the water jug program demonstrate artificial
intelligence or is it no different than any algorithmic computer program?
11. What kind of search is employed in the water jug problem?
12. The water jug program demonstrates no understanding of the process of
devising a solution. Does it solve the problem by trial and error? Would this
approach have limitations when applied to other problems?
13. How does planning make a complex problem more tractable?
14. How does solving the preconditions in the robot-fetch beer problem resemble
the backwards chaining of rules?
15. In the robot-fetch beer problem, how does the system know that the door must
be open before the robot can enter the kitchen?
2. Schaeffer (2001), “A Gamut of Games”
16. Why have game-playing programs had such success?
17. What is the alpha-beta search algorithm that Schaeffer credits for the success
of a number of game-playing programs?
18. Why is brute force search, which used to be avoided, now often a viable
option in game-playing programs?
19. How is the metric of “cost” measured in some game-playing programs and
how is it used?
20. How are Monte Carlo simulation and temporal-difference learning alike?
21. Do game-playing programs which model human game performance truly
demonstrate intelligence?
22. How has increased computer processing speed enhanced the performance of
game-playing programs?
23. Is the chess knowledge incorporated in Deep Blue useful for any other
purpose?
24. What is the primary reason why the best game-playing programs now
outperform the best human players?
25. Why have the game of Go playing programs been less successful than other
game-playing programs
3. Klein (1996): “Nonlinear Aspects of Problem Solving”
26. What are ill-structured problems?
27. Why doesn’t the model of search through a problem space work for illstructured problems?
28. According to Klein, ill-structured problems require continual redefinition of
the problem and constructing options from “leverage points.” Do you know how
this could be implemented in a knowledge representation?
4. Halfhill (1997): “Searching for Deep Blue”
29. What does the fact that chess grandmasters may consider only three moves per
second while Deep Blue can consider 100 million moves per second tell us about
our understanding of human problem solving in the game of chess?
30. Do you think that the technology of Deep Blue would be useful in solving
other problems? What kind of problems?
31. Do you think that increased computer speed alone will solve most of the
present shortcomings of AI?
1.5 Instructor presentation
1. Why are expert systems considered “strong” methods of problem solving?
2. Why can expert systems often perform as well as human experts?
3. Why is it important to limit the knowledge domain of an expert system?
4. Why can the same inference engine be used in different expert systems?
5. How does backward chaining work?
6. What is a certainty factor?
2. Cawsey, Alison (1998): The Essence of Artificial Intelligence – Chap. 3 (pp. 40-65)
Chap. 3 Review Questions
7. What does a knowledge engineer do? How is knowledge acquired from
experts?
8. Why are rules the preferred method of knowledge representation for expert
systems?
9. Why is the capability for reviewing how the expert system reaches conclusions
important? Why are rules ideal for this purpose?
10. What kinds of problems should expert systems be used for? What kinds of
problems should they probably not be used for?
11. What kind of problem would be best for a forward-chaining system? What
kind of problem would be best for a backward-chaining one?
12. Why are rule bases often considered shallow knowledge representations?
13. What does Bayes’ Theorem allow us to do?
14. What is the questionable assumption that must be made if conditional
probabilities are used? How does it affect the conclusion arrived at?
15. How are certainty factors different from probabilities? How are certainty
factors determined?
16. Why might the Internist system be said to use bidirectional chaining of rules?
17. Why would the belief networks used in Pathfinder be considered deeper
knowledge representations than rule bases?
18. Why do you suppose that expert system development was expensive in the
early years?
19. Why do you suppose that medical expert systems are not widely used for
diagnosis?
20. How would a general problem solver try to solve medical diagnosis problems
with means-end analysis?
3. Talebzadeh, et al. (1995): “Countrywide Loan Underwriting Expert System”
Instructor comment: Countrywide was caught up in the sub-prime loan scandal of
2008 and was convicted of making special loans to Democrat politicians. CLUES
apparently approved 100% of government insured loans.
21. Why were case-based reasoning and neural nets rejected for the knowledge
representation role for this expert system?
22. What difficulties occurred for the knowledge engineers because the
underwriters used heuristics instead of rules to make their loan decisions? How
did they acquire knowledge from the experts?
23. What would the knowledge engineers have done if the loan experts had
disagreed?
24. A universal problem for expert systems is user confidence and acceptance.
How did the CLUES developers overcome this problem?
25. In the validation phase of development, why is it important that the reasoning
used by the expert system be transparent and that it match the reasoning used by
the experts?
26. Could expert systems potentially do for bureaucratic jobs what robotics has
done for manufacturing?
Instructor comment: the CLUES developers could have validated the rules used
for loan decisions with regression analysis and saved some time. Working as
closely as they did with the underwriters, however, must have improved user
acceptance of the system.