Sample Questions for AI Exam 1

What is the Turing Test, and what is it supposed to show?

What is a state space?

What are the ingredients of a state space search?

How does the "describe and match" PSM work?

How could you use features to as part of a describe and match method?

Describe the "generate and test" PSM and explain when it should be
used. 

What is the key idea on which Means Ends Analysis is based?
(i.e., what do you have to know to do operator selection?)

Explain why Means Ends Analysis can be considered as a kind
of Problem Reduction method. 

Give a simple example of problem reduction in the "blocks world".

Contrast a blind/weak search with a strong/informed one.
Briefly describe an example of each type.  

Does a depth first search always find the optimum path if one exists?
Explain.

How can Hill Climbing fail or waste time?

What is a heuristic?  Give an example.

What are the components of the heuristic function that guides the A*
search?

Is A* closest to Depth First search or Best First search and why?

If heuristics are used to prune a search tree what is sacrificed?

Describe how a configuration space and A* can be used to find a
robot's path.

In the following game tree, 

       A
      / \
     /   \
    B     B
   / \   / \
  2  9  1  3

with A to move (maximizing), what would A select as its move if it was
doing a minimax procedure?

What would change if it was using an Alpha-Beta procedure?

What might the minimax result change if it extended the same tree to a
3 ply search?

Could you use the singular-extension heuristic in this situation?
Explain. What "effect" is it trying to combat?


Explain how backward chaining works using rules.
When do you need to use backtracking in backward chaining?

How can a rule-based system provide 'how' and 'why' explanations?

Explain the add-delete syntax for rules. 

Motivate the use of "specificity" and of "data recency" (i.e., using
recently added/changed WM elements) as conflict resolution strategies
for a forward chaining rule-based system.

Why does the RETE approach improve the performance of rule-based
systems? 


Consider the following frame system:

Name  DOG
Legs  Default 4
Owner If-Value-Added  Demon-check-if-kind
Coat  If-Value-Needed Demon-deduce-from-breed
Noise Value "bark"
Alive Value "yes"

Name  FIDO
ISA   DOG
Noise Value "growl"
Alive Default "only-just"

Do the following actions, assuming inheritance, and that PUT can
create new slots. Report what happens and why.

PUT FIDO Breed "corgi"
GET FIDO Coat
GET FIDO Legs
PUT FIDO Owner "ken"
GET FIDO Alive



Show a thematic role frame for the verb "to eat" and include at
least three roles as slots, filled by words from an example sentence.

In general terms describe how thematic role frames and roles act as
constraints when doing sentence analysis.  

Give one advantage of moving to primitive actions rather than using
verbs.

Using the Min-Conflicts heuristic explain the first 2 steps of solving
the 4-Queens problem, starting with this configuration:

   - - - -
   - Q - -
   Q - - Q
   - - Q -


What are the 4 basic junction/vertex labels in the blocksworld?
What types of basic line labels are there and what do they mean?

Draw a cube and label it correctly with line labels.

Briefly describe Waltz's procedure for symbolic constraint satisfaction.

Give a simple example of solving the CSP given below using Arc
Consistency:
  X <--not equal--> Y <--not equal--> Z
  X={blue, green}, Y={red, blue}, Z={red}

For a 3-color (red, green, blue), 3-country (A, B, C) map-coloring
problem (below):
i)  show the corresponding graph of constraints linking variables,
    making it clear what the constraints and variables are;
ii) show how a depth first search with backtracking would provide a
    solution to the constraint satisfaction problem.

  -----------
 |A          |
 |           |
  -----------
 |B    |C    |
 |     |     |
  -----------


Give an example of a rule of inference that might be used in a proof
procedure in logic. 

Explain how a proof by refutation works.

Explain which of {Deduction, Induction, Abduction} are likely to be
truth-preserving and why.

Using resolution, what resolvent could be produced by these clauses:

        ~Feathers(x) v Bird(x)
         Feathers(Tweety) v ~Colorful(Tweety)

Why do you need proof strategies for resolution proofs? 
Explain a heuristic that might help. 

What does "Traditional Logic Is Monotonic" mean?


What is Backtracking, and explain the two types.
Explain how one type attempts to overcome a weakness of the other.

How does a Truth-propagation net allow incorrect assumptions to be
detected?