B/Q351, Spring 1997: Final Examination

Due Tuesday, 6 May, 23:59

Instructions: This test is worth 25% of your grade. Answer only three of the five questions. Be as specific as possible. You may use any resources you want other than help from other people. E-mail your answers to me before midnight Tuesday, 6 May. Please include how much time you think you spent on the exam (try not to spend more than two hours). If you have questions, I'd prefer you send email to the whole class so that everyone sees the question and my answer.
  1. Search
    For each of the following problems, say how it is a search problem: what the problem states would be like, how you would identify a goal state, how you would estimate the distance to a goal state (if this makes sense), and whether a genetic algorithm might be an appropriate way to handle it.
    1. Parsing a natural language sentence (assigning a syntactic and/or semantic structure to it)
    2. Composing a piece of music
    3. Diagnosing a person's medical condition

  2. Logic
    You are given the following facts:
    1. Fred is not a Wallace and Gromit fan.
    2. All computer science majors who have seen a Wallace and Gromit movie are Wallace and Gromit fans.
    3. Fred has seen a Wallace and Gromit movie.
    You would like to prove that Fred is not a computer science major. Describe how you would do this using resolution theorem proving: (a) Write the facts and the goal as predicate calculus sentences. (b) Change the sentences to clause form. (c) Show the resolution steps that lead you to a contradiction.

  3. AI and Cognitive Science
    For each of the following, say why it may not be a suitable model of human reasoning.
    1. Bayesian reasoning
    2. Object recognition by template matching
    3. Resolution theorem proving
    4. STRIPS-type planning

  4. The Whole Course
    For each of the following, say how it is realized, or how it fails to be realized, in at least two different kinds of AI systems.
    1. Short-term memory
    2. Inheritance
    3. Robustness (fault tolerance)
    4. Pattern completion (associative memory)

  5. The Whole Course
    What do you think is the hardest problem in AI? What makes it so hard? How do you think it will be solved (if at all)? In your answer, make reference to concepts you learned in the course; this should not be the sort of answer you would have written before you took the course (though you may have the same idea about what is hard).