Format your assignment as a single PDF file. Due Tuesday, April 3 at 11:59 PM under "Problem Set 1" in Blackboard. LATE ASSIGNMENTS WILL NOT BE GRADED.

1. The Turing Test and the Loebner Prize. In the below, you're asked to take a position on a fundamental AI controversy: how should the goals of AI be defined, and how should progress be tracked? You'll be graded in roughly equal measure on a) whether it looks like you've read the relevant material, b) clarity of your text, and c) insight.
   Start by reading the following three links, then answer the questions below.
   The Loebner Prize
   Lessons from a Restricted Turing Test
   Loebner's Response
   1. 0.5 points. Provide a transcript of a short (5-6 statement) interaction with Rosette, winner of the 2011 Loebner prize.
   2. 0.5 points. At what point, if any, were you sure the bot was not human?
   3. 2 points. In roughly half a page of text, state whether you believe the Loebner prize is helpful for AI. Be sure to address multiple arguments from Shieber and Loebner in your answer.
   4. 2 points. If you had to invent a new annual AI competition, what would it be? Your competition should be designed to measure progress in AI, spur innovation, and generate excitement. In about one paragraph, clearly define the objectives, scoring criteria, and rules of the contest. In a second paragraph, argue why your contest would be preferable to the Loebner prize.

2. Search. In a fictional registrar's office, classrooms are assigned to courses based on multiple factors. Say we have a set of rooms R, and a set of classes C where each class is characterized by a meeting time interval and a set of acceptable classrooms (e.g. those that are large enough for the class enrollment, have required projection facilities, etc.). An assignment of classrooms to classes is valid if each class is assigned to an acceptable classroom, and no two classes meet at the same place at the same time.

   Because of the many factors involved, it's sometimes tough for the registrar's office to arrive at a valid assignment. You've taken an AI course for a week, so the university turns to you for guidance. Your task is to design a search algorithm that starts from the current class-to-room assignment (which is invalid) and finds a path to a valid assignment using the smallest number of changes from the current assignment as possible. The number of changes is equal to the number of classes that move to a different room (note, the meeting times cannot change).

   Answer the following questions:

   1. 1 point. Define the classroom assignment problem as a search problem. Specifically, in about one sentence each, precisely define the states, operators, start state, and goal test you will use.
   2. 1/2 point. Does this problem require an optimal search procedure? In a sentence, why or why not?
   3. 1/2 point. What is the maximum branching factor of your search tree?
   4. 1/2 point. Using iterative deepening with depth-limited search, what is the approximate time complexity of your algorithm in the worst case? (Use big-O notation here).
   5. 1/2 point. Will your search have problems with dupliicate states? If no, why not? If yes, give a brief example.
   6. 1/2 point. Define an admissible heuristic for your search problem, and state why the heuristic is admissible.
   7. 3/2 points. Give a concrete example problem with two rooms and two classes in which some "uninformed" search procedure returns the correct solution yet expands fewer states than A* search using your heuristic from above. State a convincing and clear argument why this is the case (it may help to draw the search tree).
      (Note: with just two rooms and two courses, your answer to this question may require that the "tie-breaking" behavior of A* on your example resolves in a particular way. That is acceptable.)