COMP 590-096 Fall 2010

Assignment 1

Due date: September 14, 3PM

Instructions

Please answer the following questions. Feel free to consult the lecture slides, textbook, and external sources. Briefly justify each of your answers and give references (web links) where appropriate. These questions are meant to be thought-provoking. For most of them, I am not looking for a single "right" answer, but for a well-justified answer that shows critical thought.

Submission: Write up your answers in a file in PDF or Word Document format. The name of your file should be lastname_firstname_assignment1.pdf or lastname_firstname_assignment1.doc. Please be sure to also put your name at the top of your document. Submit your file through Blackboard as follows:

  1. Log in to Blackboard and access COMP590.096.
  2. Select Tools | Digital Dropbox from the navigation menu on the left.
  3. Use Send file to submit an assignment for grading. Add file merely stores a file in your personal dropbox.
  4. In Digital Dropbox, the file should be listed as Submitted, and not just Posted.
Please contact the course TA, Danny Kumar (ndkumar -at- cs.unc.edu), if you have any trouble with Blackboard submission.

Late policy: You lose 25% of the points for every day the assignment is late. If you have a compelling reason for not being able to submit the assignment on time and would like to make a special arrangement, you must let me know ahead of the due date.

Question 1. Do a Web search to figure out which of the following tasks can be successfully solved by today's AI systems. For each item, answer "yes," "no," or "it depends." Support your answers with relevant links and/or your reasoning.

  1. Play a decent game of ping-pong (table tennis).
  2. Safely navigate a car in the desert.
  3. Safely navigate a car down Franklin street.
  4. Buy a week's worth of groceries on the Web.
  5. Buy a week's worth of groceries at Harris Teeter.
  6. Discover and prove a new mathematical theorem.
  7. Play champion-level Go.
  8. Solve crossword puzzles better than most people.
  9. Translate spoken English into spoken Swedish in real time.
  10. Converse successfully with another person for an hour.
  11. Perform a complex surgical operation.
  12. Unload a dishwasher and put away all the dishes.
  13. Give competent legal advice in a specialized area of law.
  14. Read a musical score from a sheet of paper and play it back.
  15. Write a funny story.

Question 2. The following comparison of computational resources is taken from Figure 1.3 in the third edition of the textbook:

  IBM Blue Gene circa 2008 Human brain
Computational units 104 CPUs, 1012 transistors 1011 neurons
Storage units 1014 bits RAM 1011 neurons
  1015 bits disk 1014 synapses
Cycle time 10-9 seconds 10-3 seconds
Operations per second 1015 1017
Memory updates per second 1014 1014

Using this information as a starting point, analyze the relative computational power of today's supercomputers vs. the brain. Which one, in your opinion, has the advantage? Is it fair to compare the two on the basis of the above information alone (e.g., might there be important differences in terms of computing architecture that are not reflected in the table)? How much role do computational resources play in the ability of computers to achieve human-level AI?

Question 3. Answer each of the following questions and briefly justify your answers:

  1. Are reflex actions (such as flinching from a hot stove) rational? Can a simple reflex agent be rational?
  2. Is it possible to behave rationally in an unobservable environment?
  3. What is the relationship between learning and rationality?
  4. Would evolution tend to result in systems that act rationally? What goals would such systems be designed to achieve?
  5. Are human beings rational agents (in the sense of maximizing expected utility)? If not, does it mean that the "rational agent" framework cannot be the right one for AI?

Question 4. Attend next week's guest lectures by Jur van den Berg and Ron Alterovitz. Select one particular task or application mentioned in one of their talks, write down a PEAS specification for that task (try to be as specific as you can about the performance measure), and classify the task environment according to each of the seven categories listed in the third lecture and in Section 2.3 of the book. What, in your opinion, are the biggest challenges in implementing an agent to cope with this task environment?