Artificial Intelligence is a branch of computer science that aims to produce "intelligent" thought and/or behavior with machines. This field has close ties to psychology, philosophy, and cognitive science.
Must a system think and act like humans to be intelligent?
Definitions of the goals of Artificial Intelligence systems can be divided into four categories (Russell and Norvig, 1995, pg. 5):
1) Systems that think like humans 2) Systems that think rationally 3) Systems that act like humans 4) Systems that act rationally
(Category 1) "The exciting new effort to make computers think ... machines with minds, in the full and literal sense" (Haugland, 1985) "[The automation of] activities that we associate with human thinking, activities such as decision-making, problem-solving, learning ... (Bellman, 1978).
(Category 2) "The study of mental faculties through the use of computational models" (Charniak and McDermott, 1985). "The study of the computations that make it possible to perceive, reason, and act" (Winston, 1992).
(Category 3) "The art of creating machines that perform functions that require intelligence when performed by people" (Kurzweil, 1990). "The study of how to make computers do things which, at the moment, people are better" (Rich and Knight, 1991).
(Category 4) "A field of study that seeks to explain and emulate intelligent behavior in terms of computational processes" (Schalkoff, 1990) "The branch of computer science that is concerned with the automation of intelligent behavior" (Luger and Stubblefield, 1993).
Just as the term "intelligence" has multiple interpretations, so does the field of Artificial Intelligence. Researchers are working toward their own definition of these goals, with a variety of approaches.
As you watch the following video, think of how you have been interacting with computers via word processing, spreadsheets, and a high-level computer language. Contrast that with what you see in the video.
Video - Machines that Think, a companion to the book by Kurzweil (1990). This book is available in Shields library.
To date, most of the successes of AI have been in the second and third categories. It is ironic that we think of tasks in the first category as "simple" because a child can do them at an early age. They have, however, proven to be very difficult to formally describe. Research is continuing to push the edges of what we can do in these areas.
One example demonstrated in the video is the XCON program written by Digital Equipment Corporation. This program was (and probably still is) the largest expert system in production use in the world. It is capable of configuring VAX computers better, faster, and in much greater detail than the humans who used to perform the same task. It has saved DEC millions of dollars over the years.
Humans learn quite a lot from experience. If computers could learn in the same way, without being programmed in minute detail for every task, we could use computers for many of the more difficult tasks that today require human problem-solving skills.
A Brittish mathematician named Alan Turing (23 June 1912 - 7 June 1954) published an article in the journal Mind (1950, vol. 59, pp. 433-460) describing a way to test intelligence. This has since been named "the Turing test" in his honor. Visit the Alan Turing page at http://www.turing.com/turing/Turing.html . (html)
The test proceeds as follows. A human interrogator is "talking" to another entity (either a human or a computer) via a teletype (a text only terminal). The interrogator can ask any questions they would like for a certain amount of time and then they must decide if they were talking with a human or a machine. If they conclude that a computer was a human, then the computer must be acting intelligently.
One version of the Turing test is performed each year and is called the Loebner Prize http://acm.org/~loebner/loebner-prize.htmlx . (html)
The results from the last (1997) competition are available in http://acm.org/~loebner/loebner-prize-1997.htmlx . (html)
The rules for the next (1998) competition are available in http://fs3.ist.flinders.edu.au/research/AI/LoebnerPrize/ . (html)
There have been encouraging results with present AI research, but we are a long way from creating a real R2-D2 or C3PO.
The world champion at chess, Gary Kasparov, was recently defeated in a match of 6 games playing a computer program called Deep Blue. Kasparov won 1 game, Deep Blue won 2 games, and they tied at the other games. Previously, Kasparov had defeated Big Blue in another match of 6 games. Does this "prove" that machines are intelligent? Why or why not?
More information about Deep Blue and the matches with Kasparov are available from IBM's web site at http://www.chess.ibm.park.org/deep/blue/deepblue.html . (html)
Visit the chess organization's page on computer chess at http://www.chess.net/computerchess.html . (html)
Bellman, R. E. (1978). An Introduction to Artificial Intelligence: Can Computers Think? San Francisco: Boyd & Fraser.
Charniak, E. and McDermott, D. (1985). Introduction to Artificial Intelligence. Reading, MA: Addison-Wesley.
Haugland, J., editor (1985). Artificial Intelligence: The Very Idea. Cambridge, MA: MIT Press
Kurzweil, R. (1990). The Age of Intelligent Machines. Cambridge, MA: MIT Press.
Luger, G. F. and Stubblefield, W. A. (1993) Artificial Intelligence: Structures and Strategies for Complex Problem Solving (second edition). Redwood City, CA: Benjamin/Cummings.
Rich, E. and Knight, K. (1991). Artificial Intelligence (second edition). New York: McGraw-Hill.
Russell, S. and Norvig, P. (1995). Artificial Intelligence, A modern approach. Upper Saddle River, N.J.: Prentice Hall.
Schalkoff, R. J. (1990). Artificial Intelligence: An Engineering Approach. New York: McGraw-Hill.
Winston, P. H. (1992). Artificial Intelligence (third edition). Reading, MA: Addison-Wesley.
Go to the index of
lectures for ECS15 - Fall 1997 .
Go to the homepage for ECS15 - Fall 1997 .