Easy Science Fair Experiment Testing AI
20Q Artificial Intelligence Project to Demonstrate Turing Test
Aug 13, 2009
An easy science fair project uses the 20Q handheld electronic game to demonstrate the effectiveness of the Turing Test. This project requires the participation of at least 10 volunteers, and one 20Q game.
AI Science Fair Project Background
20 Questions is a simple child's game that requires one person to ask up to 20 questions in order to guess what another person is thinking of. An artificial intelligence program was written to take simulate this task of deductive reasoning, and became wildly successful on the Internet. Radica Games, Ltd. purchased rights to the program, and began producing a small handheld electronic game that plays 20 Questions.
The Turing test was devised by Alan Turing to test for artificial intelligence. This test requires a judge to distinguish between the responses of a human and an artificial intelligence computer program. Using volunteers, and a 20Q game, this experiment can determine whether the 20Q game passes a simulation of the Turing test.
Artificial Intelligence Experiment Steps
The first step in this experiment to test 20Q for artificial intelligence is to decide on an object to be the 'Answer'. It will be necessary to use the same object for the entire experiment. Once the object has been chosen, 4 volunteers should be designated as Questioners, and 6 volunteers as Judges. Each of the following steps must be isolated from the volunteers, ensuring that neither the Questioners nor the Judges have any access to the information as it is gathered.
Play 20 Questions with the handheld 20Q game, typing up a transcript of each question and subsequent answer. Do not allow any of the volunteers to see any of the questions or answers. Play 20 Questions with each of the Questioner volunteers individually, in private, while typing a transcript of each question and answer. Do not allow any of the Questioners or Judges hear any of the questions or answers. Maintain a private record of which transcript went with which Questioner, and which went with the 20Q game.
Judging the 20Q Turing Test
Format each transcript identically, so that the only difference between the transcripts is the content of the questions and answers. Show the transcripts to the judges individually, and ask each judge to pick which transcript came from the 20Q game. According to the Turing test, if the judges are evenly divided, or the majority of the judges is wrong, the 20Q game qualifies as true Artificial Intelligence. If more judges can tell which transcript belongs to the 20Q game, it fails the Turing test.
 |
