Nearly 60 years ago, British mathematician Alan Turing proposed conversation as a test for determining whether an artificial device can think. "If, during text-based conversation," Turing said, "a machine is indistinguishable from a human, then it could be said to be 'thinking' and, therefore, could be attributed with intelligence."

To eliminate any potential bias by humans toward computers, Turing said the idea of artificial conversation should be tested by having humans engage in a chat session on a computer terminal. At the other end of the chat would be either a human or a computer program; the test is for the human to judge whether he or she is talking to another human or an artificial device.

The threshold for artificial intelligence, Turing said, is whether a computer program can fool humans into thinking it is alive 30 percent of the time.

Almost There

On Sunday, a version of the Turing test called the Loebner Prize was held at Reading University in the United Kingdom under the direction of Professor Kevin Warwick. The prize is named after Hugh Loebner, president and CEO of the New Jersey-based Crown Industries, who has pledged $100,000 to the first person to write a program that passes the Turing test. In addition, Loebner awards $2,000 to the entry each year that is the most human-like (i.e., fools the most judges).

Warwick said people would be surprised at how close the machines are coming to passing the conversation test.

"What we found today," Warwick said, "is that if you have 10 judges, if we take 10 interrogators, then all of the machines have fooled some of those interrogators into thinking they are human. The best machines are falling at the 20 percent mark, so they're fooling two out of 10. What Alan Turing said all those years ago, in 1950, to pass the Turing test, they need to be hitting at the 30 percent mark. So the machines are definitely getting better."

This year's prize went to Elbot, a program written by German consultant Fred Roberts. The program fooled three out of 12 judges into thinking it was human.

Greater Challenges Await

As Loebner himself has acknowledged, the ability of a computer program to fool a handful of humans into thinking it, too, is human is merely the beginning. Additional levels of the Turing test contemplate longer time periods (giving humans more time to catch the computer program in structure or subject matter) or the processing of different kinds of information . For instance, to be truly intelligent, is it necessary for a computer to be able to understand and interpret audio or visual information?

Beyond any doubt, however, the latest results from the Loebner Prize demonstrate that the successful passing of the basic Turing test is not a question of if, but when.