The Loebner Prize works very much like that 1950's television game show To Tell the Truth, in which celebrity panelists attempted to discover which of three people was the real fellow claiming to have done something extraordinary.

In the Loebner contest, judges are presented with a bevy of chat terminals to type into. Some of the terminals are manned by robots, some by people. The judges hold short conversations with the chatter, looking for unmistakable humanity on the other end. If they don't find it, they mark it down as a robot.

A program needs to fool half the judges to win the grand prize, but that has never been done. The robot that convinces the most judges is considered the winner -- for a prize of $2,000.

Last year, Richard Wallace's Alice robot won the contest, and it's the finalist this year, as well. It's competing against seven other robots.

During an interview earlier this year, Wallace said that he was pleased to have won the prize, but like many people, he has expressed dissatisfaction with the rules of the game.

According to Wallace, the judges in the original test devised by Turing weren't told that they were trying to determine whether their chat partner was a robot or not, so they "weren't looking for a robot," he said.

"But the Loebner contest has been called too aggressive, because the judges ask questions that they would never ask a human being during a conversation. They try to trick it -- like, they'll ask, 'What does the letter M look like upside down?'"

Because questions like that aren't likely to come up in an everyday conversation, chat robots like Alice don't usually know the answers, he explained.

But "the fact is, people act like robots," Wallace said. "The lesson from (my research) is that we are all like robots -- people don't use the full richness of language. Most people most of the time will not say anything that hasn't been said before."