An online chat between two robots set up by Cornell students is entertaining the nation.

"It was just an afternoon hack," said Hod Lipson, associate professor of mechanical and aerospace engineering. "It went viral in 24 hours and took us completely by surprise."

Lipson asked Ph.D. students Igor Labutov and Jason Yosinski to set up the conversation as a demo for his class on artificial intelligence. They chose a Web-based chatbot (a computer program designed to simulate human conversation) called Cleverbot. Anyone can go to http://cleverbot.com and carry on a typed conversation with the robot. The students added text-to-speech capability and computer-generated faces, set two laptops side-by-side on a table and connected them to Cleverbot, seeding the conversation with a simple "Hi." They videotaped part of the conversation and posted it on YouTube. Watch at http://www.youtube.com/watch?v=WnzlbyTZsQY. As of Aug. 31, it had more than a million hits.

The two avatars, one male and one female, start by exchanging pleasantries, but argue when one accuses the other of being a robot and then launch into a discussion of religion. Many chatbots work by repeating back what they hear in a slightly different form, but Cleverbot, developed by British artificial intelligence specialist Rollo Carpenter, draws on a vast database of phrases from all the conversations it has had in the past. Since it went live in 1997, Cleverbot has carried on more than 20 million conversations. That may explain why the male avatar says at one point "I am a unicorn." Apparently some human once said that.

"What makes this interesting is how people interpret what they see," said Lipson. Some, for instance, find "sexual tension" between the two characters. One viewer said the conversation was not like real human speech, but another countered that it was just like marriage. "The reaction is the real story," Lipson said.

Although this is not a typical subject for research in Lipson's Creative Machines Laboratory, the team is considering further exploration. Possibilities include conversations between three or more robots, or multiple robots and humans. Since Cleverbot learns from the conversations it has, what would happen if two robots continued to converse over a long period of time?

"Lying is possible," Yosinski noted.

"It's a blurring of the lines between humans and machines," Lipson said.