Sun, Sep 24, 2006 page12

• Chatbot program scoops prize for being human-like

  • BY Bobbie Johnson / THE GUARDIAN
  • • 
    • 
    • 

  Her name is Joan; she is just a few years old and very talkative, and now she is officially the world's most human computer program.

  Last Sunday, the computer-generated character scooped the prestigious Loebner prize for artificial intelligence, seeing off some of the world's most hi-tech creations in the process.

  Joan's maker, the British entrepreneur Rollo Carpenter, emerged victorious after a day of intense competition as a quartet of software experts attempted to pass the famous "Turing test" and scoop a purse of US$100,000 in the process.

  

  A woman chats with the virtual software personality ``George'' on a computer screen in London on Sept. 12. George, who is 39 years old, single and light-hearted, is looking for friends on the internet.

  PHOTO: AFP

  The test began well before computers were capable of such complex interaction. In 1950 the British mathematician Alan Turing, who famously helped crack German military codes during the second world war, asked how scientists would determine whether computers had successfully achieved truly artificial intelligence. His suggestion was that a computer that could give conversational responses that were indistinguishable from a human's could be said to be thinking for itself.

  That challenge turned into an annual competition when New York philanthropist Hugh Loebner offered US$100,000 to the first program to successfully fool a panel of judges into thinking it was a real person.

  Despite 16 years of trying, no technology has made that breakthrough, and the grand prize remains unclaimed. However, Joan was deemed to be the most human competitor on show after a blind test held at University College in London. The judges held typed conversations -- similar to an internet chat program -- with all of the challengers, before they declared her the winner.

  That victory earned Carpenter a US$2,000 prize and a bronze medal -- and also helped him retain the title he first won last year with George, another program driven by Jabberwacky, the technology behind Joan.

  What's unusual about Jabberwacky is that it uses its experience of interaction with humans to create a store from which to draw its next response. The system has been entered for the Loebner Prize contest since 2003, and now seems to be building up an impressive lead over its rivals. That may be helped by the fact that it is always able to enlarge its database through conversations online: if you want to try it out, go to jabberwacky.com.

  "I'm very pleased," said Carpenter. "In a way this is very subjective test... but I believe that for practical purposes, people in working conditions will believe it's a human."

  Carpenter is now working with call center companies to help train his robots into understanding how to better interact with humans -- with the possibility of using them to replace human workers in the near future. But experts say that a program that will successfully pass the Turing test is still some years away.

  "We're still a long way off, but I think he's onto something," said Duncan Graham-Rowe, one of the judges.

  But Carpenter is used to the long haul. The very first version of the program was written in 1982.
• Most Popular

  • 1

    Taiwan ranked as the cheapest travel destination

  • 2

    Scott McTominay fires Napoli to top of Serie A

  • 3

    Suppliers to follow TSMC’s US plans but may be years away, Minister says

  • 4

    Fraud fugitive returns to Taiwan after six years on the run

  • 5

    Intel names new Taiwan head to boost partnerships