BBC Home
Explore the BBC
BBC News
Launch consoleBBC NEWS CHANNEL
Last Updated: Tuesday, 20 September 2005, 10:02 GMT 11:02 UK
Brit's bot chats way to AI medal
Loebner Bronze Medallist Rollo Carpenter with Hugh Loebner
Mr Carpenter (r) won the competition with George, part of Jabberwacky
A British computer chat program, called George, has won an international prize for holding the most convincingly human-like conversation.

George and its creator Rollo Carpenter competed against three other talkative bot finalists in New York.

Reigning three-time winner, Alice, came fourth in this year's Loebner Prize.

The competition is based on the Turing Test, which suggests computers could be seen as "intelligent" if their chat was indistinguishable from humans.

George managed to convince the judges enough to earn himself the Hugh Loebner's Bronze Medal, and $3,000 (1,660), which goes to the most convincing entrant.

The gold medal and Grand Prize of $100,000 (55,400), which goes to the bot that completely fools the judges, has remained unclaimed since the competition's inception in 1990.

British first

Mr Carpenter told the BBC News website that the win was a first for an AI (Artificial Intelligence) that learns from its interactions.

"The time of the learning AI is now here. In a number of fields, technologies that observe and apply patterns in data to real-world situations have already come to the fore," explained Mr Carpenter.

"Though Jabberwacky, and the George character within it, remain distinctly unusual in their behaviour, the fact that they work at all, and that they've improved greatly over the last year, is a testament to the power of context."

In previous years, Mr Carpenter has entered the competition with the program, Jabberwacky. This year, George was the entered as a slightly different variant of that program.

George is a "character" which has learned its conversation skills from the interactions it has had with visitors to the Jabberwacky website, and through chats with Mr Carpenter.

Mr Carpenter thinks that in the not-too-distant future, it may be that programs or robots talk and act in place of humans, mimicking human behaviour.

Along with Jonathan Freeman, he has proposed a type of "Personal Turing Test" which would test a program's ability to convince a judge that it is a person known to them.

Image of the judging at the Loebner Prize
1: R Carpenter - "Jabberwacky"
2: V Veselov - "Eugene"
3: S Watkins - "Tony"
4: R Wallace - "Alice"
George, although a part of the Jabberwacky AI program, was intended to be an "individual" in a way that Jabberwacky on its own was not, said Mr Carpenter.

"Two things have made the difference this year: 2.5 million new entries [chat interactions/data inputs], taking the total to 5.7 million, and new techniques that extract a distinct personality - a character - from a tiny subset of the data - in this case, George," he explained.

"Anyone can create their own bot 'just by talking to it' at the site, preserving their own or an invented persona in silicon."

During the competition, the computer programs had to "chat" anonymously with four judges. The judges also chatted with four humans, each paired with a program. The judges then awarded "humanity" points to each of the pairs.

At one point in George's "conversation" with one of the judges, it accused the judge of not being human and discussed its lack of hobbies.

Alice had won the international competition for the most convincing entry in 2000, 2001 and 2004. Its creator, American Richard Wallace, started work on the software in 1995.

Alice chatbot wins for third time
20 Sep 04 |  Technology
German chatty bot is 'most human'
20 Oct 03 |  Technology
Chatbot bids to fool humans
22 Sep 03 |  Technology
AI helps gamers keep on playing
04 Sep 02 |  Technology
Hindi chatbot breaks new ground
27 Aug 02 |  Technology
Chatty computer wins again
15 Oct 01 |  Sci/Tech

The BBC is not responsible for the content of external internet sites


Americas Africa Europe Middle East South Asia Asia Pacific