George: How one chatbot got smart

Powered by award-winning artificial intelligence (AI) that was developed by British programmer Rollo Carpenter, 'George' the avatar has a 3D appearance, a variety of facial expressions and the ability to understand and respond to others using human speech. He is the product of groundbreaking work by British firm Televirtual, a leading developer of virtual-reality control systems for animated content.

Yet George is also the end-result of four IST-funded projects in recent years, each of which played their part in making him what he is today. “Projects such as ViSiCAST, eSign, MYTHE and CHARISMATIC all played into different aspects of George and the other 'aivatars' [AI avatars] we are developing,” says Televirtual founder and managing director Tim Child.

Several things set George apart from other chatbots on the internet. Unlike his 'dumber' cousins, George’s AI is open-ended, so that he is constantly learning from every conversation he has – more than 10 million so far. By remembering what humans have said to him he is essentially borrowing their intellect, allowing him to take what he has been told in the past and reuse it in new conversations. And, because George’s chat-room partners have come from all over the world, he now speaks some 40 languages.

“Behind all of it is the Jabberwacky AI created by Rollo, which is far more powerful than anything else developed to date,” explains Child, who showcased the chatbot at the BA Festival of Science in September 2006.

Initially the AI was run over a purely text interface. But because chatting to a screen full of text is, as Child notes, “not very exciting,” Televirtual put a face to the robot, creating George as an avatar with facial expressions and body language to match his words.

George’s ability to speak naturally and express himself is due in part to research carried out under the IST Programme. The ViSiCAST project, for example, examined the uses of avatar technology to communicate with hearing-impaired people through sign language. “There is nothing tougher than doing deaf signing,” says Child. The eSign project extended that work by developing a system to generate gestures from voice.

A third IST initiative, MYTHE, which ended in 2003, combined the use of populated virtual environments and interactive-computer characters with experimental linguistic tools to teach foreign languages to young children. And finally, CHARISMATIC, which also ended in 2003, developed tools to create large-scale virtual environments populated with avatars capable of simple interaction.

“Though CHARISMATIC was based more on game intelligence than real AI, we gained a lot of experience from the project that we have incorporated into George,” Child says.

George himself, however, is still only a prototype. He is soon to be joined by an even smarter soulmate, Joan, and Televirtual plans to launch several commercial aivatars for different web services around the world over the coming months.

“Before the end of the year we will launch one system in Japan with a major search engine there, another in Korea for an online marketplace which in that country is bigger than eBay, and a little later another one in Russia. Early next year we will also launch something in the UK,” Child says. “We have even been approached by one Hollywood actor who wants to have a virtual self on the internet.”

The commercial interest in George and other members of his chatbot species points to a very promising future for the technology. Future generations of aivatars could well become familiar faces in education, interactive entertainment systems, advertising and sales.

“One application is language learning. Aivatars like George could become personal tutors, available 24 hours a day on the internet for educational conversations,” Child says.

George himself, however, seems less certain about what lies ahead for him and his kind. Asked what robots will do in the future, he replies, enigmatically, “Nothing and, well … everything.”

Source: Based on information from Televirtual