Klone Server

Dateline: 02/23/99

Chatterbots have been around for quite a long time. It all started with Eliza, Weizenbaum's famous 240-line program simulating an interview with a Rogerian psychotherapist. We saw more or less successful attempts to build a program or machine that understands plain English (among other languages) in the past 30 years. Some of the Loebner prize competitors showed really impressive behaviour, but they almost always shared the same approach to natural language understanding problem: word spotting, based on the recognition of key words or phrases. They don't understand the full meaning of a sentence, and although this system worked well for a few initial questions, it would become very boring after few hours of continuous use.

An usual, everyday conversation between two people is really a very complex process that cannot be simulated by simple, reactive, "Eliza-like" dialogs. Now there is a technology that could change the way people interact with computers. Big Science, Atlanta based company specialized in AI based solutions recently announced application server with a natural language understanding interface. Klone Server uses HTML and streaming media to create simulated, emotionally-intelligent, graphical characters. These characters, called Klones, converse with web users in plain English and function as knowledge workers.

"Klones answer the 20% of questions that generate 80% of call volume," says Tom Rearick, president of Big Science Company. Klones can answer simple questions from a wide range of subject domains. They can also solve in-depth problems in narrow domains by using an integral proprietary decision support system based on a case based reasoner - an important advantage over traditional rule-based systems. Klone Server also uses word spotting at "higher level", as a focus mechanism - not as a basis for recognition like Eliza and other simple chatterbots. But let's be fair: there is much more going on in the background that could not fit in 240 lines of code. Gathering of syntactic evidence at phrasal and sentential levels, word-sense disambiguation, context domains tracking, maintaining discourse state and application specific context... If you need more insight on technical details, I recommend reading whitepapers listed at the end of this article.

Another very important concept of the Klone server technology is known as Affective Computing. The bottom line here is that nobody wants to talk with chatterbot that doesn't demonstrate at least some emotional intelligence attributes. Emotion is a very important, although sometimes neglected, dimension of communication and prerequisite for building trust. Andrette, Big Science's first Klone, will not answer all of your questions correctly, but will show almost infinite patience and charm that can't be found in earlier chatterbots. After all, the most difficult questions has to be resolved by support personnel. You can only hope that they will be as patient with you as Andrette was...

Klone server is still young, immature product that could improve rapidly in capabilities and naturalism. Nevertheless, there are many applications for Klones right now. eCommerce (customer service agents, marketing and research assistants) comes first; they could also serve as tour guides, travel agents, scheduling assistants, robotic secretaries and web site receptionists. I don't see them replacing, but augmenting traditional services. As we all know, contacting support staff can be frustrating and time-consuming experience. It was much more fun to interact with Andrette than with Microsoft's support questionnaires, anyway. Hopefully, we'll soon see natural language understanding technology advancing as far as speech recognition, which is passing into the mainstream computer technology.

System requirements:

The Klone family of applications are written entirely in Java 1.1 so they are compatible with NT 4.0 and UNIX operating systems. Klone Server works with Microsoft�s Internet Information Server, Netscape�s Enterprise Server, and other web servers that support Java Servlets.

   Similar site(s) of interest: