Storobia Archive


Learning AI Chatbot

Jabberwacky is an AI chatbot developed by Rollo Carpenter, founder of Icogno Ltd. Unlike most other chatbots, Jabberwacky does not have a pre-built database of responses. Instead it learns from interactions with users. Everything it says has previously been said to it by someone else.

Jabberwacky has been around in one form or another for a long time. The initial program, Thoughts, began in 1989. The system went online in 1997 and has now recorded millions of interactions.

Rollo Carpenter is a strong believer in the importance of context: it is not enough to build up a database of input-reponse pairs. A response that is good in one context might be totally inappropriate in another. Jabberwacky uses context information when deciding which response to give.

Learning from user input has pros and cons. On the one hand Jabberwacky can cme out with surprisingly apt and incisive comments if the conversation goes down a route the bot has already learned. Say a clever line to Jabberwcky and sooner or later the bot will use that line on someone else.

The downside is that the bot lacks a distinct personality. It thinks it's a human, male and female, young and old, rich and poor. It's also worth noting that web users often teach Jabberwacky rather less than polite responses which it could later repeat back to anyone.

It is possible to use the Jabberwacky system to teach a new bot from scratch, so that it learns a particular personality. Unfortunately that obviously means that you can't leverage the power of the massive database already built up from millions of web chats.

Jabberwacky's main personalities are George and Joan. These won Loebner Prize medals in 2005 and 2006 respectively. In 2006 George appeared on BBC TV resulting in a surge of interest in Jabberwacky.

This is an old page archive from Storobia. Please read the site terms of use