Cleverbot Developed by Rollo Carpenter
Rollo Carpenter, a British AI developer, launched Cleverbot as a publicly accessible web application in 1997, extending his earlier Jabberwacky project. Cleverbot learned conversational responses directly from accumulated human inputs rather than from a hand-coded rule base, and went on to accumulate hundreds of millions of logged exchanges.
Background
Through the 1980s and into the 1990s, most chatbots worked from scripts. A developer would write a list of patterns, attach a reply to each one, and the system would match whatever the user typed against that list. ELIZA, built at MIT in the mid-1960s, was the template: clever pattern-matching that could sustain a short exchange but had nothing behind it. If you pushed past the patterns, the illusion broke immediately.
The fundamental problem was that someone had to write every rule by hand. That took time, and it imposed a ceiling. A system could only respond to situations its authors had anticipated. Conversations are not like that. People change subject, use slang, make jokes, trail off mid-sentence. No rule set written in advance could cover much of that territory.
Rollo Carpenter had been thinking about this differently since around 1988, when he began building Jabberwacky, a program that stored what people actually said to it and used those stored exchanges to generate replies. The idea was that a system exposed to enough real conversation would eventually get better at sounding like part of one. In the early years, though, it lived on Carpenter’s own machine and saw only a limited number of exchanges.
What happened
In 1997, Carpenter put Cleverbot on the web, an evolution of the Jabberwacky work built and published through his company, Icogno Ltd. Making it publicly accessible changed everything about the scale of data it could collect. Instead of the controlled trickle of a private experiment, the system was now talking to anyone who found it.
Cleverbot did not generate replies by following rules or by predicting the next word from a statistical model. It used instance-based learning: when someone typed something, the system searched its entire store of previous human inputs for the closest match and returned what a human had said in response to something similar before. Every new conversation added to that store. The system was, in a real sense, learning its responses directly from the people who talked to it.
Carpenter’s background, including a degree in physics from the University of Bristol, informed a practical and empirical approach to the problem. Rather than building elaborate theoretical machinery, he built something that could accumulate evidence and get better as it grew. Over time, Cleverbot logged hundreds of millions of exchanges, with the eventual total exceeding 300 million interactions. That corpus was the product, as much as the system itself.
The approach had obvious limits. Cleverbot could produce replies that were fluent-sounding but contextually strange, because it was retrieving from past exchanges rather than reasoning about the present one. It had no memory of what had been said earlier in the same conversation. But those limits did not stop people talking to it, and the sheer volume of data it gathered gave researchers and the broader public a concrete picture of what data-driven conversational behaviour looked like before neural language models existed to attempt the same thing another way.
Why it mattered
Cleverbot demonstrated that a conversational agent could improve through continuous exposure to real human dialogue at scale, without relying on scripted responses or manually constructed knowledge bases. Its public accessibility meant it gathered an unusually large corpus of natural conversational data, eventually exceeding 300 million interactions, providing a practical illustration of data-driven language behaviour long before large-scale neural language models became mainstream. The project also contributed to public and academic discussions about chatbot evaluation, including a notable 2011 Turing Test-style trial at the Techniche festival in which Cleverbot was rated as 'human' by 59.3 per cent of judges.
People
Organisations
Icogno Ltd, Cleverbot Com
Sources
- The Man Behind Cleverbot.Business Insider.Secondary
- Cleverbot, About.Icogno Ltd / Cleverbot.com.Official
Cite this page
AI Achievements. (1997). Cleverbot Developed by Rollo Carpenter. Retrieved 2026-08-22, from https://achievements.ai/milestone/cleverbot-chatbot-by-rollo
@misc{achievements_cleverbot_chatbot_by_rollo,
title = {Cleverbot Developed by Rollo Carpenter},
author = {{AI Achievements}},
year = {1997},
url = {https://achievements.ai/milestone/cleverbot-chatbot-by-rollo}
}