Rollo Carpenter is a British artificial-intelligence developer best known for the conversational programs Jabberwacky and Cleverbot. His approach differed from the rule-writing tradition of ELIZA and A.L.I.C.E.: rather than scripting responses, his bots learn from their own conversations, remembering how humans replied in the past and reusing those replies, so the systems improve as more people talk to them.
Carpenter’s highest-profile result came at the Techniche 2011 festival in Guwahati, India, where Cleverbot took part in a public Turing-test exercise. Cleverbot’s own page about the event records that on September 3, 2011, the bot was rated 59.3 percent human, close to the 63.3 percent achieved by the real humans in the test, and quotes Carpenter reacting that the figure was “higher than even I was expecting, or even hoping for,” while acknowledging that “there is still a difference between human and machine.”
His work bridged two eras of conversational AI. Jabberwacky and Cleverbot were not the hand-coded bots of earlier decades, but neither were they the deep neural models that followed; they showed how much conversational fluency could come simply from accumulating and replaying human dialogue at scale.
For a general reader, Carpenter is a useful figure for understanding that “learning from data” arrived in chatbots well before modern machine learning made it the default, and that the question of whether such a bot could fool people was being tested in front of live audiences years before today’s assistants.