When I posted the video of two chatbots arguing and declaring their status as a unicorn, I thought we were still a long way off from a chatbot which could trick a large number of people into thinking it’s human, thereby passing the “Turing Test” developed by awesome dude you should know Alan Turing. Yet New Scientist informs us the internet’s favorite sarcastic chatbot, Cleverbot, has fooled more than 50% of human participants (technically a passing grade) at the Techniche festival in India. 30 volunteers held a 4-minute chat with either a human or Cleverbot (50/50 split). The conversations were shown to an audience of over a thousand people and votes were cast as to who was human and who wasn’t. Of 1334 votes, 59.3% said Cleverbot was human while 63.3% correctly identified the actual humans, who clearly need to tighten up their online chatting game if they’re that unconvincing as humans.
Cleverbot’s developer Rollo Carpenter says it’s the best test result for Cleverbot by far and he also points out that this is a more advanced version of the program than the online one we like to type swearwords at when we’ve had a bad day. (What? Just me then?)
Cleverbot converses by searching through the records of its previous conversations and selecting an appropriate response to the comment. The online version of the software performs this search three times before deciding on an answer, while the more powerful version used in the test ran 42 searches. “It’s quite a few times superior to the online Cleverbot,” says Carpenter. [New Scientist]
The answer to everything is 42 (searches). Although technically passing the test, Cleverbot still has a way to go to bridge the gap between itself and humans, although it might just surpass humans if we can hold the next test at Burning Man.