Spent an hour or so trying to convince cleverbot that bees make a buzzing sound and at the end after asking again it replied they make a “meh” sound which is progress though certainly not the right answer.
Watching the responses cleverbot provided it became clear that cleverbot was not going to trust any single person’s responses, but instead will ask others who chat with cleverbot. This is why at times cleverbot can appear quite lucid and even come close to passing the Turing test before again succumbing to nonsensical one liners.
When on some line of discussion where people are likely to respond in the same way it will seem like talking to another person, and in a way that is what’s happening. Though when it’s relying on previous conversations it is easy to forge a new one Cleverbot can’t directly relate to. As a recent radio lab post discussed, although AI has come a long way in some ways, it still has a long way to go.