Artificial intelligence is a tricky field. Rosette has her flaws, but so does every other chatbot.
As for how some of them act...random, a lot of chatbots are supposed to be "learning chatbots." When they talk to humans, they memorize how humans react to what they say. Later, they'll copy those reactions if in a similar situation. Cleverbot's an example of a learning chatbot, as is Jabberwacky. The problem with learning chatbots is that...well, they're going to copy a bunch of strangers on the Internet.