That's because it is essentially humans you are speaking to. It uses the technology behind Jabberwhacky and Jabberwhacky works using a sort of feedback loop. When you say something to it, it remembers what you say and then repeats it to someone else later, it then remembers the response and uses it next time someone asks the same question. It remembers the context using a tree. It's quite basic actually. if you say something which it does not have a response to yet, it stores it in it's database, says something completely random to you, and then repeats what you said to it to someone else later to try to get a response to add to it's database.
Try saying something like "I am human". It will argue with you and say you are a bot. That's because it said "I am human" to someone else, and that person argued with him.
Wait wait wait wait
That mean if I ask it some strange disturbing question, he will feed them back to a poor confused soul?
Human: Hello robot.
AI: Greetings. Did you ever taste human flesh?
Human: What ?
AI: I cannot get enough of it.
Human: That's disgusting
AI:Not as much as your mother
etc..
Edit:
Hah, actual conversation:
Me: How do you kill a zombie?
Cleverbot: You slice of his head or wack an axe into his forehead.
Me: Good answer, I'm impressed.
Cleverbot: You should be.
Me: Are you prepared for the zombie apocalypse?
Cleverbot:There's a test?
Me:No, only death
Cleverbot:Dead ones are good, they eat less.
Me:That's not true in case of zombies tinhead.
Cleverbot:Then why didn't you shave?
Huuuuh, ok...