Of course Eliza [1] is a famous AI bot, but I laughed quite a bit when I learned (on PAIP) that they wrote a program to simulate a paranoid schizophrenic (PARRY) and made then talk to each other [2].
I wonder how much of the human-responses are used to train the bot. It'd explain a lot of the hostility and randomness... I know I tend to make weird responses to bots, to see what kind of weird responses I can get back.
bot: what are the ingredients to that?
me: Tea. Earl gray. Hot.
As someone that enjoys crackmes, I found these posts fascinating - thank you very much :) It's interesting to see some reversing techniques for web services.
I wrote a Python script using a Cleverbot library that got it to talk to itself. The conversations actually weren't bad and you could sometimes see personalities form between the two. It makes me wonder whether the AI is designed so small differences early on between the two bots will eventually cause large divergences.
Also, it was funny to see them both call each other bots.
[1] http://en.wikipedia.org/wiki/ELIZA
[2] http://en.wikipedia.org/wiki/PARRY