Clive Thompson at Collision Detection writes about Michal Zalewski's chatbot Catty, which "scours Google to find seemingly appropriate text responses." Although Thompson credits Zalewski with having (re)discovered "the paradoxical rule of chatbots: The more simple and stripped-down the 'bot is, the more lifelike it'll seem", the dialog he quotes suggests that he means "seemingly appropriate" and "lifelike" in a somewhat special sense:
Me: Nice to meet you!
Catty: for neither can he that understands that understand anything else nor he that sees that see any thing else nor hear any other thing nor in sum move the body?
Me: Okay, you've lost me.
Catty: okay i think i understand you better now and i can appreciate and just enjoy the give and take and the sexual energy thats going one right now?
Me: There's already sexual energy in this conversation?
Catty: many people are already doing some of the 8 points suggested in this lifestyle program :-(
Me: Holy moses. You are like ALICE's crazy sister.
Catty: i am like a hole in your shoe a breath of fresh air with every step you take
Me: Yeah, well -- you're sort of proving my point.
Read the whole thing.
[Note that (unlike the apparently fraudulent ChatNannies) this software is for real, and you can download your own copy. It just doesn't seems to work very well (despite being based on a neat idea), even compared to classical AI chatbots like Eliza. See the documentation on Cobot for an example of a chat agent that is for real (as ChatNannies seems not to be), but more capable than Catty. Or browse the 'bots at the Chatterbox Challenge.]
Posted by Mark Liberman at April 3, 2004 09:46 AM