How to answer a question: a simple system | DDI
Nick Sullivan stashed this in Software!
Fascinating look at how to deconstruct a question an answer it with AI. Perhaps someone should build a front end for google that implements this algorithm as a proxy to google.
You can't be sirius.
I think Siri is based largely on wolfram alpha. And yes, Cleverbot is interesting. But I wanna play around with this algorithm anyway. :)
The line is "Surely you can't be Sirius."
To which I reply, "Of course I'm Sirius. And don't call me Shirley."
The singularity is here when computers can't be distinguished from reddit users in conversation.
"Out of the 1,334 votes cast, Cleverbot was judged to be 59.3% human, compared to the rating of 63.3% human achieved by the actual humans. A score of 50% or higher is often considered to be a pass of the Turing Test."
To be indistinguishable from humans it needs to be ruder: http://pandawhale.com/convo/2485/how-can-you-tell-whether-youre-communicating-with-a-computer-or-a-human-barking-up-the-wrong-tree
Cleverbot takes quotes from prior interactions with humans and feeds them back. This is the reason it's so realistic. It fails when people circle back on those statements because the snippets will sometimes contradict each other; Cleverbot will say it's single, then married, then male, then female, etc.
Cleverbot doesn't realize it is contradicting itself, so there's no way to make it explode by exposing its contradictions to itself.
Which reminds me, Wikipedia's Does Not Compute article is a fun read:
The Simpsons writers often parody this phrase and its implications. In one episode, Homer said he wanted to make a robot repeat the words "It does not compute" until it exploded by giving it illogical commands. In the episode Trilogy of Error, Lisa's school project, a grammar-fixing robot, explodes after hearing too much bad grammar from the mobsters led by Fat Tony. The robot repeats the phrase "Bad grammar overload!" as a parody of the original phrase...