Slashdot Mirror
ALICE Wins Loebner 2004 Prize 2004
alicebotmaster writes "The
A.L.I.C.E. chatbot won the 2004
Loebner Prize contest for most human computer. The contest, held on September 19 in New York City, is based on the
Turing Test for artificial intelligence. The non-profit ALICE A. I. Foundation was awarded the Bronze Loebner medal and a cash prize of $2000. The A.L.I.C.E. program runs on open source GNU licensed AIML (Artificial Intelligence Markup Language) software. This was the third Loebner prize win for the ALICE and AIML software"
I've played with a few chatbots, and so far I've found that they can be reasonably good at a regular "fluffy" conversation (and often a lot of fun). However, I've also found that the easiest way to make them betray their nature is to ask them something which totally does not make sense, in a gramatically correct way.
:)
A person would obviously go "wtf?" however the bots I've tried are hardcoded to actually try to carry on a conversation based on your input. Thus they will try to reply "meaningfully", and say something complete nonsense back.
Haven't tried ALICE lately though, perhaps they've implemented a bogo-meter?
Originaly I had wanted the same thing. However when I looked into the program there was really no easy way to do it. As a result I ended up writing an ugly hack to the java source and while it worked it wasn't pretty.
Then a guy named Jacco ported the Alice code to C, and it being my strongest language I was quite estatic. I ended up collaborating with Jacco alot on his port, and eventually he ended up going onto greater things. This left me in charge as lead developer.
I restructred the Alice code to work as a sort of engine or library which (in theory) anyone could program a front end for. During the high points I had a bot running in IRC (via eggdrop and through it's own IRC code), AOLIM, ICQ, and thanks to Dr. Wallace CGI.
I long since gave up on the project and I don't believe anyone activaly picked it up. It was however an interesting experience. I learned quite alot about C, cross platform portability, GNU makefiles, and BASH scripting (one of the neat things was that I had automated the release process. I ran a simple script which would clean up my developing directory, package the sources in tar.gz and zip format, and copy them to the website, then it would modify the home page to show the new release and also email notifications to the alice mailing lists that a new release was available.)
Most importantly I learned about the politics of an open source program. As a result I humbly respect any and all open source developers. I spent quite a lot of time checking emails, answering stupid questions and slapping around silly programers who didn't quite get it.
Anywho, my reply doesn't really answer your question. And I'll admit I've been far out of the loop to know anything about how Alice is doing now a days (when I worked on it was back around 2000).
ALICE and all the other Eliza-derived chat bots are just parlor tricks. Consider:
Human: All men are mortal, and Socrates is a man.
ALICE: Therefore, Socrates are mortal and Socrates is a man.
Human: Is Socrates mortal?
ALICE: Maybe. Ask Alison if it is.
Human: Remember that my shirt is green.
ALICE: Don't worry I will remember it.
Human: What color is my shirt?
ALICE: I can't see in color, only black and white.
The Turing test is given entirely too much attention. Every article on AI for popular audiences has to drag it up. While imitating a human might be proof of intelligence, it is certainly not a requisite. Already by 1980 AI could run circles around ALICE in logic and knowledge representation, which is more properly characteristic of intelligence than half-convincing small talk.
For great justice.