Posted by Matthew on Wednesday September 4, 2002 @05:16PM
from the unknown dept.
Matthew writes: IBM researchers today announced that they have developed a software agent that passes the Turing test. The Turing test states that software is artificially intelligent when reasonable humans cannot tell that they are communicating with a computer. They did this by incorporating a new technology called “Artificial Stupidity” or AS.
Code-named “Anna Nicole”, the software agent interfaces with real people constantly via a special AOL Instant Message interface and an E! Television show. AOL Instant Messages are the computer’s only bi-directional interface with the outside world. The software runs on IBM’s famous Deep Blue computer, but now takes so little compute power that a port to Linux is in the works.
“The breakthrough came when we realized that we had gone too far with Artificial Intelligence. People didn’t believe our earlier attempts because they were too smart. Let’s face it, when you ask someone in chat what the square root of PI is, and that person begins answering, you know you’re not talking to a human. That meant that we were failing the Turing test every time.
“Artificial Stupidity is, at its core, a simple process. The computer makes an immediate estimation of the amount of compute time required to answer a specific question. If the estimation goes over two seconds, the computer aborts the thread and replies with a randomly selected reply from this list: “What?; Whatever; Calm down and give me your address; Dude, get serious; yeah right; okay Einstein; You’re making me hot; Was that Pepperoni?; or I’m going to have to forward you to second tier support.”
By incorporating AS in to their AI efforts, IBM has developed a product with immediate sales potential. They’re now working on speech recognition and synthesis modules to support industrial applications such as Tech Support, Drive Through and Telephone Order Taking, and 411 and 911 Operator. “We’ve found that by tweaking the speech synthesis with a heavy foreign accent, people speak a lot more slowly and loudly, which makes the voice recognition much easier, and they’re far more likely to repeat themselves. Those small optimizations are making this a “here, now” technology.