Building a better bot

AI bots come in many different varieties – from regular IM Bots, to web bots, to those really awful customer services bots that never seem to answer any questions you have. I’ll admit it: when I was a kid, I enjoyed talking with everyone’s favorite AIM Bot, SmarterChild. If I wasn’t blasting insults in his face, I would enjoy seeing in what areas his AI would fail.

But what is the best AI bot out there, if there is even one? Is there such a program that can trick a human into thinking it is actually having a conversation with another human?

That is exactly what the Loebner Prize intends to find. The Loebner Prize is designed to seek out an AI bot which can fool its entire panel of judges into thinking that they are having an online chat with another human. As of this year, a winner for the grand prize of $5,000 (fooling all the panelists into thinking an AI bot is a human) has yet to be discovered. Instead, there is the Bronze Prize ($3,000) for the robot which performs the best out of all of the contenders.
This year’s winner was Elbot. Created with the goal to someday become an advanced enough system to replace the need for human customer service, the AI bot is actually quite sound. Having chatted with it for a while I was almost startled by how well it engages conversation. There were a lot of times where the responses would simply leave me scratching my hair. Yet many times it would do something as advanced as use it’s AI to create grammar patterns, or something as simple as telling me a joke.

Could one really program an AI bot sophisticated enough to carry on a conversation? If we are talking about a conversation on the internet, then it seems quite likely; after all it isn’t like conversations over AIM are exactly Pulitzer material, now…