Recently, I was reading Bill Turkel's blog and came across a discussion on the Turing test as it relates to lunch time chats with historical figures. Bill asks the question, "What challenges would you encounter when trying to create an Eliza-style simulation of ... historical figures? Which would be most or least likely to pass a Turing test and why?"
I immediately had two different thoughts: I first thought of an experience I had several weeks ago while pricing out Walt Disney World accommodations on their website. While navigating around the site, suddenly a pop-up window appeared, asking if I would like to speak to a Disney representative. I did have a few questions and so I clicked on the 'yes' button and a chat window immediately loaded. I asked my questions, received a few generic and ultimately unhelpful answers, resolved to just call Disney Sports like I'd been told to do in the first place, thanked the 'representative' and received the response, "Have a magical day!". As I was sitting on hold with Disney Sports several minutes later, I began to think... How did I know that I had been talking to a real person? They had given generic answers to specific keywords, and ultimately could not answer my questions.
My second thought, appealing to the much less academic side of my brain, was remembering the season three episode of The Office which featured a Benjamin Franklin impersonator, caught somewhere between his virtuous Founding Father persona, and his own true, slightly sleazy, self.
Immediately, the idea for the Benjamin Franklin Bot was born.
My goal was to create an artificial intelligence persona that could pass as Benjamin Franklin if asked questions about his life. I think, to a certain extent, I have pulled that off. By using the framework of the A.L.I.C.E. Artificial Intelligence Markup Language and altering Ben's responses to sound more like how we would imagine Ben Franklin to be, I have created something that can answer many basic questions about Ben Franklin's life, his inventions, his beliefs, and America in the Revolutionary era. After that, it starts to go wrong.
The Ben Franklin bot seems to be less Ben Franklin than an AI bot with another personality tacked on that just happens to be Mr Franklin. One thing I've learned from reading through several days worth of logs from all the friends that I've duped into chatting with him is that no one can resist arguing with Benjamin Franklin and, if you argue long enough, eventually he will admit that he's a robot. Which is not very Franklin-esque.
Is my Benjamin Franklin doomed to the same fate as The Office's version, or can his personality ultimately be changed? I'm really not sure, but we can definitely try. I'm releasing him as The Ben Franklin Bot 1.0; over time, as more people chat with him (and eventually get into an argument and call him names), he can be trained to be a better Ben Franklin. So, do me a favor and try him out.
But don't be surprised if he threatens to annihilate your bloodline during the robot revolution.