A few weeks ago there was a Robot Games & Expo held at Fort Mason here in San Francisco. The reason I know this is because at the Internet Archive's public Friday lunch the day before, Mario mentioned that he was looking forward to seeing an AI robot make its way through a maze, a la rats in a maze looking for cheese, that classic experiment.
I can't quite remember why I began thinking about conversational bots, but somewhere along the line I did. You know, those things that are launched into IM, or maybe even a phone support conversation tree that gets you far enough that your enquiry is given context before you speak to a human.
Combine that with a conversation with John Burns, Director of Advanced Technology Planning at Jstor, who was telling me about all sorts of cool things like a keyword being more relevant if it's found towards the beginning of a book, etc.
Then, I thought it might be interesting to throw the maze-wandering AI bot fueled with some sort of conversation or lexicon into a set of information, say, like the Library of Congress Subject Headings, and have it try to find the "cheese" heading. Then, run it again, say a million times.
I wonder what its paths would look like. Would they improve? Would they generate any sort of meaning? Would the connections that a bot like this found be useful to a search?
PS - When I was poking around robotics and Uncanny Valley-esque things on You Tube, I found a clip about Eva, "a synthetic approximation of a socially-interactive human." The clincher for me in this clip was when the researcher who built Eva was interacting with "her" and she was helping him find a staff member in the building. Hilarity ensued (for me) when she replied While you wait, do you want to learn some physics?
If there are any students out there who might be interested to try something like this, please do write to me. glo at archive. And that's a dot org, by the way.
I can't quite remember why I began thinking about conversational bots, but somewhere along the line I did. You know, those things that are launched into IM, or maybe even a phone support conversation tree that gets you far enough that your enquiry is given context before you speak to a human.
Combine that with a conversation with John Burns, Director of Advanced Technology Planning at Jstor, who was telling me about all sorts of cool things like a keyword being more relevant if it's found towards the beginning of a book, etc.
Then, I thought it might be interesting to throw the maze-wandering AI bot fueled with some sort of conversation or lexicon into a set of information, say, like the Library of Congress Subject Headings, and have it try to find the "cheese" heading. Then, run it again, say a million times.
I wonder what its paths would look like. Would they improve? Would they generate any sort of meaning? Would the connections that a bot like this found be useful to a search?
PS - When I was poking around robotics and Uncanny Valley-esque things on You Tube, I found a clip about Eva, "a synthetic approximation of a socially-interactive human." The clincher for me in this clip was when the researcher who built Eva was interacting with "her" and she was helping him find a staff member in the building. Hilarity ensued (for me) when she replied While you wait, do you want to learn some physics?
If there are any students out there who might be interested to try something like this, please do write to me. glo at archive. And that's a dot org, by the way.