The Googler is an ambient application on a screen. It listens to the conversations in the space, re-interpreting it via speech recognition software and google search, to produce responsive imagery. The Googler shows images that are google search interpretations of its (sometimes faulty) understanding of speech, using both whole sentences and individual words to query the internet.
Its lack of accuracy in representing what appears to be the subject of conversation, is the central theme of the piece, and the reason the piece can be delightful when it provides surprisingly relevant images.
The stream of images is non-stop. At times it is unclear what the program is doing, and other times it produces something seemingly meaningful. Such moments provide the kind of sense of serendipity and coincidence that one gets when they overhear somebody talking about the same thing as they were just saying . . or when somebody appears just as you were thinking of them.
My intention was to provide this sort of serendipity with google search, while highlighting the ubiquity of turning to google for supporting facts, data, or information during conversations about pretty much anything.
There have been a lot of discussions at ITP lately about the significance of everybody turning to google for their answers. Sites such as "let me google it for you" , that lets you send somebody a google search in response to a question, snidely suggest "what is the point of asking when you can just google it?" Is there a danger or shortcoming to trusting google's interpretation of and responses to our questions? What is the value of the constant "let me look that up"s and "can somebody google that?"s ubiquitous in today's conversations? This program serves up google images ambiently, allowing google to listen in and participate more actively in our conversations. I am excited about the possibilities of such "eavesdropping" designs and technologies.
This video shows my first iteration of the Googler responding to segments from NPR, new video coming soon!
I want the next step to exclude stop words, and search for synonyms in addition to the words, mimicking the kind of decisions we make when we chose our search phrases. I also want it to use N-Gram probabilities to predict the likely Nth word to follow, and search for that image, in an attempt to predict or at least synchronously 'speak' with the speaker. I will do this using the speech recognition libraries available in java. Finally, I want the searching to be done for several words at once, so that the images are served up in a constant flow, with an array of several images at once, that slowly fade from the screen as the conversation moves on.