Collective/Artificial Intelligence

I stumbled onto an interesting device a few months ago, which, in relation to our discussion of the collective intelligence embodied in the concept of Web 2.0 and the ilovebees ARG, seems interestingly relevant to our class. Google Talk, programmed by Google researcher Douwe Osinga and available on his website (unfortunately, the website is down at the moment of writing--check out his blog instead) generates text from Google search results. The user enters a phrase (usually the beginning of a sentence, like "A bird in the hand is") and the device completes the phrase with the most common subsequent word in the search results (in this case "worth"). After dropping the first word in the phrase, the device repeats the process infinitely, producing a potentially endless stream of 'consciousness' coming from Google searches. Indi Samarajiva's article about Onsinga's devices includes a pretty representative example of the device's output: beginning with "jesus will return," Google responded with "to Kings Associated Press July End did did Nature build the worlds largest Sex personals site!" The response indicates both the achievements and shortcomings of the device--it sometimes reveals how words are most frequently combined throughout the internet, in this case to advertise a pornographic personals site, but other times it produces random gibberish from the uncountable quantity of words on the Internet.

Osinga's device runs off the collected intelligence of the pages of the internet. Admittedly un-intelligent at this point, the device manages to generate text from the massive production of language by every contributor to the internet. It depends on the multi-user production of internet texts--the definitive quality of Web 2.0. Obviously, the device most successfully completes cliches because of their frequent use. However, in general, the device still produces intelligible language. In a way, the device allows Google to speak--it independently constructs new texts from limited user input. However, without the capacity to learn the rules of language from its continued inputs and outputs, the device will remain un-intelligent. In the attempt to create natural-language searches and user-interfaces in general, the ability to create unique verbal output in response to verbal input represents an important step toward mergent behavior. The goal, it seems, would be to creation Google dialogue--a technology which combines chat-bot and internet searching capability which can not only process natural language inputs, but produce natural language outputs from the searched documents. The interesting thing is that, by combining the user-created texts on searched websites into an approximate natural-language response, the device would potentially be putting individual users into a single dialogue with millions of web contributors. The device, if it could learn the appropriate behaviors, could possibly distill the collective intelligence of the internet into an artificial intelligence that "talks" to individual users.

Beyond the incredible potential for information gathering such an interface would create, it could represent a new step for the goal of AI--rather than programming all the processes of one human brain into a single computer, perhaps AI would process and reformat the collected human intelligence already found on the internet.