Demonstrating the filter bubble

Five or six years ago or so, Eli Pariser wrote a book entitled the Filter Bubble. The core idea in this book was that online information companies personalize our online experience serving our personal priorities rather than presenting the world as it actually is. Google and Facebook make the best examples. In the early days of Google, search results relied on “page rank”. This was a method for estimating the quality of search results based on the links among web pages. Simply put, the more folks who linked to your content, the better your content must be. Google eventually went beyond this approach taking advantage of your own online behavior and attempting to show you what their algorithms estimated would most meet your needs. Facebook does a similar thing. Pariser’s concern was that this approach supports existing beliefs and attitudes rather than providing the most accurate information. 

I tried to evaluate Pariser’s claim in a way I thought made sense. I assumed that Google would know a lot about me when I used the computer on my office desk, but would know little about me when I went into the computer lab just down the hall. I decided to use the polysemous word “apple” as my focus on technology would label me as likely to be interested in Apple computers and not the apple as in fruit. I reasoned that the search from my office would show lots of items related to technology and the search from the computer lab would include more hits on fruits. While this seemed logical to me, I found little difference. Most of the hits were on technology. 

I just encountered a blog post associated with the search company DuckDuckGo that offers a way to demonstrate the self-serving Google search bias. DuckDuckGo is a Google competitor and suggests as an advantage that it respects user privacy. The DuckDuckGo researchers asked different people to simultaneously conduct the same searchers. The searches targeted terms related to current political controversies as this is the context within which many are now concerned about the filter bubble. The data returned were compared across users to count differences and the differences did demonstrate the bias Pariser had warned readers about.

There must be a classroom project in here somewhere. If students in 1:1 environments consistently use the same computer, it would seem that the DuckDuckGo methodology would be easy to duplicate. Having students collect and analyze these data would be a great way to discuss concerns related to what we all find online.