On Saturday I presented the first part of this two part post. It outlined what I believe will be a better internet. By this I mean an internet that “knows” you. An internet that doesn’t give you search results you don’t want. An internet that doesn’t introduce you to people you won’t like. Read the post for more details. Today’s post is about HOW that better internet will get to know you. I have two thoughts on how this will happen for simplicity I’ll call the first one math and the second one cliques.
The math or “Google” model is based on linking disparate data points to make sense of you. There is a certain elegance to this type of model, in that it will mine the data of millions of users to discover things about us that we never knew. The math model for example may be able to identify that because you like playing scrimmage hockey and are over 50, you may like arthritis medicine. It can tell this because most pickup hockey players over 50, like arthritis medicine. The systems set to figure out these trends would do it using regression models; comparing every potential data set randomly till trends were identified. So, while it would have eventually figured out that hockey players above 50 like arthritis medicine, it would also have noticed that hockey players above do not like perfume. The trick here is that it’s anonymous. You don’t tell the math method who your friends are, it places you in a bunch of different demographics and does it for you.
In a cliques based better internet, the internet will know you because of what you and your friends like. This is what I’m calling the “facebook” model. This “social” model is not based on millions of disparate data points, it’s based on a few data points that it can tie directly to you. It can tell that one of your friends bought arthritis medicine and then two of your mutual friends bought it and assume that you might like some. The trick here is that it’s context based. It takes some information that it knows (who your friends are) and tries to figure the rest out.
Here are some common examples of both systems:
- Google search is the best known example of the “math” method. It groups you, anonymously with other people who have searched for the same keyword. It uses a formula to figure out what you might like. As it evolves it will learn to use other data points. For example, on your GPS enabled mobile phone, Google automatically applies a location context to your search. Imagine if it used past searches, profile information, frequently visited websites, browsing habits, etc… to provide additional context.
- Facebook is of course an example of the “cliques” method. Today Facebook suggested I become a fan of the Pittsburgh Steelers because 16 of my friends are. There are some limitations with facebook today because my buying habits are in no way similar to many (most?) of my friends. However, that’s beginning to be solved by allowing me to group my friends. In the future it will look at my closest friends when suggesting something to me.
- Hunch is another amazing example of the “math” method. It uses some random games to find out whether or not I will like something. While this is exciting, the really interesting part is that some of these games are automated now. If you’re looking to really surprise yourself, try the Twitter predictor game. Simply put in your twitter ID and see what Hunch can learn about you from your Tweets and your Tweeps. According to their blog, Hunch can predict how you’ll answer 85% of its questions from that information alone. I covered Hunch’s potential in more detail in this old post.
Got other examples? Add them to the comments.
