Yesterday Simon posted an insightful essay about blogs functioning as a news filter. Blogs as filters—what an idea. By adding sites to his blogroll that he knows pull data he is interested in, Simon filters his news intake quite effectively. It got me to thinking... If Blog rolls function this way perhaps the blogosphere itself does much more.
Sure, Simon's blogroll is a filter from his perspective, and a good filter it is. But what analogy can be drawn for the whole 'sphere?
From a larger perspective, a group of bloggers doing what they do, posting, reading, blogrolling, linking, and tracking-back (or trackbacking?) functions like a large, complicated neural network.
Every description of a neural network starts with a model, and this one is no different. Starting from the idea of a perceptron, each blogger is a neuron in the network. Each neuron at different points in time fires, based upon some function of its input. In the blog-network model, this happens when a blogger posts an entry. The blogger has received enough input to cause them to compose an entry. Each blogger also has a set of incoming connections in the form of a blogroll. Like the neuron whose synapses connect to other neurons to receive their output as its input, the blogger maintains a blogroll to receive input from other bloggers. There you have it—that's the basic model.
Let's get into the details.
A blogger neuron receives input from the output of other bloggers. Not every other blogger out there, just a set few. That few are his or her blogroll or regular reading list. Other external input comes from sites that are not regular blog-like sources, which in this model I consider outside the network. These sources are analogous to external input fed into a standard neural network such as a training set or input upon which to operate. We can measure (to some degree) what incoming connections are being simulated by looking at what the blogger is linking to. Harder to measure are the strength of the links (weight) connecting the blogger neuron to its inputs.
Like a recurrent network, the 'sphere records state through trackbacks, referrer logs, comments, and Technorati, feeding stimulation in the opposite direction of reading and linking. Who is paying attention to my output? These reverse stimulation mechanisms provide the answer. A network that uses this structure is called recurrent and can predict sequences, or like a Hopfield network, perhaps simulate "robust content-addressable memory".
One hole in my analogy here is the lack of usual training that is involved with generating neural networks. How can you propagate error backwards on the blogosphere and correct error? Blogger's are individualistic about their tastes, making it hard to give even a guess as to what their activation function must be. Forget sigmoidal, it seems random at times from an outsider's perspective.
Without a differentiable activation function, error minimization is hard. If you don't know what direction to move in to decrease the error, you can hardly minimize it.
Here, the blogosphere resembles more a Kohonen map (or Self Organizing Map) than a neural network. By moving similar units closer together (blogrolling and other forms of association), the network organizes itself without supervised training. There doesn't have to be a "training set" with defined correct answers.
The problem with this model is that any sane researcher training neural networks defines a way of drawing conclusions from the network. This might be a set of output neurons, one of which firing indications a particular classification of the input, for example. How can someone do that using the blogosphere?
I don't have any specific answers to that, but one thing does come to mind. We have the perfect platform upon which to build an application that does just that. Google. Because as Simon points out, it isn't just a search engine; it's "the world's largest and most scalable platform for developing huge web-based applications".
So one day, someone will develop the gigantic planet eating robot, with an ultra-sophisticated blogonetwork web app as a brain, drawing on the power of Google and the blogosphere to do its evil work. I for one am glad to be a part of it.