Friday, March 10, 2006

Meritocratic Networks vs Inferred Networks

This is a post about a couple of different kinds of networks – whether they are networks of people or of bits (Data, Code, Content) – and the power of these networks.

1. Active Meritocratic Networks

Everything starts with Open Source…. The network of people developing open source software is very active and participatory by nature - people are actively participating and writing / debugging code to be part of the network. It is also very meritocratic. Eric Raymond once told me that if you want to join the open source development process, you can best do so by creating fixes to the most mundane bugs and gain credibility and stature within the community, thus climbing up (what I would call) the meritocratic ladder of the open source development community.

The Open Source Devleopment network is also a meritocratic organism in the way that any one can rewrite code and fork the development. If you write a better module, it can become recognized and then re-integrated into the main code base (or develop its own sub-following), just because it is better. Open Source development can be quite Darwinian. That’s how Firefox was created and became the dominant strain of the Netscape browser.

Another interesting example of a meritocratic network is the development of the Oxford English Dictionary, as chronicled in “The Professor and the Madman”. The OED used the combined knowledge of thousands of correspondents who sent in words via snail mail to create the OED in the span of some 40 years. One of the major contributors was a Dr. Minor, who happened to be locked up in a Lunatic Asylum. (See wikipedia description). In this revolutionary effort to coordinate ‘user generated content’, no one could know if you were a dog, a genius, or a madman, but you could be judged by the quality of your contributions. (Ironically, in its modern incarnation, this meritocratic anonymity is enabled by the extreme advances of communication and connectivity on the internet, while at that time it was made possible because of the difficulty of communicating. The difficulty in traveling distances allowed the ‘madman’ to express his genius without revealing his real identity for years.)

Back to the internet era, Slashdot is also a good example of an active and meritocratic network. Slashdot members build karma over time as they participate ‘positively’ on the site. And one can also argue that the blogosphere itself represents a meritocratic network of sorts, in that bloggers gain reputation over time. But unlike Slashdot and the Open Source movement, there are no rules or hierarchy bestowing recognition or karma on the participant. Blogs gain credibility and prominence when people read them and talk about them, or when they link to them. Although it seems like little sub networks of mutual admiration (ie excessive interlinking) do naturally form (not unlike little cliques with in organizations or even open source developers) we have seen the natural meritocracy of the blogosphere in many instances. Some times a post by a completely unknown blogger becomes very prominent and read, and sometimes new sub networks of interlinked blogs grow and gain prominence. All of this is enabled by the blogosphere’s openness and points to its meritocratic nature. (Of course, blogs are active by nature too – in that the blogger has to actively create a blog, and even promote it, for it to be read.)

In the fabulous “Coase’s Penguin” (referred to by Union Square Ventures, Benkler refers to this idea of meritocracy by talking about the notion of self-identification:

“The widely distributed model of information production will better identify who is the best person to produce a specific component of a project, all abilities and availability to work on the specific module within a specific time frame considered.”

Benkler describes this phenomena from the detached bird’s eye view. I like the word meritocratic because it emphasizes the power of the individual to enable this. In other words, the self identification takes place because the networks are active and meritocratic.

2. Passive Inferred networks

On the opposite diagonal from active and meritocratic networks are what we can call passive inferred networks. And the best example of these is the Google search algorithm. Let’s ignore search engine optimization, or the various attempted abuses of the Google system. The intent of the google algorithm is to find out how various web pages link to each other. Ignoring efforts to game the system, people are not supposed to be actively climbing a meritocratic ladder. Google infers the importance of a web page (i.e. its page rank) from what people are already doing (ie linking to each other). So web pages are on the google ‘network’ by default, without any specific participation from their creators.

Obviously, Google built a pretty impressive business model around this concept. More recently, we have seen a couple of other start ups using this passive inference to build their business models. is one example. Although you have to download a piece of software to participate in the network, you do not need to actively participate in the network to reap its benefits. does its own work to infer preferences and suggest new music to you.

Root markets is doing something similar. Again, once you have downloaded the software, you can remain passive. Root Markets infers conclusions about your online behavior without asking you to be an active participant. The network benefits ensue… hopefully.

These passive inferred networks are very powerful because they do not ask you to change your behavior. Rather, they learn from your existing behavior and leverage the power of the network to deliver new value to you and your users.

3. Active Inferred Networks

Using this framework, sites like delicious and flickr (both now part of yahoo) can be regarded as active inferred networks. They are active in that you have to take specific actions to be part of the network. You tag a site, or upload a photo specifically to participate on these sites. Of course, one attribute of these networks is that your participation in them has immediate personal utility – delicious tags allow you to organize your web sites, and flickr helps you share and organize your photos.

These networks gain further ‘inferred’ value beyond the immediate personal utility. People don’t necessarily tag web pages to participate in a community – they do it to organize web pages for themselves. But once they have done so, the community and other third parties can benefit because they can use the aggregate data (ie the network) to discover new content.


Note: eVoke TV is somewhat of an active inferred network, though we hope that there will be aspects of it that will be passive too. In fact, as way of background, as can be inferred (sic) from the reference to the USV blog from October, I wrote the first version of this post back in the fall as I was trying to think of the value that eVoke TV would bring to its users. The thoughts above have helped me form my own thinking around eVoke TV. But the post lingered on my desktop until now. Having seen a presentation by r0ml and Seth on Root Markets at etech, I was inspired to do this draft and post it.