I started on Friday talking about aggregation of information and its potential in producing objective and spherical knowldedge. The day after (on Saturday) I noted the emergent behavior such networks can develop.

Just a reminder; these series of posts are supposed to highlight the difference between having a set of experts in your network/community/system/blogroll/name-it-as-you-wish and having an open public, without posing any restrictions. In other words the series are about the fundamental difference between Google’s Knol and wikipedia.

Today I will continue this series of posts, explaining the four preconditions necessary for a correct aggregation.

Diversity; an example

Let us consider a blog with liberal political content, which mostly references other liberal blogs. It is logical to presume that the knowledge aggregated from these blogs will also be liberal – an one sided view of the political system. The network these blogs create, will in its turn mostly gather users who also agree with this perspective. The impact to the collective information is obvious; it will treat all political matters liberally. But apart from that, the network will consist of a small liberal group of people excluding other opinions and accordingly viewers. An objective network (meaning one that sees subjects under different perspectives) cultivates constructive discussions and therefore draws the interest of a wider public. It also sets the tone in which discussions take place and a common goal that people can join in attempting to reach.

Diversity of perpectives

Emergent systems manage to function so well, because a collective macrobehavior is achieved when the system consists of a multitude of agents, each one attributed with a simple operation. In other words emergence is accomplished (among other reasons) due to the multitude of different agents. Emergence relies on the diversity of its agents.

The Condorcet Jury theorem, which I also mentioned in an earlier post, relies on the fact that in a diverse group the chances that at least someone will propose a new, radical idea is increased. Also influence between individuals or clustering of opinions are less likely to appear in a diverse group. This means that the information collected will approach an objective view of the subject.

Diversity of expertise

A diverse group does not only imply collecting a set of different perspectives about a topic. It also means creating a group of individuals with different grade of expertise and knowledge. Homogeneous groups have the tendency to refrain from investigating alternatives, they find it harder to continue learning and thus bring less new information in the community. Less experienced members will provide fresh aspects and propose questions, which under different circumstances would not be expressed.

Conclusion

We must keep in mind that the focus of a community is not to consist of wise individuals. Instead it concentrates on making wise decisions. I do not suggest here that a diverse group of uninformed individuals could collectively succeed more than one of experts. But an assemblage of people with various degrees of insight, may give better results than a few specialists.

Naturally, when i use words like ‘creating a community’ or ‘producing collective knowledge’, I do not mean that you can just choose who contributes to your network. (you might be able to guide it a bit, but in the end its out of your hands). On the contrary the traits, challenges and participators of your network will act as a magnet to a diverse group.

The question that emerges is: ‘Do you want to be open to everybody and hope for the best, or do you want to invite only experts and observe a clustering of knowledge and perspectives?

In the late ’90s Marvin Minsky published a book called ‘Mentopolis’. He documented the human brain as a distributed network, consisting of a multiple agents, where each one of those agents is responsiple for just one operation. In the picture below, for example, he proposed that in order for our brain to recognize an apple all these agents should be set in motion. The ‘color’ agent should collect his information and send it to the ‘look to’ agent, who in his turn would communicate with the ‘place’ agent and so forth. My interest in this network (called the find-machine by Minsky) is not its credibility but its properties and attributes.

Emergent networks

The system Minsky composed was a typical example of an emergent network, namely a system with multiple agents dynamically interacting in multiple ways, following local rules and oblivious to any higher-level instructions. Minsky visualized a perfectly functioning system, with absolutely no central control. The nodes (meaning the agents) are interacting in order for their microbehavior (sorting color, size, etc.) to result in a macrobehavior (perceiving the object). Such organizations are present in nature (see the work of Deborah Gordon on the emergent behavior of ants), computer software and even in the structure of cities and are giving us a glimpse of networks, which correctly aggregate information.

Emergent systems function so perfectly, because they work with neighbor interaction, feedback, pattern recognition and indirect control. They are designed to learn from the ground level, to take advantage of local knowledge for an upper goal. Through interaction, they are capable of recognizing patterns and indirectly controlling the whole system.

Emergent social web

I’m not implying that the social web undertakes a completely emergent behavior. We are dealing neither with oblivious users nor with pattern recognition systems (at least not yet). But still there are perfectly functioning communities, which adopt the traits of an emergent behavior (probably slashdot, wikipedia and the linux operating system being the most profound examples). There is not any administrator – at least not in the traditional sense – leading the community. The users are self organized, sometimes each one responsible for a specific activity and always working together to provide quality material. Under that perspective we are experiencing the formation of online emergent networks, which are developing a life of their own – a life without any central control.

But what makes such behavior so successful? As I argued on my previous post regarding aggregation of knowledge (and your additions are mostly welcome on this), their success lies on:

• diversity
• independence
• indirect control
• self-organization

Conclusion

If such systems (and among them is the World Wide Web itself) manage so successfully to collect knowledge without any central power, why should we accept the control of any authority, which would define who posts which article and who links where? Years of experience show us that such ‘problems’ of the web can regulate themselves.

In following posts I will concentrate explicitly on each of the above-named traits of emergent networks with the hope of justifying my thesis, that expertise is not the only path to knowledge.

For this post the book of Steven Johnson: Emergence and of Marvin Minsky: Mentopolis (where the photo also comes from; original was in german, I translated it) where of great assistance.

In my previous post about Google’s Knol and the role of the author I posed the question, in what extent do collaborative networks need author(itie)s to aggregate information correctly. And by correctly I mean, objectively - taking note of all sides of the subject.

Condorcet Jury Theorem

First I’ll try to explain why aggregation of knowledge can actually bring better results, than the opinion of a single expert. I’m based on the assumption of the Condorcet Jury Theorem, which supports that the probability of a correct answer by a majority of the group increases toward 100 percent as the size of the group increases. The theorem is based on the hypothesis that people are answering a question with two possible answers (one right and one wrong) and that their answers are not random – on the contrary they have more than a 50 percent probability of being correct.

Naturally, extensive criticism has been leveled at the binary logic of the Condorcet theorem, since a question has usually a wide spectrum of answers. But recent studies have shown, that even when the group is dealing with multiple options (instead of a true and false selection), there is still a high probability that it will actually conclude to the right answer, as long as the individuals tend to choose the right option.

The Catch

But there is a catch to the theorem: Correct aggregation of information does not simply rely on a large group of people. This group should fill a number of preconditions (which can in a way also be identified as an emergent behavior). During my research and study (and partly also during my own personal thinking) I collected these preconditions:

• diversity
• independence
• indirect control
• self-organization

The Concept

I’ll start a series of posts about each specific point. In that way I want to advocate for wikipedia’s system (or any collaborative network, that does not encourage ownership) within the scope of the discussion about google’s Knol. Namely, my assumption is that any participatory system that promotes ownership (authorship) and control, will eventually produce one-sided information.

If you have any additions or thoughts on the subject, improvize - contact me.