Thursday, October 11, 2007

A Truly Distributed Creative System

Posted to idc, October 11, 2007

John Hopkins wrote, on idc:

You cannot have a truly distributed creative system without there
being open channels between (all) nodes.
I don't think this is true.

Imagine an idealized communications system, where links were created directly from person to person. If all channels were open at any given time, we would be communicating simultaneously with 6 billion people. We do not have the capacity to process this communication, so it has the net effect of being nothing but noise and static. Call this the congestion problem.

This point was first made to me by Francisco Valera in a talk at the University of Alberta Hospital in 1987 or so. He was describing the connectivity between elements of the immune system, and showed that most effective communication between nodes was obtained at less than maximal connection, a mid-way point between zero connectivity and total connectivity. Similarly, in human perception, we find that neurons are connected, not to every other neuron, but to a subset of neurons.

What this tells me is that what defines a "truly distributed creative system" is not the number of open channels (with 'all' being best) but rather the structure or configuration of those channels. And in this light, I contend that there are two major models to choose from:

- egalitarian configurations - each node has the same number of connections to other nodes
- inegalitarian configurations - nodes have unequal numbers of connections to other nodes

Now the 'scale free' networks described by Clay Shirky are inegalitarian configurations. The evidence of this is the 'power law' diagram that graphs the number of connections per member against the number of members having this number of connections. Very few members have a high number of connections, while very many members have a low number of connections - this is the 'long tail' described by Anderson.

The networks are scale free because, theoretically, there is no limit to the number of connections a member could have (a status Google appears to have achieved on the internet). [*] Other inegalitarian networks have practical limits imposed on them. The network of connections between airports, for example, is an inegalitarian configuration. Chicago is connected to many more places than Moncton. But the laws of physics impose a scale on this network. Chicago cannot handle a million times more connections than Moncton, because airplanes take up a certain amount of space, and no airport could handle a million aircraft. This is another example of the congestion problem.

What distinguishes the inegalitarian system from the inegalitarian system is its the number of 'hops' through connections required to travel from any given one member to another (this can be expressed as an average of all possible hops in the network). In a fully inegalitarian system, the maximum number of hops is '2' - from one member, who has one connection, to the central node, which is connected to every other node, to the target node. In a fully egalitarian system, the maximum number of hops can be much higher (this, again, is sensitive to configuration).

As the discussion above should have made clear, it should be apparent that fully inegalitarian systems suffer as much from congestion as fully connected systems, however, this congestion is suffered in only one node, the central node. No human, for example, could be the central node of communication for 6 billion people. This means that, while the number of hops to get from one point to another may be low, the probability of the message actually being communicated is also low. In effect, what happens is that the inegalitarian system becomes a 'broadcast' system - very few messages are actually sent, and they are received by everyone in one hop.

In other words - maximal connectivity can result in the *opposite* of a truly distributed creative system. It can result in a maximally centralized system.

I'm sure there's a reference from critical theory or media theory, but what would to me define a truly distributed creative system is 'voice' (sometimes called 'reach'). This could be understood in different ways: the number of people a person communicates with, the average number of people each person communicates with, the minimum number, etc. My own approach to 'voice' is to define it in terms of 'capacity'. In short, any message by any person could be received by all other people in the network. But it is also defined by control. In short, no message by any person is necessarily received by all other people in the network.

One way to talk about this is to talk about the entities in the network. When you look at Watt and Barabasi, they talk about the probability that a message will be forwarded from one node to the next. This, obviously, is a property of both the message and the node. Suppose, for example, that the message is the ebola virus, and that the node is a human being. The virus is very contagious. If contracted to one person, it has a very high probability of being passed on to the next. But suppose the person is resistant. Then he or she won't contract the virus, and thus, has a very low probability of passing it on.

The other way to talk about this is to talk about the structure of the network. The probability of the virus being passed on increases with the number of connections. This means that in some circumstances - for example, a person with many friends - the probability of the virus being passed on is virtually certain. So in some network configurations, there is no way to stop a virus from sweeping through the membership. These networks are, specifically, networks that are highly inegalitarian - broadcast networks. Because the virus spreads so rapidly, there is no way to limit the spread of the message, either by quarantine (reducing the number of connections per carrier) or inoculation (increasing the resistance to the message).

In order to create the truly distributed creative system, therefore, you need to:

- limit the number of connections for any given node. This limit would be based on what might be thought of as the 'receptor capacity' of any given node, that is, the maximum number of messages it can receive without congestion, which in turn is, the maximum number of messages it can receive where each message has a non-zero chance of changing the state of the receptor node.

- maximize the number of connections, up to the limit, for any given node. This might be thought of as maximizing the voice of individual nodes. What this does is to give any message from any given node a good start - it has a high probability of propagating at least one step beyond its originator. It cannot progress too fast - because of the limit to the number of connections - but within that limit, it progresses as fast as it can.

- within these constraints, maximize the efficiency of the network - that is (assuming no congestion) to minimize the average number of hops required for a network to propagate to any other point in the network.

These conditions combine to give a message the best chance possible of permeating the entire network, and the network the best chance possible of blocking undesirable messages. For any given message, the greatest number of people possible are in a position to offer a countervailing message, and the network is permeable enough to allow the countervailing message the same chance of being propagated.

What sort of network does that look like? I have already argued that it is not a broadcast network. Let me take that one step further and argue that it is not a 'hub and spokes' network. Such networks are biased toward limiting the number of hops - at the expense of voice, and with the risk of congestion. That's why, in hub and spoke networks, the central networks become 'supernodes', capable of handling many more connections than individual nodes. But this increase in capacity comes with a trade-off - an increase in congestion. This becomes most evident when the supernode attempts to acquire a voice. A centralized node that does nothing but reroute messages may handle many messages efficiently, but then the same node is used to read those messages and (say) filter them for content, congestion quickly occurs, with a dramatic decrease in the node's capacity.

Rather, the sort of network that results is what may be called a 'community of communities' model. Nodes are highly connected in clusters. A cluster is defined simply as a set of nodes with multiple mutual connections. Nodes also connect - on a less frequent basis - to nodes outside the cluster. Indeed (to take this a step further) nodes typically belong to multiple clusters. They may be more or less connected to some clusters. The propagation of a message is essentially the propagation of the message from one community to the next. The number of steps is low - but for a message to pass from one step to the next, it needs to be 'approved' by a large number of nodes.

When we look at things like Wenger's communities of practice, we see, in part, the description of this sort of network. Rather than the school-and-teacher model of professional development (which is a hub and spokes model) the community of practice maximalizes the voice of each of its members. It can be called a cluster around a certain topic or area of interest, but the topic or area of interest does not define the community, it is rather an empirical description of the community (and thus, for example, we see people who came together as a hockey team in 1980 continue to be drinking buddies in 1990 and go on to form an investment club in 2000).

Maximally distributed creativity isn't about opening the channels of communication, at least, not directly. It is about each person having the potential to be a member of a receptive community, where there is a great deal of interactivity among the members of that community, and where the community, in turn, is a member of a wider community of communities. Each person thus is always heard by some, has the potential to be heard by all, and plays a role not only in the creation of new ideas, but also, as part of the community, in the evaluation and passing on of others' ideas.


[*] I just want to amend my previous post slightly.

I wrote: "The networks are scale free because, theoretically, there is no limit to the number of connections a member could have..."

This should not be confused with the definition of a 'scale free network', which is specifically, that "a network that is scale-free will have the same properties no matter what the number of its nodes is."

But the relationship between my statement and the more formal definition should be clear. If there is a limit to the number of connections created by the physical properties of the nodes, then the mathematical formula that describes one instance of the network (a small instance) cannot be used to describe all instances of the same type of network.

No comments:

Post a Comment

Your comments will be moderated. Sorry, but it's not a nice world out there.