Human Networks And Causality Cones

The study of networks has changed the way we think about our world and the way that societies organise themselves within it. In particular, the discovery that many real world networks can be thought as small worlds, in which most nodes are not neighbours but can be reached by a small number of jumps, has had a profound impact.

But until now, most studies have focused on networks as static affairs in which the links between nodes do not change in any significant way.

That is beginning to change as data becomes available from mobile phones and RFID tags that tease apart the nature of human behaviour and interaction in detail that has never before been possible.

Today, Lorenzo Isella at the Institute for Scientific Interchange Foundation in Turin and friends reveal an interesting example in which the interactions between in humans in similar but not identical circumstances leads to subtle but important differences in the network of connections between them.

Isella and co have examined data taken from RFID cards that recorded the interactions between people at two different events: an exhibition at the Science Gallery, a museum in Dublin, and a conference at the Institute for Scientific Interchange Foundation in Italy.

These data sets are quite different. At the Science Gallery, researchers recorded 230,000 interactions between 14,000 people over a period of three months. At the conference, they recorded 10,000 interactions between 100 people over three days.

People’s behaviour at these events were obviously different. At the museum, people arrived at different times and streamed through the gallery in just a few hours. At the conference, the attendees tended to stay on site and make repeated contacts over several days.

So it’s hardly surprising then that the average number of contacts made at the conference was more than double those made at the museum (roughly 20 v 8).

The networks were also different. It turns out that the pattern of interactions between attendees at the conference formed a small world network while the pattern at the museum often did not. So it is much harder to link visitors to the museum to each other using a small number of steps through the network.

These differences have an important implication which Isella and co were able to draw out by studying the way that infectious agents such as memes or viruses might spread through the respective networks.

At the conference, information and viruses would spread relatively easily, in a manner that has been well studied for small world networks. Anyone at such a conference could easily start a rumour or spread an infection that reaches all other attendees

However, it was much harder to spread infection at a museum. The crucial reason for this is that the network has an important temporal element. A visitor who arrives at the museum early in the day could start a rumour or spread a disease that has the potential to reach all or many other visitors that day. However, a visitor at the end of the day can only spread an infectious agent to those already there or who come later.

In some ways, this is a trivial observation: you can’t infect a person who has left an area before you arrive. Infection can only occur with some kind of causal interaction. But this has been more or less ignored in the study of static networks.

That will have to change as the data gives scientists access to the dynamic nature of networks. They’ll have no choice but to consider the fascinating issue of causality and the role it plays in networks. It’s possible to imagine ‘causality cones’, like light cones in physics, which map out regions of the network that individuals can and cannot influence. That, in turn, raises the prospect of a kind of relativity science applied to dynamic networks.

There’s not telling where that kind of thinking will go. Network science has matured rapidly in the 20 years since it became mainstream. But the study of dynamic networks reminds us that there is much more to come.

Ref: arxiv.org/abs/1006.1260: What’s In A Crowd? Analysis Of Face-To-Face Behavioral Networks