In the last year, tons of smart technology thinkers have been noticing the fascinating parallels between on-line networks and human, social ones. Indeed, that idea is partly behind the item I posted a few days ago — about “why are some blogs popular”, and power-law distributions.

But recent research has found that human networks and computer networks can actually be very different. Mark Newman, a physics professor at the University of Michigan, studied both and found something interesting: In human networks, people who are social gravitate towards other people who are social. But in computer networks, highly-connected nodes frequently connect to millions of dead-end nodes. Computer networks are a bit socially agnostic; humans — well, we want to hang with the cool people.

This has some intriguing implications, as Newman notes in an interview at Technology News and Review:

In social networks, where popular people are friends with other popular people, diseases spread easily, said Newman. At the same time, however, this type of network has a small central set of people that the disease can actually reach. “They support epidemics easily, but… the epidemic is limited in who it can reach,” he said.

The opposite is true for the Internet, the Web and biological networks, said Newman. This makes these types of networks more vulnerable to attack than social networks are.

The implications for vaccinating people and for protecting networks like the Internet against attacks are not good, according to Newman. The networks that we might want to break up, like social networks that spread disease, are resilient against attacks; but the networks that we wish to protect, like the Internet, are vulnerable to attack, said Newman.

Social networks hold together even when some of the most connected nodes are removed. This may be because these nodes tend to be clustered together in a core group so that there’s a lot of redundancy, according to Newman. This means that vaccination and similar strategies are less effective than in other types of networks.

Attacks on the largest nodes of disassortative networks, however, affect the network as a whole more because the connections are more broadly distributed across the network. “This suggests that if nodes were to fail on the Internet, it would have a bigger effect on the performance of the Net than we might otherwise expect,” he said. “In a way, it is telling us that the Internet is fragile.”

Newman found that the number of highly-connected nodes that need to be removed to destroy disassortative networks is smaller by a factor of five or 10 than the number needed to destroy assortative networks.

It’s certainly true that the Net, for all its vaunted robustness, is oddly fragile. Wanna cause some serious world damage? Okay — just arrange to destroy the thirteen root name servers that organize which Net addresses point to where. Poof: The whole Net goes down. This would not be easy to do, since most of these 13 servers are buried in bunkers and protected by armed guards. But that’s precisely the point: The security of the root-name server system is not computational or network-based — it’s not inherent. It’s secure because there are guys with big-ass guns protecting them. Social networks are far more redundant.