"From now on, if we capture foreign ships and their respective countries try to attack us, we will kill them (the hostages)," Jamac Habeb, a 30-year-old pirate, told The Associated Press from one of Somalia's piracy hubs, Eyl. "(U.S. forces have) become our No. 1 enemy."

While this rhetoric serves pirates' purposes, in conveying a menacing facade, if we view the history of events leading up to these statements through the prism of rational choice we can easily identify them as simple cheap-talk.

Now, extend this simple model onto the events that have occurred over the past several month off the Somali coast. Up to recently, the game had been in equilibrium; the state's posterior belief at each stage [hostage event] about the pirates' type was that they were willing to use violence, and as such gave into their demands to rescue the hostages to minimize the loss of life (i.e., maximize the state's utility).

Over the weekend, however, the U.S. Navy took an off-equilibrium action by attacking the pirates to rescue the hostages. In knocking the system out of equilibrium, the dynamics of the system become more well-defined. In the old equilibrium, the use of violence was ex ante inefficient for pirates, since they were able to maximize their utility without using it. The use of violence, then, becomes a costly signal for the pirates, as it reveals tactics, capabilities and weaknesses.

This more robust understanding of the dynamics provides insights into the pirates' actual type; that is, pirates were either always willing to use violence, but did not do so in the old equilibrium because it was costly; or, they never were willing to and never will be, because thei beliefs about the state's willingness to use violence had been made clear, and killing hostages in the future will not increase their reward.

Regardless of this distinction, the threat of violence is clearly cheap-talk, since it does not inform the state of the pirates' type, and should not affect their actions. Put plainly, the pirates are either threatening the status quo (continued willingness to use violence), or are making a non-credible threat (new willingness to use violence). Any future policies to cope with the pirate situation in Somali, therefore, should not use these threats as a foundation as they reveal nothing about pirates' willingness to use violence.

Update: Interesting discussion on how the 'game' has changed at orgtheory.net

Special thanks to @ChrisAlbon @ryanbriggs @pattonmat and @NYkrinDC for the lively exchange on Twitter earlier today that inspired this post.

Photo: sitfu.com

A Home Office spokesman confirmed search engine optimisation training is part of the government's anti-radicalisation strategy. "In order to support mainstream voices, we work with local partners to help develop their communication, representational and leadership skills," he said.

"This support could include media training, which can help make their voices heard more widely, and support the development of skills which allow communities to be more effective in debate."

The Register points out—rightly—that the preponderance of evidence on such manipulatory activities suggests it simply does not work. Blogger friend Tim Stevens of ubiwar.com is an expert in this field, and his co-authored report for the International Centre for the Study of Radicalisation, "Countering Online Radicalisation: A Strategy for Action," exposes these shortcomings explicitly.

From a technical perspective, there are also several limitations to such a strategy. First, Google's PageRank is proprietary, therefore, any "gaming" of its indexing would be fundamentally flawed by an analyst's assumptions about how it works. Second, though the intricacies of the algorithm are secret, it is a well-known that at its core PageRank uses Eigenvector centrality to ranks pages (rank is based on how many other high ranking pages link to you, not just number of links). The power of Eigenvectors is in their ability to accurately measuring the natural structural dynamics of large networks, which often take on an exponential or scale-free form. A conscience policy attempting to manipulate the natural mathematics of of these structures is likely a lost battle.

As a final thought, this news underscores the magnitude of Google's power, as it can now even be viewed as an instrument of national security. Google has already shown its ability to predict outbreaks of influenza in the United States; why then, would they not be able to also track and predict the spread and outbreak of extremism?

Part of the problem is that a robust formal literature on this topic does not exist, which would allow researchers to test the predictions of these models empirically. A group of European researchers have attempted to address this in a new paper, "Decision Processes of a Suicide Bomber - Integrating Economics and Psychology." Accounting for three theoretical types of attackers; posthumous effects, announcement effects and defection effects; the authors construct separate formal models to explain the rationality of committing suicide terrorism for each type. Their models mix theories from psychology, which the authors argue are necessary to show that the decision to defect from a suicide terrorism event is rare event. As they state in the conclusion:

It was shown under which circumstances the decision to commit a suicide attack can be time-inconsistent and what mechanisms might prevent time-inconsistency. Integrating the psychological concepts of cognitive dissonance and terror management theory into our economic analysis, we demonstrated why—although predicted by standard economic theory – defection is a phenomenon rarely observed. As individuals try to reduce the cognitive dissonance associated with the decision to commit a suicide attack, they select biased information that supports their decision to commit an attack.

Though interesting, the reliance on the psychological literature does not seem necessary to reach this conclusion. The most important contribution of this paper is the notion that an attacker's type matters for their utility to attacking. This leads to a potentially more interesting extension of this paper. One could define an extensive form game, in which nature decides the type of an attacker, a terrorist group then decides whether to pick that individual for a suicide attack with incomplete information regarding type; then finally, the attacker has to decide whether to carry out the attack or defect. Defining the perfect Bayesian equilibrium of such a game may be very informative, and allow future researchers to test its predictions empirically. Without getting ahead of myself, however, has anyone come across an application of game theory to terrorism?

Photo: Wikipedia

The authors' primary contributions is their recognition a fundamental difference exists between how covert networks form and operate compared to traditional overt networks; and as such, our framework for analyzing these network must match this divergence. For this papers, the authors focus on how the need to communicate and organize in secret drives this difference, but rather than test their claims empirically (in a lab, as is common), they develop a graph theoretic foundation to develop testable claims. As they state:

In this paper we analyze the problem of covert network structure design taking [the need to remain secret] explicitly into account from a theoretical perspective. The strategic interaction between a covert organization and its opposing forces is not explicitly taken into account but implicitly modeled by having secrecy as a key design parameter. We consider both secrecy and information processing efficiency as key network design parameters and we analyze several different scenarios corresponding to different assumptions on those parameters.

The authors have clearly given a lot of thought to the theoretical consequences of having to minimize the flow of resources over a network, while at the same time maintaining its effectiveness. One could spend a lot of time pondering the proposition they present, and I recommend that those interested dig deep into their ideas. I could not help but think that many (if not all) of the theories presented could be tested by simulation in an appropriately designed ABM; particularly the formation the an "optimal covert network" presented towards the end of the paper (pictured above). Without doing any testing of my own, but only having experience studying covert networks, the structure they present does not reflect how covert networks form in my experience, which leads to the obvious question: what makes this structure optimal?

Photo: ScienceDirect