If the military is going to disrupt insurgent cells or understand how revolutionary movements congeal, it needs to perceive the connections between people that hide in plain sight. Naturally, the Pentagon's mad scientists are turning to Marvel Comic's one-eyed master spy for a hint.
Nick Fury isn't *really *going to teach Darpa how to understand social networks. But Darpa's repurposing his organization's moniker, SHIELD, for precisely that. Darpa's version of SHIELD, barely out of the concept phase, is called Novel Techniques for the Synthesis of High Fidelity Social Network Data, which sounds less swashbuckling than Col. Fury's old outfit. But if it succeeds, it'll help reveal out the real-life agents of HYDRA.
Basically, SHIELD is a modeling effort. It needs to produce a model of human behavior, interaction and connectivity without using much information from actual social networks like a specific terrorist group. Not many companies or families are willing to provide modelers with their information, for the obvious privacy concerns. (Hence the anonymity of SHIELD, natch.) And what few network org charts are out there might not provide sufficient data to generalize.
So when in doubt, call on small businesses to generate social network models using "synthetic data" -- facsimiles of social interaction using "appropriate social and behavioral theories" to figure out the "relational networks" between Person X and Person Y. As a trained spy, Nick Fury does that as easily as he chomps a cigar. But Darpa needs a rigorous "statistical/probabilistic description" if it's going to predict who inside a terror cell is going to rise to the top if the leadership gets killed or captured.
A model that traces ties between people isn't good enough. The networks of interest to the Department of Defense -- arms dealers, spies, terrorists, accidental revolutions -- are "dynamic," changing over time. So Darpa needs SHIELD to "explicitly take temporal variation into account." In other words, something that weights the frequency and durability of ties between people over time to provide more than a superficial sense of how they interact with each other.
Think you can build something like that for the Pentagon's biggest brains? First, bone up on your Markov chains. Then consult this cheat sheet, helpfully provided in the SHIELD solicitation: "Determine the number of users in specific subpopulations, e.g., in age ranges, locations, etc.... Determine the patterns of interaction and friendship, and which subpopulations are interacting? What is the amount and frequency of interaction? When is it occurring (time of day, day of week, month of year)?"
Once you've graphed some of those interactions, see if you can cull some of the data if it appears that the strength of relationships between certain people isn't very strong. And should any of them become stronger or weaker, can your model reflect those changes? Don't expect any help with the data itself, by the way. Darpa's solicitation warns, "the data sets used are the sole responsibility of the performers – the government will not supply any data."
It's a tall order. Understanding how actual people interact -- with their subtleties, concealed intentions and misunderstandings -- is a life's work. Models for understanding the structure of Sim Insurgency and how it behaves carry the risk of becoming unrealistic by being uber-rational.
But models that can accurately reveal dynamic social networks carry a massive benefit for the military and intelligence agencies, which struggle to understand how civilian social structures feed into insurgent groups. Not even Nick Fury can suss out the hidden connections imperiling mankind all the time.
*Image: Marvel Comics, via beyondhollywood.com
*
See Also:
