SIGNALS INTELLIGENCE DIRECTORATE
NSA
FORT MEADE, MARYLAND
Samantha Stout sat at her desk. A wall of large computer screens—eight in all—faced her. Each screen looked as if a child had sat down and started typing random letters and numbers into it, except for one screen that had a tile of news feeds from various parts of the world running with the volume down.
It was past midnight.
Samantha was the individual inside the NSA who was responsible for testing and integrating the new software program that had been developed by DARPA. DARPA stood for Defense Advanced Research Projects Agency, an agency inside the Department of Defense responsible for the development of cutting-edge technologies for use by the military. The program, called Rolex, was raw but potentially very powerful. DARPA had developed a way to capture certain nonintelligent electronic signals and convert them into symbols that could then be translated into words. In other words, it could grab pure data streaming through the sky from pre-internet forms of technology and convert it into words. It was Samantha who’d figured out how to aim Rolex at North Korea and specifically certain electronic frequencies that the NSA knew were being generated by the military. By writing an algorithm that was layered on top of Rolex’s algorithms, Stout had trained Rolex to be able to know what the signals pattern looked like when the North Koreans launched missiles. The problem was, Rolex only worked with historical data. It was an armchair quarterback, able to pinpoint when a launch had taken place. Her challenge: get Rolex to parse, assess, and react in real time. Rolex needed to be predictive versus reactive in order to know when the order to launch by the North Koreans had taken place—and therefore how long the U.S. had before the missiles actually left the ground.
Samantha called a classmate of hers from the California Institute of Technology, a woman named Kami Gray who worked at a hedge fund in New York City. Kami, she knew, wrote algorithms that the giant hedge fund used to look at patterns in the stock market and then predict what was going to happen next based on those patterns. Kami had already done with different software what Samantha needed to do with Rolex: read terabytes of real-time data in nanoseconds in order to predict the future.
Without telling Kami any classified information, she explained the challenge.
“You need to be able to know the pattern in the first few seconds,” said Kami. “You need to slice off the first few seconds of the signals events you’ve already catalogued, then isolate them versus all other signals.”
“Exactly.”
“I’m going to send you an access key to an algorithm I wrote that enables us to see certain patterns in commodities prices and then drive a trade. It’s designed to react to the pattern within a fraction of a second. If the pattern is the one we trained it to watch out for, our computers automatically start buying or selling, as the case may be.”
“Sounds perfect,” said Samantha.
“Do you know your way around MATLAB?” said Kami, referring to the computing environment her algorithm had been built in.
“Yes,” said Samantha.
“Obviously don’t tell anyone.”
“I owe you dinner next time you’re down here.”