As wrong a choice as Poindexter was to lead this project, and as tone deaf as he may have been in its presentation, he had the right idea. He had been thinking about it for decades. One of the computer’s great contributions—this ability to store and manipulate vast amounts of data—seemed mundane but was in practice so revolutionary that it was transforming modern life, whether performing a Google search, stocking the shelves at a Walmart from an international supply chain, shipping packages anywhere in the world overnight, or mapping the human genome. So why not put that capacity to work tracing a terrorist network—recognizing clues in what would appear, even to teams of skilled analysts, to be random events?
Poindexter’s concept did more than survive. It would come to undergird the entire war effort: storing every scrap of intel about al Qaeda and related groups gathered by the nation’s very active military and spy agencies, transforming them into data, and then plumbing that data for leads. The hunt for bin Laden and others eventually drew on an unfathomably rich database, accessible to anyone in the world with the proper security clearance, whether a marine officer at an outpost in Afghanistan or a team of analysts working in Langley. Sifting through it required software capable of ranging deep and fast and with keen discernment—a problem the government itself proved less effective at solving than were teams of young software engineers in Silicon Valley. A start-up called Palantir, for instance, came up with a program that elegantly accomplished what TIA had set out to do. Founded in 2004 by Alex Karp and Peter Thiel—the latter is the billionaire cocreator of Paypal and an early Facebook investor—Palantir developed a product that actually deserves the popular designation Killer App. Newly minted software engineers from the best computer schools in the country were put up in a seven-thousand-square-foot workspace in Palo Alto. It was stocked with junk food and video games and nicknamed “the Shire,” the home of the Hobbits in Tolkien’s Lord of the Rings. (The company itself is named after a magical stone in the Tolkien saga that confers special powers of sight and communication.) The software produced from this very unlikely source would help turn America’s special forces into deadly effective hunters. Palantir is now worth billions, and has contracts with, among others, the CIA, the NSA, the Defense Department, the Defense Intelligence Agency, the FBI, the National Counterterrorism Center, and the Department of Homeland Security.
The pace and urgency of war have always accelerated the development of technology and encouraged novel uses of devices that already exist. After rapid initial success toppling the regime of Saddam Hussein in Iraq, American forces found themselves under increasing attack by Sunni extremist groups, the most violent of which was a new branch of al Qaeda, under the direction of an innovative killer named Abu Musab al-Zarqawi. His group mounted a campaign of roadside bombs and brutal suicide attacks, many of them designed to kill Iraqi civilians indiscriminately—the sort of attacks that bin Laden, in hiding, considered mistakes. Indeed, the mass killings eventually helped turn the Sunni majority in Iraq against the insurgency, marking the turning point in the war. But at the same time, under the direction of General Stanley McChrystal, JSOC was hammering away on insurgent cells of the local al Qaeda killers with increasing effectiveness, mounting mission after mission in rapid succession, capturing and killing at a pace that such operations had never before been able to sustain. They found Saddam Hussein hiding in a hole in the ground in late 2003. Zarqawi himself was killed by an American bomb in 2005. McChrystal’s success, considered to be one of the major military accomplishments of modern times, was something he called “collaborative operations,” by which he meant the fusion of “special operators”—teams of elite shooters from every branch of the service—with this new computational ability, which amassed data from all of the other inputs. The task force built a massive database at Camp Victory in Iraq, and then another at Bagram in Afghanistan, blending the big picture with the small. It meant bringing a different kind of warrior to the front, one more accustomed to clicking a mouse than pulling the trigger.
Guy Filippelli was one of them. A young army captain, a West Point graduate with a master’s degree from Oxford, in 2005 he was asked by his commander in Afghanistan to visit the walled-off facilities of the task force—the special ops unit—and show them what he could do with his computer. Filippelli calls himself a geek. He had started writing computer programs as a high school student before heading to West Point’s growing computer science department. He was helping the command staff at Bagram design systems to better control “information flow,” plugging intel collected from the sites of raids in the field and from the interrogations of detainees into a growing national terror database. He arrived inside the cloistered walls of the task force full of enthusiasm for his work, certain his lecture would excite these frontline troops. The shooters and their staff could not have been less impressed. Filippelli’s subject matter was highly technical and abstract, cutting edge, and very cool to him, but he was talking to a roomful of soldiers whose adrenaline rush came from… free falling from high altitudes or getting shot at. Their world was the extreme opposite of virtual. So the next time the young captain got a chance, this time with a smaller group of soldiers, he tried a different tack.
“Listen, I know you guys are a thousand times better at this stuff than I am and are probably already doing all this, but let me show you what I’m doing and I’ll be out of your hair in ten minutes.”
At first it was something easy. The task force was used to simply locking up suspects in the detainment facility as they awaited questioning. Filippelli had built a database for detainees, and had also mapped the facility’s population by tribal affiliation, background, kinship, and other factors. Putting a detainee in the wrong place, for instance, with a group from his own village, meant that his comrades would rapidly coach him. Filippelli could show how those poorly placed were significantly less useful afterward in interrogation. So where you put them in the facility was important.
“Look,” he said. “You’ve picked up this guy. Why did you put him with these guys? You could have done this…”
And with that, he closed his laptop and started for the door.
“Thanks for your time,” he said. “Let me know if there’s anything else I can do for you.”
“Wait,” the men protested. “Tell us a little more about this.”
Gradually, he found himself working more and more with the task force, showing them how crunching data could vastly improve their efficiency. The applications went way beyond storing detainees. The name of the game in warfare is to learn faster and act faster than the enemy. So, as Filippelli and others doing the same kind of work came to see it, the contest had to do with time cycles. If it’s a detainee who could be held for, say, only twenty-four hours, how do I use that time most efficiently? What questions should he be asked? What do I need to learn in order to ask him the best questions in the time allotted? And that was just one piece of the puzzle. Looking at the larger mission, the special ops teams needed to get inside the information cycle of their enemy. In the past, after a successful night raid where a member of an insurgent cell was killed or arrested, by morning, or even within a few hours, every critical member of that group would know about it and would have taken evasive action. Information spread quickly. Cell phones would be ditched, computer discs destroyed, bomb-making facilities moved—the bad guys would scatter. But if you could get inside that response time—if you could beat their information cycle and learn enough from the first raid through either interrogation or, say, scrutinizing a seized cell phone or hard drive—you might be able to launch a new raid or even multiple raids before word of the first one had gotten out.