Kill Chain: The Rise of the High-Tech Assassins (27 page)

The Distributed Common Ground System headquarters at Langley is not an institution comfortable with outside scrutiny. Throughout my visit, I was an alien presence. Touring the facility with a squad of escorts, my little phalanx was preceded at all times by a serviceman holding high a red torch to signify that an interloper from outside the classified universe was in their midst and therefore that nothing secret should be shown or uttered. The huge rooms, lined with workspaces crammed with multiple screens, resembled nothing so much as a Wall Street trading room, except that here the screens display images of Afghan hillsides collected by a Reaper drone or a patch of the Pacific ocean sent by a Global Hawk drone at 60,000 feet or the horn of Africa from an orbiting satellite or the coast of Iran from a U-2 spy plane, any and all of which, in theory at least, could be called up by the young men and women consigned to gaze at the screens day after day. They may be coordinating a strike, archiving video so that it can be called up later, or simply gazing at a targeted house, car, or person far away. Larger screens high on the walls displayed the location of “platforms,” the drones and planes and satellites gathering the knowledge being piped into the system in a ceaseless torrent, the equivalent, so the air force informed me, of 700 copies of the
Encyclopedia Britannica
per day.

It is a system of extraordinary complexity and expense. Simply to ensure that all the different nodes and sites remain interconnected at all times involves a massive investment in bandwidth and fiberoptic communications. Even rendering the incoming imagery viewable requires a considerable engineering effort to “clean up” what may arrive as unintelligible images.

Thus, ever since William Perry championed the privatization of defense operations and support functions in the 1990s, outsourcing key missions to civilian contractors has taken up an ever-increasing share of military operations and budgets. This makes it legally difficult, given the sanctity of contracts and corporate litigiousness, to cut spending in this area. The esoteric world of “D-Sigs” is no exception. Just as private contractors handle drone takeoffs and landings before handing them over to the military crews to conduct actual strikes, so corporations not only built this complex electronic nervous system but also to a considerable extent operate and maintain it.

A simple check on Internet job postings from corporations on contract to service the system helps to convey the scale of the business. Openings at just the Langley node, for example, were appearing daily, with no sign of a slowdown even as Washington rang with talk of austerity and “a hollowed out military.” A typical day’s sample in early March 2014 advertised openings for, variously, a “systems administrator” (the position that Edward Snowden put to good use) required by CACI International, a “subject matter expert” sought by Sehike Consulting, an “intelligence capabilities analyst” required by Digital Management, while General Dynamics was looking for a network engineer. All positions required at least a Top Secret Clearance, and most mandated SCI (special compartmented information), which usually meant signals intelligence. Salaries ranged between $120,000 and $170,000 annually, though of course the contractors would be adding a hefty overhead when submitting bills to the taxpayer.

The Langley unit has a resident chaplain and a psychologist who patrol the area between the computer banks ready to offer counseling to anyone unduly distressed by scenes of remote death and destruction in which they may have participated via video. But though much has been made of the combat stress suffered by servicemen and women in such conditions, Colonel Hernando Ortega, surgeon for the Air Force Intelligence Surveillance and Reconnaissance Agency, has expressed a different view of the stresses of what he calls “tele-warfare.” Addressing a Washington think-tank seminar in 2013, he divulged the fact that “our guys are below the general civilian population as far as risk for PTSD.” To murmurs of disbelief from his audience of defense intellectuals he insisted that “we haven’t had any pilots with PTSD.… We had, I think, one sensor operator, maybe.”

One side of the huge room at Langley was reserved for the signals intelligence workstations where NSA assignees monitored the cell-phone locations of targets, eavesdropped on their conversations, or checked links based on calling patterns, all crucial elements in Deptula’s drive to fulfill his thirst for “knowledge.” They enabled the marriage of visual and signals intelligence, so that a specific SUV in a wedding-party convoy can be pinpointed thanks to signals from the high-value target’s cell phone and then incinerated with a Hellfire missile. This does indeed constitute an impressive feat of technological intelligence, assuming, of course, that they have the correct phone number.

“I overhauled the system,” Deptula told me, “made it global, so that any station could be involved in any operation with a phone call.” The scale was impressive—the whole world enclosed in one thinking spiderweb—even if the end result was a tsunami, a thousand hours
per day
of full-motion video (defined as 24 frames per second) plus further streams of intercepted calls and associated signals intelligence that the current 5,000-strong complement of air force Distributed Common Ground System analysts will never have time to review.

Lamenting the quantity of information collected, whether images or signals, is a traditional meme of intelligence officials. Deptula, for example, liked to warn that “we will soon be swimming in sensors and drowning in data.” However, neither he nor any of his fellow generals has ever suggested that the answer might be to collect less. As General James R. Clapper, future director of national intelligence, once said, “I cannot see a situation where someone is going to say, ‘Hey, I can do with less of that.’” This statement suggests that perhaps the object of the exercise, wittingly or otherwise, is not the production of useful information but simply the building of a bigger bureaucratic empire with a bigger budget. Clearly, however, no one could admit to this. Therefore the orthodox response to the “drowning-in-data” lament has been to invoke the urgent need and imminent prospect of turning the business of analysis over to machines. As Deptula himself has declared, “making this automatic is an absolute must.”

This is the Holy Grail, pursued ever since the distant days of Task Force Alpha, with ever-more participants joining the chase. Exponential expansion in processing power and software improvements has certainly made it easier to extract relevant items from a mass of data and “connect the dots,” as the hackneyed phrase has it. Yet this program embodies a mechanistic approach to warfare, as in the 1941 air-war plan that projected the defeat of Germany with the destruction of a set number of targets, or Warden’s “five rings” concept for defeating Iraq in 1991. Lack of information leading to misidentification of truly critical nodes has been routinely blamed for the long string of post-1941 failures of this target-list approach to war.

This mind-set has been extremely beneficial for the various interested parties, including most recently the contractors who are building and servicing the air force DCGS (Distributed Common Ground Systems) as well as those (largely the same cast) who are creating its army and navy counterparts, DCGS-A and DCGS-N. While the air force version primarily serves to assist in the execution of airstrikes, the $2.3 billion army system is supposed to help soldiers on the ground assess present and future threats, such as the names, faces, relationships of known enemies, favored sites for planting IEDs, and so on. DCGS-A attracts a great deal of well-merited abuse from a host of critics, in and out of uniform, who attest to the difficulty of using it and its frequent breakdowns.

Many of these critics, including soldiers in the field, swear with equal vehemence to the merits of the data-analysis system offered by Palantir, a Silicon Valley corporation with origins in the PayPal fraud-detection division. Much of its appeal derives from its ease of use: “It’s a database with an Apple-ish interface,” one contractor in the automated intelligence business told me, “and they’re really good at selling themselves.” Palantir, like DCGS, is an intelligence fusion system but one that has applications across a wide range of fields, from Wall Street to disaster relief. “Palantir works because it’s a commercial system, constantly refined,” one longtime Pentagon consultant explained to me. “The army system is produced by a bureaucracy that works in partnership with the usual suspects, Northrop, Raytheon, General Dynamics, Lockheed, and they put together something horribly complicated, unwieldy, and expensive.”

In its early years Palantir worked exclusively for the CIA, which had initially funded it through its In-Q-Tel start-up arm. Basically, the system categorizes information in an easily readable form, making use of whatever disparate databases the user can access and wants to use. So, for example, it could display a list of cities in Libya ranked according to the number of suicide bombers they have produced or in graphic form to show maps of where the most productive cities are. An army officer in Afghanistan sent me a Palantir-generated color-coded display, showing which senior officers in the Afghan unit he was advising had good relations with senior officials in the Kabul government and which did not. Bolstered by intelligence agency accolades for its undisclosed feats in tracking al-Qaeda terrorists, Palantir has expanded its market to Special Forces, law enforcement, and JP Morgan, where it detects mortgage fraud (by outsiders). I queried a marine friend serving in an isolated Afghan outpost about its alleged ability, as one enthusiastic congressman put it, to “detect IEDs.” He responded that it is indeed a “great system … since the enemy in this part of the world is habitual, you can ‘predict’ where possible and likely IEDs will be, based on historical trends.”

This is clearly an eminently useful function, as long as enemy habits don’t suddenly change and as long as the hard data, in this case records of previous IED attacks, have been correctly entered. There is no reason to suppose this would not be the case; bombs, especially when they kill and maim, are easy to define and record. But even at the simplest level, the automation optimistically foreseen by Deptula and others represents a more ambitious goal: a system that will “extract insight from information” (as the Palantir website neatly defines
analysis
) from sensors in real time, painlessly delivering prepackaged analyses to “customers” for further action as desired.

Ruminations on the problems of managing enormous quantities of surveillance data and the exciting possibilities for analyzing it automatically are a frequent topic on those occasions when inmates of the military-intelligence-industrial complex meet and confer, such as at the annual GeoInt convention in Florida (dubbed by insiders “the intelligence community’s spring break”). Less attention is paid to more mundane matters, such as the shortcomings of existing systems for reasons of technical unfeasibility, incompetence, or greed. The MQ9 Reaper drone, for example, General Atomics’ successor to the Predator, was introduced into service in 2005 and is now the backbone of the drone fleet. “They developed and fielded it in a hurry,” an official in the Office of the Secretary of Defense explained to me. “Jumper [the air force chief of staff who promoted Predator] was retiring. He was an enthusiast for UAVs, but they knew Moseley, the incoming chief, was not so keen, so they pushed it on Jumper’s watch.” Larger and heavier than the Predator and capable of carrying more weapons (including 500-pound bombs), Reaper is extremely expensive to buy (more than $30 million a copy) and maintain ($5 million per year), much greater than older manned combat planes such as the F-16 and A-10. Though advertised as capable of patrolling for up to 30 hours, it manages less than half that when carrying its limited load of armaments, and it crashes at least twice as often as F-15 and F-16 manned fighters.

Nor should it be assumed that the Reaper is better equipped than Predator to survey the ground beneath and thus avoid confusing women and children with “military-aged males,” a problem known in the video-surveillance world as “slants.” In fact, it carries essentially the same sensors as the Predator that killed a marine, staff sergeant Jeremy Smith, and a navy corpsman, [medic] Benjamin Rast, in Upper Sangin, Helmand Province, Afghanistan, on April 6, 2011, because it could not distinguish their distinctive helmeted and armored profiles from turbaned Taliban. The engagement was monitored by a DCGS substation in Terre Haute, Indiana, manned by the Indiana Air National Guard, where an “imagery supervisor” concluded that the two men in the process of being targeted by the Predator appeared to be shooting
away
from fellow marines, indicating they were friendlies. He communicated this to the “tactical communicator,” who passed the information on to the Predator base in Nevada, but the message was received by two mission intelligence coordinators who somehow failed to communicate this important news to the Predator pilot and sensor operator sitting a few feet away. They were looking at exactly the same video but were convinced that the muzzle flashes indicated the men were shooting
toward
the marines, hardly a testament to the much-touted resolution of drone videos. Furthermore, as the pilot subsequently told investigators, he thought the targeted men were enemy partly because, on infrared, their images tended to be “much hotter than friendly forces.” The subsequent investigation concluded that no one was to blame. Nor did anyone seriously question the baroque complexity of these arrangements, with “imagery supervisors” trying (via a “tactical communicator”) to get a vital message to “mission intelligence coordinators” (one of them a trainee) so that they could tell a “pilot” that he was misreading a murky image of a confused firefight eight thousand miles away or suggest that perhaps a live pilot in a plane overhead in direct communication with the ground force might have done a better job.

Unsurprisingly, the Association of Unmanned Vehicle Systems International, the increasingly potent drone lobby (“Advancing the unmanned systems and robotics community through education, advocacy and leadership,” according to its mission statement) does not like to dwell on such mishaps. Now boasting 7,500 members and a board of directors well larded with defense aerospace industry stalwarts, the group lobbies for all things drone, especially their freedom to share the skies of America with traditional civilian air traffic. Among other successes, the group, by its own account, literally wrote a 2011 law mandating the Federal Aviation Authority to allow this. For the industry, law enforcement is a promising market, and so Congress obligingly appropriated money to the U.S. Border Patrol to buy six Reapers. Reviewing their operation, the Government Accountability Office found that as of 2011 these Reapers had enabled the capture of 5,103 undocumented aliens and drug smugglers at a cost of $7,054 per captive. However, some subversive border control official had arranged to rent a Cessna light aircraft that did not require a huge support team (171 people for a Reaper patrol) and equipped it with a simple infrared sensor. Performing the same duties as the Reapers, the Cessna operation yielded at least 6,500 captives at a cost of only $230 per person, 3 percent of the Reaper tally. Needless to say, the experiment was not repeated, and Congress soon appropriated more Reapers to guard the frontier.