The Naked Future

by Patrick Tucker

  For instance, let’s say you’re a cop looking for a robbery suspect late at night. You know the general part of town the perpetrator is in but need more information to nail down a location. Let’s say you have access to a big data set indicating tens of thousands of arrest records and you can query that database to learn the type of place that most suspects of this crime go after a robbery. It could be home, girlfriend’s house, mom’s house, bar, et cetera. Let’s say you can also bring up a map to show you all of the closed-circuit television (CCTV) cameras in the area of the incident and even which stores are open late, where you might be able to find a witness who saw something. You can further ask a network of community members and other cops to mark on a digital map where they saw something that could be useful—an article of clothing left behind by the subject, a stolen item, a sighting of someone matching the subject’s description. You now have not just one map but several that you combine to rapidly narrow down where to find your suspect and even obtain the evidence for conviction. Esri, working with police departments around the world, is putting that command center view on laptops and even phones. This is the challenge that occupies King: “How do I get that information into the officer’s hands so that he can be there at that same time?” The NYPD, in partnership with Microsoft, has also been developing these sorts of capabilities for New York City beat cops through a program called “the dashboard.”12

  To understand how that contributes to a more naked future, you have to imagine that dashboard dissemination of capability continuing, eventually, on to consumers. In the same way computers used to be the size of rooms and were available for experts to use, then became objects people could access on desktops, and are now objects in our pockets, the dissemination of this type of big police data is going to follow the same path.

  In the next ten years there’s no technological, economic, or even legal reason why every individual with a smartphone shouldn’t be able to download a live crime map showing both current and expected hot spots. Predictive policing won’t just be something that happens around you, it will be a process that you participate in directly. Information will grow much more rich and meaningful when it’s combined with other bits of local data and personal information. I asked Mike King if he saw this eventuality as likely. “It’s happening today,” he answered. “When I talk about Esri moving through mobile and other opportunities, that’s the idea of getting this water to the end of the row.”

  Ever better and more timely reporting of crimes and incidents are key to the continual improvement in predictive policing. Increasingly, that reporting is being done by bots.

  New York, Milwaukee, and nearly seventy other cities around the United States use a sensor network system called ShotSpotter, which uses acoustics to detect and pinpoint gunfire the moment the trigger is pulled. In California and along the eastern seaboard, cameras snap pictures of license plates as cars enter and leave specific areas of various cities. That’s on top of a growing CCTV infrastructure, which, in 2011, comprised more than 45 million systems around the globe. Growing by 33 percent per year, the global CCTV market is forecast to become a $3.2 billion per year industry by 2016.13 Not all of those cameras will be attached to buildings. In 2015 police departments around the country will begin testing aerial drones to establish a permanent eye in the sky in cities around the country, as authorized by H.R. 658, the FAA Modernization and Reform Act of 2012 signed into law by President Obama.14 As a result of this bill, thirty thousand unmanned aerial vehicles will be crisscrossing America by 2020, according to Todd Humphreys of the University of Texas at Austin.15

  There’s a huge industry incentive to make it seem as though the growing web of cameras, microphones, sensors, and robot planes keeping watch over us is making us safer. Unfortunately, predictive policing won’t automatically fix any of the long-term issues that plague our criminal justice system or change the way many cops interact with residents in poor neighborhoods. Zero-tolerance policies—of which predictive policing programs often serve as a component—are really effective at putting people behind bars. In a country with the highest prison population rate on the planet, that’s like taking a machine that produces a terrible product, say, exploding strollers, and “improving” it not by changing the design of the strollers but by enabling it to produce many more exploding strollers far faster and more cheaply. Even in places where every criminal is truly a threat to public health (which is no place), pumped-up arrests will exacerbate prison overcrowding, recidivism, and so forth.16

  But the most striking abuses of predictive policing programs and surveillance in general will likely soon emerge from China. China will surpass the United States as the world’s largest market for surveillance equipment by 2014, according to a report from the Homeland Security Research Corporation (HSRC). The manufacturing hub of Guangdong Province, which is near Hong Kong, boasts a $1-million-camera-security system. China is today spending more on internal security than it is on defense, but many in the West, including NATO and the U.S. DOD, claim that Chinese military funding and “public safety” funding overlap a great deal.

  Predictive policing in the wrong hands looks less like a boon to public safety and more like a totalitarian hammer. Some predictive policing tactics have already been used to stifle dissent and protest in the United States. In 2003 Miami police targeted and arrested several demonstrators prior to a major protest against the Free Trade Area of the Americas (FTAA) agreement. Today, police around the country routinely employ espionage tactics to predict and preempt spontaneous punk and dance shows (under the expansive and poorly written 2003 RAVE Act, sponsored by Joe Biden, which can be used to arrest concert promoters for the behavior of their patrons). If you’re a police chief or mayor, preempting a protest is less risky than trying to disrupt one in progress, especially in an age where the kids you will be pepper spraying carry TV studios in their pockets.

  This is bigger than busting garage punk shows, squashing Occupy marches before they take place, and shutting down raves before the speakers are even plugged in. It’s bigger than the enforcement of vaguely worded local nuisance ordinances. The same tactics that can give police advance awareness of local protest events can also be used to predict civil demonstrations, marches, and clashes halfway around the world.

  Acting locally is now visible globally.

  Seeing the Riot Through the Trees

  The date is June 30, 2012. Computer scientist Naren Ramakrishnan is in his Virginia Tech lab watching a map of the Americas on his computer screen. A band of hundreds of red dots hovers over Mexico City; another band is over the Brazil-Paraguay border. The dot cluster is ringed by concentric circles of yellow, green, and blue. It looks almost like a radiant heat map, as though the capital of Mexico and the Brazilian border town of Foz do Iguaçu are on fire, but they aren’t—at least, not yet. These dots represent geo-tagged tweets containing the terms “país,” “trabajador,” “trabaj,” “president,” and “protest.” The controversial Enrique Peña Nieto is about to be officially elected the president of Mexico and the geo-tagged tweets represent a march taking form to protest his election.

  In 2012 Nieto represented the return to power of the Partido Revolucionario Institucional (PRI). Despite the insurgent-sounding moniker, the PRI is very much the old-power party in Mexico, having governed the country for seventy-one years until 2000. It has long been associated with chronic corruption and even collusion with drug cartels. Nieto, a young, handsome, not conspicuously bright former governor of the state of México is seen by many as something of a figurehead for a murky, well-funded machine. Having met him I can attest that he can be very charming, smiles easily, and has a firm handshake. As a governor, he is best known for allowing a particularly brutal army assault on protestors in the city of San Salvador Atenco. The June 30 red-dot cluster over Mexico indicates a lit fuse around the topic of Nieto on Twitter.

  At 11:15 P.M., on July 1, as soon as the election is called for
the PRI, the student movement group Yo Soy 132 (I Am 132) will spring into action, challenging the results and accusing the PRI of fraud and voter suppression.17 The next month will be marked by massive protests, marches, clashes with police, and arrests. This is the future that these red dots on Ramakrishnan’s monitor foretell.

  The cluster in Brazil relates to a sudden rise in the use of “país,” “protest,” “empres,” “ciudad,” and “gobiern.” In a few days twenty-five hundred people will close the Friendship Bridge connecting the Brazilian city of Foz de Iguaçu to the Paraguayan Ciudad del Este, another episode in the impeachment drama of Paraguayan president Fernando Lugo.

  As soon as clusters appear on Ramakrishnan’s computer, the system automatically sends an alert to government analysts with the Intelligence Advanced Research Projects Activity (IARPA), which is funding Ramakrishnan through a program called Open Source Indicators (OSI). The program seeks to use available public data to model potential future events before they happen. Ramakrishnan and his team are one of several candidates competing for IARPA funds for further development. The different teams are evaluated monthly on the basis of what their predictions were, how much lead time the prediction provided, confidence in the prediction, and other factors.18

  The OSI program is a descendant to the intelligence practice of analyzing “chatter,” a method of surveillance that first emerged during the cold war. U.S. intelligence agents would listen in on the Soviet military communication network for clues about impending actions or troop movements. Most of this overheard talk was unremarkable but when the amount of chatter between missile silo personnel and military headquarters increased, this indicated that a big military exercise was about to get under way.19 This analysis was a purely human endeavor and a fairly straightforward one, with one enemy, one network to watch, and one set of events to watch out for.

  In the post-9/11 world, where—we are told—potential enemies are everywhere and threats are too numerous to mention, the IARPA considers any event related to “population-level changes in communication, consumption, and movement” worthy of predicting. That could include a commodity-price explosion; a civil war; a disease outbreak; the election of a fringe candidate to an allied nation’s parliament; anything that could impact either U.S. interests, security, or both. The hope is that if such events can be seen in advance, their potential impact can then be calculated, different responses can be simulated, and decision makers can then select the best action.

  What this means is that the amount of potentially useful data has grown to encompass a far greater number of signals. For U.S. intelligence personnel, Facebook, Twitter, and other social networks now serve the role that chatter served during the cold war. But as Ramakrishnan admits, Facebook probably is not where the next major national security threat is going to pop up. So intelligence actively monitors about twenty thousand blogs, RSS feeds, and other sources of information in the same way newsroom reporters constantly watch AP bulletins and listen to police scanners to catch late-breaking developments.

  In looking for potential geopolitical hot spots, researchers also watch out for many of the broken-window signals that play a role in neighborhood predictive policing, but on a global scale. The number of cars in hospital parking lots in a major city can suggest an emerging health crisis, as can a sudden jump in school absences. Even brush clearing or road building can predict an event of geopolitical consequence.

  Spend enough time on Google Maps and you can spot a war in the making.

  Between January and April 2011, a group of Harvard researchers with the George Clooney–funded Satellite Sentinel Project (SSP) used publicly available satellite images to effectively predict that the Sudanese Armed Forces (SAF) were going to stage a military invasion of the disputed area of Abyei within the coming months. The giveaway wasn’t tank or troop buildup on the border. Sudan began building wider, less flood-prone roads toward the target, the kind you would use to transport big equipment such as oil tankers. But there was no oil near where the SAF was working. “These roads indicated the intent to deploy armored units and other heavy vehicles south towards Abyei during the rainy season,” SSP researchers wrote in their final report on the incident.20 True to their prediction, the SAF began burning border villages in March and initiated a formal invasion on May 19 of that year.

  Correctly forecasting a military invasion in Africa used to be the sort of thing only a superpower could do; now it’s a semester project for Harvard students.

  Much of this data is hiding in plain sight, in reports already written and filed. In 2012 a group of British researchers applied a statistical model to the diaries and dispatches of soldiers in Afghanistan, obtained through the WikiLeaks project. They created a formula to predict future violence levels based on how troops described their firefights in their diaries. The model correctly (though retroactively) predicted an uptick in violence in 2010.21

  Simple news reports when observed on a massive scale can reveal information that isn’t explicit in any single news item. As I originally wrote for the Futurist magazine, a researcher named Kalev Leetaru was able to retroactively pinpoint the location of Osama bin Laden within a 124-mile radius of Abbottabad, Pakistan, where the terrorist leader was eventually found. He found that almost half of the news reports mentioning Bin Laden included the words “Islamabad” and “Peshawar,” two key cities in northern Pakistan. While only one news report mentioned Abbottabad (in the context of a terrorist player who had been captured there), Abbottabad is located easily within 124 miles of the two key cities. In a separate attempt to predict geopolitical events from news reports, Leetaru also used a thirty-year archive of global news put through a learning algorithm to detect “tone” in the news stories (the number of negatively charged words versus positively charged words) along fifteen hundred dimensions and ran the model on the Nautilus, a large, shared-memory supercomputer capable of running more than 8.2 trillion floating point operations per second. Leetaru’s model also retroactively predicted the social uprisings in Egypt, Tunisia, and Libya.22

  News reports, tweets, and media tone are correlated with violence. Predicting the actual cause of violence is more difficult. Yet researchers are making progress here as well. In Libya, Tunisia, and Egypt, the price of food, as measured by the food price index of the Food and Agriculture Organization of the United Nations (FAO), clearly plays a critical role in civil unrest. In 2008 an advance in this index of more than sixty base points easily preceded a number of low-intensity “food riots.” Prices collapsed and then bounced back just before the 2011 Arab Spring events in Tunisia, Libya, and Egypt.23

  If you’re a humanitarian NGO, knowing where and when civil unrest is going to strike can help you position relief resources and staff in advance. If you’re a company, you can pull your business interests out of a country where the shit’s about to hit the fan. But to law enforcement, predicting the time and place of an event of significance is less important than knowing who will be involved.

  Unlike predicting an invasion, piecing together a model of what a particular individual will do involves a lot more variables. Not only is it more challenging technically, it’s also more costly. Researchers can’t just run lab experiments on who will or won’t commit a crime, so research has to take place in the real world. But experimentation here runs up against privacy concerns. In recent years researchers have found a clever way around these thorny issues by looking toward captive audiences, individuals in situations who have effectively relinquished any expectation of privacy.

  CHAPTER 10

  Crime: Predicting the Who

  THE date is the Wednesday before Thanksgiving 2025. The setting is Dulles International Airport. Today is the busiest travel day of the year and the airport is crowded with parents dragging children dragging stuffed animals from gate to gate. But while there is no shortage of people in the airport, a single key feature distinguishes it from a similar setting as we would encounter it
today. The people aren’t standing in line. Nor are they attempting the difficult task of disrobing while moving through an X-ray machine. They aren’t carrying their belts or shoes or being patted down by TSA agents. They’re just walking to where they need to be or else waiting for their plane to board. There seems to be no security whatsoever.

  The only apparent bottleneck in the flow of traffic occurs near the entrance to the departure gates, a spot where, just a few years ago, patrons would have encountered enormous detectors of metal, gas, and powder. Instead, visitors to the future Dulles are asked to walk up to a machine, stare directly into a lens, and answer a few questions.

  The visitors approach the kiosk one after another, perform the task, and are moved quickly through . . . save one man, whose responses to the requisite questions are somehow off. The machine has not given him clearance to move forward. This man is the bottleneck. After his third attempt, two broad-shouldered TSA agents appear and stand beside the passenger.

  “I think this thing is broken,” the man informs them. The agents smile, shake their heads, take the man firmly by the elbow, and lead him away. The machine is not broken. The other passengers viewing this event don’t know where the man is being led and express no concern. They understand only that an inconvenience has been moved from their path. They will catch their flight. Because the secondary search area is used only in rare circumstances and there is no backlog, even the man who has been pulled away will not be delayed too long—assuming, of course, the system determines he poses no legitimate threat.

  An early version of the above-described program is already in place in strategically selected airports around the country (the metal detectors have not yet been moved out). The object of this screening is not luggage, exterior clothing, or even people’s bodies, but rather people’s innermost thoughts.