by P. W. Singer
“Yeah, that’s not going to be possible,” he replied. “Student government voted a campus-wide policy of no face rec sharing with law enforcement.1 The cameras here collect it for the patient service and staff access authorization, but after that all the faces get fuzzed over. A comp sci student updates the program each year for their senior thesis.”
“You gotta be kidding me,” Keegan fumed. “Can your IT folks at least share over the pixilated feed?”
“Yeah, we can get them to do that.”
“Thanks,” she said.
He looked down and spoke into a mic mounted on his vest. “Wait a second . . .” the cop said.
A cold blast of adrenaline shot down Keegan’s spine. In getting the feed from the campus security, had they just tripped a notice to main FBI?
“I know you,” he said. “The bot too. You’re the one from that rooftop perp chase, the subway, and all that.”
“Yeah, that’s us,” Keegan said.
“No wonder you’re all that they sent,” he said enthusiastically. “That video of the beatdown you did on those dudes at the Mall . . . Ruthless! How did it feel to lay someone out like—”
“Agent Keegan,” said TAMS. “There is an urgent message for you.”
She turned to the machine, thankful for the interruption, but wondering why it was notifying her verbally when it could have popped it on her vizglasses.
“I am sorry, officer,” said TAMS. “It is urgent and requires privacy.”
“No worries at all,” the guard said. “You can go on through. I’m posted down here for the next six hours. Your guy won’t get past me.” Then he looked at TAMS and smiled. “Just don’t go beating up too many people . . . without me.”
“OK,” TAMS said. The robot turned and walked to the far end of the area, toward an alcove that, for the moment, was only occupied with an empty patient’s bed. “I have obtained network access to the hospital video surveillance system,” said TAMS.
“Why didn’t you just push the info to my lens?” Keegan asked.
“It appeared that you required an exit from the conversation,” TAMS said.
It was right. And it also meant that the CIA had provided its TAMS with a few added protocols, including the ability to deceive. Such a feature made perfect sense for an intelligence-community machine; hell, it made sense for any social interaction with humans.2 But she wasn’t sure how comfortable she was with a robot that could tell even a little white lie.
“New command protocol,” she instructed. “When on operations with me, do not speak false information unless I specifically authorize it.”
“OK,” TAMS said.
But now that she knew the machine could tell a lie, how could she be certain its response was the truth? She buried that riddle. Now wasn’t the time.
“Display what you have,” she said, and multiple windows of streaming video popped up in her vizglasses, showing views of the lobby, hallways, parking garage, and various departments. In each, hundreds of evident hospital staff, patients, and visitors moved back and forth, their faces all with the same smear of pixilation.
“Anything you can do about the anonymizer?” she asked, hopeful that TAMS’s neural networks would be able to find the underlying patterns in the obscuring mosaics of digital noise.3
“No,” the machine replied. “It has been designed with interception in mind, using a combination of three-stage quantum cryptography and blockchain principles.4 Even if decryption were possible, any change of the pixilation will result in a change of the underlying image itself.”
“I guess they got an A on their senior thesis then,” Keegan muttered to herself.
“Yes,” TAMS replied.
She shook her head at the sheer amount of knowledge it had access to and instructed, “Show me the schematics of the hospital and crossmatch any activity from the cloud that connects to the NICU.”
A 3-D rendering of the hospital structure built on her screen. A cluster of the anonymized faces in the third floor began to be overlaid with data points. Different colored tags marked everything from an individual who had NICU work history on their online resume to someone wearing a Watchlet that posted social media images geo-tagged to the NICU’s location.
“We need to winnow it down,” Keegan said. “Pull out anyone who’s not tagged and doesn’t fit any personal identifying features we know of Todd.”
At a sweep, scores of figures were erased from the screen, every pixilated human body without a data point disappearing. Then, wave after wave of deletions played out in more staggered form. Mothers-to-be disappeared like apparitions, but the children they held, and those lying in protected warmers, remained. After a few seconds, the children’s figures evaporated as well. Within a few seconds, all that was left after the data-driven rapture was a hospital of just under a hundred men.
“It’s still too many,” said Keegan. “Is there anything that ties them together?”
Seconds passed before TAMS spoke. “Of the remaining possible suspects, there are two main clusters with limited overlap. The first is the use of password-protected network stations or other secured systems.”
“Hospital staff. Pull out anyone who’s made multiple entries before tonight.”
Roughly half the bodies disappeared from the screen.
“The second cluster has made phone calls from the hospital during the past sixty minutes. Based on direct-access database analysis, they have been contacting individuals with whom they have tightly meshed social connections, close family and friends.”
“New dads,” Keegan said. “Delete them.”
At once, the remaining bodies faded from the image. All that was left was a single man in a doctor’s coat, his face obscured by a mosaic of black, gray, and white dots, leaving the NICU.
Georgetown University Hospital
Washington, DC
The choice was seemingly a binary one:
Save the children. Stop a mass murderer. Which was more important?
Keegan weighed the outcomes over mere microseconds, infinitesimal calculations factoring in everything from the certainty of the number of children in the NICU to the untold catastrophes that might follow if Todd went free. But then cutting through it all was the realization that she was no mere calculating machine. It didn’t have to be a binary choice, if she was willing to give to a machine that most human gift of all. Not trust that TAMS would do what she expected, but trust that it would do the right thing on its own.
“I’ll follow Todd and you go to the NICU,” she commanded. If they had to split up, she had to admit to herself that TAMS was better suited to figure out whatever Todd had done there.
She reached out to grasp the robot by the shoulders. She could only hope it had learned enough to register that signal as something more than mere tactile pressure. “TAMS, stop whatever attack Todd has planned against those kids. I’m trusting you to save them.”
“OK.”
It bolted through the Emergency Department waiting room, followed by surprised shouts and cries of anger as it weaved between patients and staff.
Keegan pulled up directions that marked the quickest route in the building to cross paths with Todd. Following the line of small green arrows projected in her field of view, she sprinted down the hallway.
As she ran, she tried to check in on TAMS, blinking open onto her viz a POV shot from the machine’s forward-facing cameras. All it showed was a disorienting picture of jarring motion, maybe a stairwell, maybe another hallway. Before she could shift back to the 3-D building map to mark its position, she stumbled, tripping over a wheelchair folded up against the wall that she’d missed in her focus on the machine’s mission, and not her own. She struggled to stay upright as the impact jarred her spine and her leg took the weight of the fall at an odd angle. Numbness started to creep up her leg, and she grabbed the handrail that ran along the hallway to steady herself for a moment.
As Keegan took off again at a slower jog, she double-blinked to close
the window on her vizglasses. It was time to take the training wheels off. Whatever TAMS did, it’d have to do it without her watching.
For TAMS, it was now as much a race through the digital world as it was through the hospital itself.
As the machine ran toward the NICU, its physical sensors detected obstacles and projected optimal pathways freed from human limitations. Its feet sprang off positions on the floor selected for their greater traction by minute reflections of light off the tile, while its fingers reached out to press against food trolleys, gurneys, and water fountains, anything to give extra speed.
Rounding a corner, the bot’s camera detected the foot of a bed being wheeled into the hallway by a knee-high orderly bot and determined that its speed and direction would intersect with its own path.1 Rather than stopping, TAMS leaped, one foot extending onto the hallway railing, the other pivoting to step onto the wall itself. As it passed over the gurney, literally running up the side of the wall to keep its maximum pace, the patient below began to wonder about the side effects of her pain meds.
At the very same moment of its physical movements forward to its target, TAMS’s analytic systems were moving backward through the video trail of Todd. Layers of neural networks deconstructed the footage of the pixilated man’s movements through the hospital, battling away to identify any anomalous behavior compared with that of all the other human-sized objects in the NICU’s video history. In less than two seconds, it identified a marker moment from just before it identified Todd. Imagery showed the blurred man pulling a small tablet computer out of his backpack. He typed on it for a few moments, inserted a cable, and then slid it into one of the bedside monitors next to one of the human-sized objects, digitally identified by a barcode tag on their left ankle.
Running profile matches to Todd’s prior attacks, TAMS’s network analysis determined that the imagery time stamp matched that of a software download into the hospital control systems managing the facility’s liquid oxygen supply. As it ran calculations of how long it would take for the download to complete and for the software’s embedded instructions to force an explosion of the highly combustible gas, the emotional burden of the prospect of the entire neonatal wing of the hospital—every newborn baby, parent, and staff member—being consumed in a fiery blast did not factor into its calculations. Yet now TAMS understood the stakes in a different way: the specific outcome that had to be prevented in order to complete Keegan’s command.
When it entered the waiting room, TAMS pulled up short before the NICU ward entrance. A few parents started to murmur at the unaccompanied machine’s sudden arrival. It looked nothing like the hospital bots, designed to appear docile and approachable. The murmurs grew louder when its lower legs extended an entire foot to give the machine an elevated position to see their children through the viewing window.
Now able to assess the NICU with its own sensors, TAMS confirmed Todd was already gone. It pushed the information to Keegan but judged any other action on that pathway of the decision tree to be of lesser priority to its ordered task. The robot then ran multiple simulations on the expected time of standard human pathways and found none that allowed it to meet its overall goal. Punching its left fist forward, TAMS shattered the viewing window. Parents screamed as glass rained down. Their shouts intensified as the robot vaulted through the broken frame into the NICU.
Ignoring the screams of the families in the waiting room behind it, the robot stalked between the incubators. Its audio sensor picked up a baby at the far end of the room crying with the desperate and breathless high-pitched mewing of helplessness. TAMS leapt over one warmer and loomed over the plastic incubator. Inside was a small boy, wearing a barcode bracelet around his ankle. The unique design of thin and thick black lines matched what TAMS had observed in the video footage, identifying the human as “Mark Rezak,” two months old, born premature at twenty-three weeks, weighing 658 grams.
As TAMS lifted the lid on the warmer, an adult ran into the room. She wore clothing that corresponded to the records of hospital staff and a name tag that read “Karina.” This matched a “Karina Eggers” in the hospital files, who was assigned as a nurse to the NICU, which was soon confirmed by a facial recognition match with her Virginia driver’s license and several online dating sites.
“No!” screamed the human identified as Karina Eggers, as TAMS scoured the cloud for any and all information on her, pulling down everything from job history to her complaints on social media about a last date gone bad. It determined she was not authorized to issue command orders to deviate from Keegan’s instructions.
TAMS dropped the incubator lid, which hit the floor with a clatter, causing the baby Mark Rezak to cry at a 23 percent higher volume. With its right hand, TAMS extended its fingers into a cradle-like claw and lifted the infant from its warmer. A tangle of blue and white wires hung down from the child, back into the incubator’s monitors.
Following the nurse, another human entered the room. Facial recognition matched her to a driver’s license for Sheryl Root, age forty-four, of Silver Spring, Maryland, as well as the state registry for a personal weapon. This was further confirmed when the gun that she was registered for in the state database matched the outline of the weapon that TAMS detected in the human’s hand, a Smith & Wesson M&P 9 with a laser sight mounted under the barrel.
“Stop right there!” Root shouted, raising the gun. While TAMS registered the text “SECURITY” on the front of the second human adult’s black jacket, neither Sheryl Root’s uniform or online history empowered her as an official agent of law enforcement, able to command its operations. TAMS then judged the threat from the observed weapon. Self-preservation was determined to be secondary to its mission. It continued with its prior decided course of action, but with a slight modulation of a pivot of its torso, so as to place itself between the projected bullet trajectory and the loudly crying human child.
The shot now clear, the security guard laid her finger down on the trigger. A tiny red dot from the pistol’s laser sight appeared on what would be the face of the robot, the red light splashing slightly across the smooth surface.
The guard’s finger began to press down on the trigger, taking a breath to hold the dot steady. In that one moment of stillness, she heard the sound of a woman’s voice. It had the lilting cadence of a mother singing a lullaby.
You are my sunshine, my only sunshine
You make me happy when skies are gray2
As the voice sang from its onboard speakers, the robot’s hand slowly began to rock the baby in a swaying motion. With its other arm, it reached into the warmer and pulled out a blue cable connecting to a tablet.
“Just . . . stop right there,” said the guard, not knowing what else to say.
The nurse stepped in front of the red laser target and slowly walked toward TAMS. “I can take him,” she said, her arms out.
“OK,” TAMS said, responding in a digital voice that was discordant from the music. The nurse froze for a moment, but then took the baby from the machine and wrapped it in a warming blanket.
With Todd’s command code prevented from being completely downloaded, TAMS assessed that Keegan’s primary task instruction had been fulfilled. It searched for Todd’s present location and found him on the rooftop of the hospital, detecting that Keegan was in the stairwell, climbing toward the same position. Without another word, TAMS turned to join them, vaulting back through the broken glass. As the robot ran down the hallway, however, it continued to play the four-year-old clip that it had pulled from the cloud of Lara Keegan singing to her baby daughter, reflecting its new learned behavior that lullabies apparently also calmed human adults.
Georgetown University Hospital
Washington, DC
Keegan climbed up to stand on the stairwell’s handrail running next to the door to the hospital roof. With one hand holding the exit sign that hung down, she drew her Sig pistol with the other. Reaching out, she carefully tapped the overhead light with the butt of the pistol. It wa
s a delicate blow, like cracking an egg, enough to be almost silent, but with it, the stairwell was left in complete darkness. When she opened the door, Keegan did not want to present an easy-to-target silhouette.
She climbed down, and after marking to memory the door’s location in the infrared projected layout of the hallway, she took off the expensive vizglasses from the Range Rover. She then placed them on the concrete steps. Feeling for them in the dark with the tip of her shoe, she stepped down as hard as she could, snapping the frames and grinding down the lenses with her heels.
It was another precaution that could make all the difference. She didn’t know who to trust anymore, but had to assume Shaw was watching.
Then she chambered a round into her Sig Sauer. A long exhale, and she delicately shouldered open the door to the roof. As she emerged into the night, a part of her brain still found a way to marvel at the stars overhead.
The rooftop was a forest of exhaust pipes, microsatellite dishes, articulating delivery drone pads on long stalk-like pedestals, and the glimmering aluminum fins of the heat sinks that passively cooled the hospital’s network hardware.1 Crouching low, Keegan moved slowly along the perimeter of the roof, rather than trying to weave through that jungle. Seven stories up, a light breeze could be felt. Overhead, a brightly blinking air ambulance cut through the night with a growl as it flared down to a hover. Keegan watched it warily, wondering if it was going to crash, another one of Todd’s attacks, or simply deliver patients to the automated orderly cart that waited below. Her question was answered as it came to a hover and then dropped out of view, landing on a pad on the next building’s roof.
She used the sound to advance quicker, turning the corner of the roof. There. Roughly 30 paces ahead.
A man in a doctor’s coat sat on the low wall that ran along the edge of the rooftop. His legs dangled over the side of the building, as if he were taking in the view. An open backpack lay nearby.