Will an AI-powered robocop keep New York's busiest subway station safe?
The New York Police Department's K5 robot cop roams the mezzanine of the Times Square - 42nd St. subway station. A broken monitor with indecipherable subway arrival times hangs above the robot, which is being rented for $9 an hour.
I met New York's subway robot cop on a temperate November Monday at midnight. I found the robot known as K5 patrolling the mostly empty mezzanine of the Times Square-42nd St. subway station, pacing from one end of the corridor to the other, pausing, like a cautious Roomba, as riders passed by.
In late September, the New York Police Department deployed a 5'2, 398-pound, hunk of metal in the city's busiest and most tourist-heavy subway station. Equipped with four HD wide angle cameras, one infrared thermal camera, 16 microphones, and wheels, the K5 security robot works the subway station, accompanied by an officer, between midnight and 6 a.m.
I needed to see it, but had to pull some teeth to find someone to accompany me. All three of my roommates left my text request on read, and a few friends initially acquiesced and later bailed. On a subway ride to Queens the day prior, I explained my K5 mission to another friend who luckily agreed to go with me. "It'd be funny to see a police officer replaced by a big cone," they said.
As I approached K5, the robot's cameras came eye-level with my face. It rolled along the hallway as I walked behind it. The blue fluorescent lights flashing across K5's armor reflected on the dirty white subway wall tiles as it moved. As if suspicious of my stalking, it stopped and stared back at me. Its movements felt so uncanny that at that moment, and -- I know this sounds silly -- I was waiting for it to say something to me.
Amid department-wide staffing shortages, New York City is renting the security camera on wheels for $9 an hour, well below the $15 minimum wage, as the city finds more "cost-efficient ways to bring about safety," Mayor Eric Adams said in a joint press conference announcing the robot in September.
Police departments and privately-owned businesses across the country are deploying robots like the K5 as a lower-cost crime reduction alternative to security guards and police. Some are diffusing bombs or responding to 911 calls. But most of them are simply taking videos in public spaces and storing that footage in a database, where behind a screen, departments and business owners can access it to monitor suspicious activity.
The NYPD and Knightscope, the self-driving autonomous security robot company behind K5, say they believe this robot could act as a physical deterrent to crime, as well as on-the-ground eyes to record those who commit crimes for future prosecution. More crime will be caught on video, and the perpetrators will wisen up to where and when they commit one (or at least be deterred by a four-eyed robot). Privacy advocates, on the other hand, say these sorts of bots function as "security theater" at best and state-sanctioned surveillance at worst. But will these robocops actually stop crime and make our world safer?
The K5 robot taking a stroll, accompanied by an NYPD officer.
Supplementing human patrol with robot police
In 2022 alone, the Times Square-42nd St. station saw 45 million passengers, the highest ridership out of all 400-plus New York City subway stations in 2022 alone, according to data released by the Metro Transit Authority.
Compared to last year, subway crime is down by 4.5% this year, and down 8.1% compared to pre-pandemic levels, said NYPD Transit chief Michael Kemper at the September press conference. However, those statistics haven't changed the public safety perception that a majority of New Yorkers hold. Some 57% of New York City residents say they are concerned about safety in public places and 46% say they have witnessed violent or threatening behavior themselves, according to a recent poll conducted by the Siena College Research Institute.
In comes Knightscope, whose mission is to "make the United States of America the safest country in the world." The company's robots patrol hospitals, malls, casinos, public parks, parking lots, and more. A few of its robots implement facial recognition, a few of them can read license plates -- and all of them use AI. They use it to navigate the locations they patrol, re-navigate those locations when a car or a person or a fallen tree gets in the way of their computed path, make decisions on when and how to autonomously recharge, detect certain people and license plates, and more, according to a Knightscope blog post that addresses customers' AI questions.
To speak with a Knightscope representative, I filled out a form and clicked a button that ironically confirmed to Knightscope that I myself was not a robot. I then connected with Knightscope executive vice president and chief client officer Stacy Stephens, who told me humans are absolutely terrible at boring, routine, and monotonous activities, like a patrol.
"If you're doing the same thing over and over again, you become complacent very, very quickly. Robots, on the other hand, do that exceptionally well," he said.
Also: Robots are servicing short-staffed restaurants. What happens to the human waiters?
A lot of what a police officer does on patrol is standing and watching his environment to detect odd activity -- to keep an eye out for a bag that's left behind on a subway platform or check in on an erratic passenger. But if, for instance, a robot is surveilling an area when a fight breaks out or someone's purse is robbed, it would take on-the-ground video, and then a human officer could still intervene. The robot also has a call button that uses two-way audio to connect riders to immediate human assistance.
The robots are incapable of spotting a criminal act on their own, Stephens said. Instead, they're monitoring for "out-of-the-ordinary detections" that the user programs into the bot.
Knightscope and the NYPD say that the K5 robot doesn't use facial recognition. But when asked what methods the robot employs to detect suspicious activity, Stephens said the robot has a watch list that pulls from license plate recognition, mobile devices, and photographs to detect known threats and bad actors.
The New York City subway system is not short on cameras. In fact, the subway network has "more cameras than a Las Vegas casino," president of NYC Transit Richard Davey said at the press conference. And more are on their way. News came out in July that a select number of stations installed cameras that use AI to track fare evasion, and two dozen more stations will implement the technology by the end of the year. The system doesn't currently flag individual evaders to law enforcement, but an MTA spokesperson and police spokesperson declined to comment to NBC on whether that policy would change.
So if the K5 does not use facial recognition, then why does the NYPD need more cameras in the form of a roaming robot? At the same press conference, Kemper said the robot is "supplementing" the expansive camera network with K5's moving cameras. Along with 360-degree video footage that could be used as evidence to prosecute criminals after they commit crimes, Stephens also said he believes the sheer presence of K5 would deter crimes in its field of vision.
Knightscope's robots have captured video of domestic violence that was eventually used for prosecution, identified the person of interest in a stolen vehicle incident through license plate recognition, aided in the arrest warrant of a sexual predator, and assisted in the identification of a gunman in a mall shooting, according to its website.
The security robot company is swimming in success. It has nearly doubled sales from this year to last and announces new contracts multiple times a week, Stephens said. Not too shabby for a company that told its investors in 2021 that it wouldn't be solvent after the third quarter of 2022, according to an NBC News report.
The goal seems to be to implement surveillance technologies like Knightscope's robots on a "large scale," its website states. With over 19,000 police agencies, 8,000 security companies, and many more private security practitioners across the country, its website suggests the need for a national database or "single objective source" for tracking and fighting crime.
Stephens said Knightscope isn't doing anything "offensive" with its robots. "It is something that can be utilized to hold somebody accountable for their actions," he said.
Also: Can AI curb loneliness in older adults? This robot companion is proving it's possible
This isn't the NYPD's first robot rodeo. Two years ago, the police department rolled out a Boston Dynamics robodog, to the chagrin of activists and council members who cut its run short and called it dystopian. This year, the NYPD is deploying the robodogs once again to be used for life-threatening situations like bomb threats or shootings.
Subway passenger Sean Carey shares a similar dystopian sentiment on robot surveillance. He moved to New York in 2014, and has spent the last two and a half years penning poems for the public in places like the station. He spends a few hours a day sitting on a collapsible camping stool at the subway. That night I met Carey, he wore a Canadian tuxedo with an "Ask me for a poem" sign hung around his neck and a Hermes Baby typewriter on his lap. For someone who spends so much time in areas New York is investing security resources in, he isn't keen on being under the city's constant surveillance.
"It forces you, rather than to be your true self, to perform. If people are forced to perform too much, they're gonna get angry," he said.
Carey also said he feels safer down in the subway than in the city streets above -- that is, without the robot cop. Safety aside, he still values the human part of patrol. At Washington Square Park, where he also sets up and types poems, he recounts one police officer who makes a point every day to interact with the street performers, vendors, and artists in the area.
"He gets to know them and gets to know what they're doing. That way [the police] come up with a nuanced approach to encouraging you to comply with whatever regulations they're enforcing," Carey said.
Diffusing intense situations requires lots of interpersonal experience and community knowledge. "I've been on the street enough to see just how much the rules are bent, interpreted, and inconsistently enforced. It's kind of up to the discretion of the officer, but there needs to be discretion," he said.
Carey sees the robot as something other than a security ruse. Earlier that night, he witnessed a family come up to K5 and take a family portrait with it, "like it was a mascot in Times Square."
'Security theater' in NYC subways?
With or without a roaming robot cop, this particular subway station teems with scenes and otherworldly sights. Monotony might come easy in an empty mall or suburban parking lot, but not the underbelly of Times Square. The question arises: is it effective for a robot to use data and predictive analytics to detect or even predict unusual behavior in such a populous subway system?
"If there's a place in America where the unusual is usual, it's the New York City subways," said Andrew Guthrie Ferguson, an American University law professor and author of the book "The Rise of Big Data Policing."
"The idea that you could sit in some computer lab and create a situation where you'd be able identify the unusual such that it should draw police suspicion -- I'd love to see it, but I just can't imagine that they're able to successfully figure out what's normal in a city that prides itself on not being normal."
Also: Robots can empower you to create your own products. Here's how
As much as the station serves as a destination, departure, or connection to another part of the city, equally matched in its utility is its spectacle. Find women selling fresh fruit and churros as you transfer trains, or a jazz band surrounded by an enthusiastic crowd in your terminal. A city distilled into underground blocks, there are naked cowboys, droves of drunken Santas in December, Broadway actors running to and from shows, tourists visiting from every corner of the earth, and so much more. A robot cop could learn a lot if it spent some time here, or, to Ferguson's point, get utterly confused.
Ferguson calls K5 a "physical manifestation of security theater," that is, a performance of safety to placate fears rather than an actual mechanism to protect people -- and it's all around us. For example, after 9/11, the NYPD did subway bag searches where they picked a few random subway stations and looked through every passenger's bag before they boarded the train.
"Of course, if you were really intent on doing something bad, you could just go to a different subway stop and do whatever you're planning on doing. But the idea was to show the public you are trying to do something even though you knew it wouldn't be very effective," Ferguson said.
Knightscope CEO William Santana Li, unsurprisingly, shares a different perspective on crime and surveillance. Li sees a future brimming with opportunities for AI to do good and stop crime, albeit through contentious use cases. Li wants to link historic data on crime, location, weather, and other patterns with real-time, on-site data for total predictive and preventative crime measures.
"It's been shown that it is possible to make a positive impact on crime through multiple studies just using historical data, but I still believe without knowing what is happening real-time on the ground right now -- that algorithm is flawed," Li writes in a Knightscope blog post.
Police are already responding and aggregating real-time crime data with data police departments store through cameras and other technology like gunshot detectors and license plate readers. Meanwhile, behind every second of video footage is massive amounts of data that the NYPD stores in its database.
But, like the humans who aggregate and shape the data, it is never objective and it includes the same biases held by its creators. Leading privacy law expert Neil Richards authored the academic article "The Dangers of Surveillance," which critically points out the risks of discrimination.
"[T]he gathering of information affects the power dynamic between the watcher and the watched, giving the watcher greater power to influence or direct the subject of surveillance," writes the law professor at Washington University, advocating for guardrails on surveillance.
While these technologies concentrate on using available data to predict and deter crime before it happens, it can only go so far to quell the reasons people commit crime in the first place. In Ferguson's "The Rise of Big Data Policing," he recounted the Chicago Police Department's mission of "mapping the social network of violence" through a heat list in 2016 that found that 70% of those shot in Chicago to be on the list and 80 of those arrested in connection with the shootings. If these technologies aren't paired with "interventions, resources, and redirection," however, crime continues.
"Without targeted (and funded) social service interventions, the algorithm just became a targeting mechanism for police. Plainly stated, mapping the social network of violence may be easier than ending the violence. Data identifies the disease but offers no cure," Ferguson wrote.
After the night ended and I'd successfully sought out K5, my friend and I took the subway home. As we waited for the S train to arrive, I looked up to find a security camera above my head. Once we arrived at our home station and got off the subway, I noticed a screen hanging on the tile wall displaying a message from the MTA: "We're always looking out for your safety." The picture on display shows a collage of computer monitors and security camera footage of a subway platform.
"Safety counts," the header read.