.Was It Racist? Santa Cruz Bans Predictive Policing

Before she moved to Santa Cruz in 2004, Brenda Griffin, the president of the Santa Cruz chapter of the NAACP, worked for a New England civil rights law firm in a big industrial city. The job left her with sad memories after witnessing the devastation caused by police violence and its aftermath. “The police officers were very brutal,” she says.

But Griffin says she had never seen someone be killed in cold blood until videos circulated in May of a Minneapolis police officer kneeling on the neck of a Black bouncer named George Floyd for more than eight minutes. The footage ignited protests around the country and prompted nationwide discussions about police violence and racial bias. All of that has given Griffin some semblance of faith that protesters may keep marching until U.S. leaders address the inequities facing the country.

“I am hopeful because the floodlights are on this issue—not just the spotlight, what the NAACP and other organizations do, but the floodlights—are on our racial inequities in healthcare and education and law enforcement. Not only in the United States, but throughout the world,” she says. “Seeing all the different ethnicities in these protests makes me hopeful that we’ll see some changes come about. I also hope that people will go the distance. We’re in a movement; we’re in a moment right now. And I’m hoping that we will continue this moment until we see systemic changes.”

The Santa Cruz City Council voted on June 23 to display the Pan-African and Black Lives Matter flags in front of City Hall every year during the month of July. That same afternoon, the council voted to establish two new groups to review police policy, preemptively ban facial recognition software, and to effectively ban a controversial predictive policing technology that Santa Cruz first started experimenting with nine years ago. Additionally, Police Chief Andy Mills announced changes to his department’s policies. He banned no-knock warrants, neck restraints and shooting at moving vehicles. He also modified parts of the department’s use-of-force procedures.

But it’s been the predictive policing decision that has received the most attention.

Santa Cruz Mayor Justin Cummings says there’s much work left to do to get bias out of police data before departments can start relying on such data to make patrolling decisions.

“If we have racial bias in policing, what that means is that the data that’s going into these algorithms is already inherently biased and will have biased outcomes, so it doesn’t make any sense to try and use technology when the likelihood that it’s going to negatively impact communities of color is apparent,” says Cummings, the city’s first-ever Black male mayor.

Santa Cruz became the first community in the country to take such action on predictive policing, which uses data to target law enforcement to areas where crime is statistically most likely to occur. Last week’s vote on predictive policing represents a significant plot twist for a high-profile invention that has garnered flashy headlines for Santa Cruz over the last nine years—and one that has deep ties to the county’s top politicians.

The technology’s future locally is not abundantly clear.

Brian MacDonald, CEO of Santa Cruz-based predictive policing company PredPol, says he doesn’t think of the council’s recent vote on the matter as a ban at all. He paints it as an opportunity for the city to press reset and make sure that predictive policing systems do not have any biases baked in. MacDonald is adamant that they do not.

According to the language in the City Council’s motion, however, the council would need to pass a new resolution to let the police start using predictive policing again.

PREDICTION MARKETS

In 2011, Santa Cruz became the country’s second community, after Los Angeles, to try predictive policing.

Although it’s impossible to ever pinpoint the reasons behind fluctuations in crime, Santa Cruz saw an 11% decrease in burglaries during a six-month pilot, compared to the six months prior, and a 4% decrease compared to the same stretch of time one year earlier. (Some contemporaneous news reports cited more significant reductions in crime, but MacDonald couldn’t confirm their veracity, as he wasn’t with the company at the time.)

The system operated based on the notion that property criminals are very often creatures of habit—that when they are successful, they tend to strike again in an area nearby. Santa Cruz Police Department (SCPD) deployed predictive policing partly as a way to cope with increasingly scarce resources during a time of Great Recession-era budget cutbacks. The pivot earned nationwide coverage, including a spot on Time’s list of the 50 best inventions of 2011. Predictive policing promotional materials from that period wrapped up the system in the metaphorical language of earthquakes: After a quake strikes, seismologists can calculate the probability that an aftershock will happen in the area and when. Property crimes, the thinking went, would follow a similar pattern. Journalists often preferred to compare the system to Minority Report—Steven Spielberg’s science fiction thriller starring Tom Cruise, wherein suspects are caught before they commit crimes.

Researcher George Mohler, a one-time assistant adjunct professor for UCSC, developed the predictive policing system with UCLA anthropology professor Jeffrey Brantingham. Together, they created PredPol, which now counts about 50 community agencies—with a combined population of 10,000,000 residents—as clients, according to MacDonald.

PredPol, which is headquartered in Santa Cruz, uses three kinds of data: crime type, crime location and crime date/time. “These three data points are the closest we can come to truly objective crime data to work with,” MacDonald says via email.

The company uses those data points to create its “heat maps”—the 500-square-foot city blocks that officers should make sure to patrol through at certain times on a given day. MacDonald says PredPol never collects any racial, demographic or socioeconomic information, nor does it use arrest data. PredPol also limits its predictions to a rather narrow window of crime: vehicle theft, break-ins, burglaries, assaults, and robberies. He says the company has seen no evidence of bias in its data collection.

“If cars are being stolen on weeknights in a predominantly Black neighborhood, our prediction boxes would show up there,” he writes. “If they are being stolen in a predominantly white neighborhood, that’s where we’ll see boxes. In either case, officers will proactively patrol those areas to deter those thefts from occurring. It’s the presence of the officers that deters the crime.”

Santa Cruz County Supervisor Zach Friend was an SCPD crime analyst and spokesperson during the department’s predictive policing rollout. He knows full well that police nationwide have a tendency to show bias when on patrols, but he says that because predictive policing relied on data from police reports, the inputs were as sound as they could possibly be.

“In fact, the greater concern in that situation was, ‘Is there under-reporting in certain populations of crime because of concern with relationship with police?’ For example, you’re more likely to report a crime in the upper Westside than you are in the Beach Flats,” he says.

Friend says that, as he recalls, predictive policing resulted in fewer patrols going to neighborhoods that may have previously been oversaturated with policing.

One prominent critique of PredPol’s methods, though, came in 2015, as published by the Royal Statistical Society’s Significance magazine. Two researchers found that, when they plugged numbers for Oakland drug crime numbers into the PredPol algorithm, the heat maps fell disproportionately into neighborhoods with more people of color.

However, MacDonald notes that PredPol never used drug arrest data for precisely that reason. PredPol’s algorithm doesn’t predict for crime types that can result from officer-initiated actions, like drug use or prostitution—so as to avoid creating a statistical doom loop, where over-policed communities become increasingly over-policed as cops find more and more crime.

Santa Cruz County Supervisor and former Mayor Ryan Coonerty once served as director of government relations for PredPol. He started hearing about possible issues around racial bias in PredPol’s algorithmic data around 2014. Coonerty, who stopped working for the company in 2013, says he takes those concerns seriously and that they should be further vetted and addressed.

PROBABLE CAUSE

Another concern about predictive policing revolves around what happens when a cop shows up to an identified hotspot.

Does the officer simply circle once or twice around in their patrol car, make their presence known and keep an eye out for any obvious crimes being committed—as MacDonald and Friend indicate that they should? Or does the officer prowl around, emboldened by the heat map, looking for the slightest hint that someone might be homeless, poor or maybe even just nervous-looking?

The answers aren’t always clear. Some privacy scholars have raised questions about predictive policing based on Fourth Amendment concerns and also asked what constitutes grounds for reasonable suspicion in the era of highly targeted police operations.

In 2012, when then-Deputy Chief Steve Clark took a Santa Cruz Weekly reporter out on a predictive policing patrol, he compared a stop to one of PredPol’s hotspots to the act of “hunting.” 

Four years later, Clark—a big supporter of predictive policing—ignited outrage locally when he took a cable news reporter out on a patrol and appeared to profile a Black man walking a bike as a possible criminal, before offering reasons why a white girl on a bike didn’t look suspicious at all. Clark, who could not be reached for comment, retired in December of 2016.

When it comes to perceptions of bias among recent SCPD leaders, Clark wasn’t the only culprit. 

Former Santa Cruz Police Chief Kevin Vogel, who retired in 2017, recently took to Facebook to make what appeared to be a racist statement. Shortly after Netflix CEO Reed Hastings donated $120 million to support historically Black colleges, Vogel announced he was canceling his Netflix subscription because he said $120 million was “a lot of money for a charity that I am not interested in donating to,” as reported on GoodTimes.sc. (Vogel did not mention the colleges directly or respond to requests for comment.) In the comments below his post, the former chief added that he also disagreed with a $1 million donation from Hastings to a data-based research organization dedicated to fighting racial bias in American law enforcement. This happened on the evening of June 19, as activists were taking to the streets for Juneteenth to protest racial bias in law enforcement.

It’s no wonder activists have had concerns about a history of implicit bias at SCPD, dating back to predictive policing’s early days.

The alternatives to the data-heavy approach, however, aren’t always clear. Generally speaking, Coonerty has concerns about what it means when officers don’t rely on data to make decisions. A deputy’s habits and gut instincts may not be colorblind either.

“It’s always better to have people informed, rather having people going out and basing things on their own hunches,” Coonerty says.

He still believes the right data-driven solution could be effective to fight crime, assuming the data is sound and the algorithm bias-free. He gives the example of the recent rash of catalytic converter thefts from Priuses around Santa Cruz.

“That’s exactly what you could predict,” Coonerty says. “People’s habits are predictable. That’s an example of where [predictive policing] could be used.”

It’s also an example of where the line blurs between predictive policing and typical investigative work. Chief Mills says there is a predictive element to investigative work of his department’s crime analyst John Mitchell. But he says the focus is on solving crimes. Mills even framed Mitchell’s work as “predictive policing” in an interview with GT last year, when the term was less politically charged.

When and how exactly SCPD started to phase out PredPol’s software is unclear. Mills told GT in 2018 that he had incorporated the system into his neighborhood policing model. But a recent staff report indicated that SCPD stopped using the formula shortly after Mills arrived at the department in 2017. While on the phone with GT, Mills checked with a police sergeant to ask for his recollection of SCPD’s timeline, and the sergeant told Mills that many officers had pretty much stopped using the algorithm before Mills arrived because they didn’t feel it was working.

SARGE CARD

When it comes to doubts about predictive policing’s efficacy, Santa Cruz police sergeants would not be the only ones to feel that way.

Although Santa Cruz was the first city to pass an actual moratorium on predictive policing, police departments around the country, including in Palo Alto and Mountain View, have dumped the software in recent years because leaders said they weren’t seeing results. And in Los Angeles, after two years of pushback from activists, the Los Angeles Police Department cut ties with PredPol in April. LAPD officials said the reason was due to budget-related concerns, although an audit last year had created questions about the program’s success.

Given how civil liberties concerns have been front and center in discussions about the technology, a review of the academic and journalistic arguments against predictive policing finds them to be thinner on hard data than expected. Many of the articles look at non-PredPol types of predictive policing, including systems in place elsewhere in the U.S. that determine the likelihood that individuals would commit crimes in the near future—a practice that blatantly resembles racial profiling. (It’s a practice PredPol opposes.) Some articles conflate the various forms of predictive policing with one another or confuse them altogether. Others say PredPol relies on arrest or drug data for its algorithms, which MacDonald insists the company has never done. 

MacDonald says PredPol is a small company that doesn’t have a public relations representative and that the business has chosen not to engage with its critics. He lists eight other companies—including IBM, Motorola and Cisco—that have been involved in the predictive policing space. He says he feels that many have received less flak, despite being, in his opinion, less transparent and more problematic in their business practices. 

Hard data aside, it is the circumstantial evidence fueling activists’ deep-seated distrust of PredPol’s work that may be more compelling.

Critics of predictive policing often place it in the broader historical context. In many ways, the precursor to PredPol was a metric-obsessed approach undertaken in the 1990s by the NYPD called CompStat, which harmed communities of color and the efforts by law enforcement to work with them. 

PredPol’s leaders may have a high-minded concept of how their data is divorced from the historical trends in faulty crime data of yesteryear, but given community concerns and the possible breaches of trust caused by former department leaders in the company’s hometown, it would behoove leadership to try to get out in front of the controversies and respond to questions from the community—especially in the midst of a nationwide conversation about reforming the police. 

It’s also not unreasonable for critics to put the onus on PredPol to prove that its software helps vulnerable communities more than it hurts them.

MacDonald did not speak up at last week’s City Council meeting, nor did any other PredPol employees or supporters of predictive policing more generally. MacDonald says he sat down with Mills a couple times after the chief’s 2017 arrival in Santa Cruz. Other than that, MacDonald hasn’t talked with any city leaders lately.

“I haven’t taken the time to meet with other Santa Cruz city leaders because Chief Mills said he wanted to take a different approach,” he writes to GT. “At some point I’d like to sit down with [Mayor] Cummings and [City Manager Martín] Bernal to give them an overview of what we do and get their feedback on what we do and how it is perceived.”

UCSC professor emeritus of Sociology Craig Reinarman, who has followed the predictive policing news, isn’t surprised to hear that PredPol didn’t step forward to publically engage. He sees it as part of a larger trend.

“This is the same thing you hear from Silicon Valley all the time—whether it’s [President Donald] Trump and Twitter or Facebook ads that are lying and so on and so forth,” says Reinarman, who has concerns about predictive policing software, based on issues he’s seen with algorithms targeting probation and parole terms and also based on questions he has about how heat maps are both generated and interpreted. “Technology’s never really neutral, and they claim it’s always neutral. So I’m not surprised to hear the people who invented this didn’t want to come out and defend it in public. They may know the most about the ways in which there are risks about reifying preexisting conditions.”

FLEET EFFORT

Going forward, Cummings has ideas for other data that PredPol or another group could use to start tracking a different aspect of public safety.

He’s curious if police officers making arrests and those who file reports may be disproportionately targeting people of color. “Could it be used, not for predictive purposes, but for evaluating police departments?” says Cummings, who first brought forward the idea of the predictive policing ban in January with two colleagues—then-councilmembers Drew Glover and Chris Krohn—both of whom faced a successful, if divisive, recall election.

Although it’s the predictive policing element that earned international headlines last week, the City Council also voted unanimously to preemptively ban facial recognition software, which has been shown to be prone to major errors, particularly in identifying faces of Black and Asian people. The Santa Cruz County ACLU advocated for both changes. The city is additionally creating two new groups—a Mayor’s Community Advisory Committee and a City Council working group—to study police issues more closely.

On top of that, the council will take a close look at the city budget at a Thursday meeting, with the plan of fine-tuning the budget throughout the summer. Given budget shortfalls related to the Covid-19 pandemic and the recent protests about law enforcement, Cummings knows there will be calls to “defund the police.” And although it may sound like a small semantic nuance, Cummings has appreciated how some of the discussion in recent weeks has shifted away from defunding the police per se and toward the idea of reinvesting in the community. Big changes, though, won’t happen overnight, he says.

“We need to have all pieces in place before we start cutting funding from different departments,” he says. “And that’s what this conversation will be all about—how do we want to be policed, what does public safety look like for Santa Cruz, and how can we do so in a way that ensures equal protection for everyone?”

Update, June 30, 7:15pm: This story has been updated to clarify when Ryan Coonerty left PredPol.