Estimated reading time: 14 minutes
Key Takeaways
- Three factors consistently drive public support for AI in policing: whether people believe the technology is fair, whether they had a say in adopting it, and whether they trust the agency’s motives. Higher concern in any one of those areas translates directly to lower support.
- Knowing more about AI doesn’t automatically make people more supportive. That relationship runs entirely through how fair people perceive the process to be. Education without credibility doesn’t move the needle.
- Dignity and respect, while central to fair policing in human interactions, did not significantly predict support for AI in this study. The authors suggest people may view AI systems as separate from interpersonal dynamics.
- The bottom line for agencies: address community concerns before deployment and after, design systems with built-in fairness and transparency, and run public education campaigns that actually explain what the technology does and what guardrails exist.
In 2011, the Los Angeles Police Department began using PredPol, a predictive policing tool that used algorithms to forecast where crimes were likely to occur based on historical data. The department used it for nine years. What it did not do, consistently, was bring the community along.
Community opposition built steadily. Critics argued the algorithm directed more officers into neighborhoods that were already over-policed, reinforcing existing disparities rather than predicting actual crime. Internal audits obtained by the news outlet MuckRock suggested the agency struggled to measure PredPol’s effect on crime. By 2020, the department dropped PredPol, officially citing budget constraints from the COVID-19 pandemic. Hamid Khan, a campaign coordinator with the Stop LAPD Spying Coalition, which had spent years organizing against the program, was more direct about what he believed drove the decision: “This was clearly the community rising up.”
Now consider Chula Vista, California, which launched its drone program in 2018. The program has become a national model for transparency. Every drone deployment is logged publicly, including the location, flight path, and purpose of each flight. “Transparency and accountability are key components in the success of our drone program,” Chief Roxana Kennedy said in a statement, adding that the department “pride[s] ourselves on ensuring the public has access to our drone flight information in upholding the trust of our community.” The program still generates questions. But it has not drawn the kind of sustained organized opposition.
Neither of these is strictly an AI story, but they illustrate a dynamic that becomes even more consequential as agencies move toward more advanced AI tools, things like facial recognition, real-time video analytics, and AI-generated risk scores. The community concerns that toppled a nine-year program in Los Angeles, questions about bias, transparency, and who gets to weigh in, are the same concerns that will shape whether AI adoption holds or collapses. The technology is newer and harder to explain. The underlying dynamic is the same.
There’s also a harder reality worth naming: some communities are going to push back on new policing technology regardless of how well the rollout is handled. Privacy concerns are real, distrust of surveillance runs deep in some neighborhoods, and no amount of transparency resolves every objection. The question isn’t whether opposition exists. It’s whether your agency gets ahead of it or gets buried by it.
That tension is exactly what a new study published in Police Practice and Research set out to measure. How should police best handle adopting AI technology to earn the strongest community response?
The Study
Researchers Ahmet Guler, Sedat Kula, and Kaan Boke surveyed 583 adults across New Jersey, New York, and Pennsylvania, asking about their support for AI in policing, their knowledge of specific AI applications, and their concerns about fairness and accountability in how those tools get used. The study was published in Police Practice and Research in June 2025.
Participants rated their support across eight items covering local and federal AI use. They were also asked whether they were aware of ten specific applications currently used in policing: facial recognition, image and video analysis, predictive policing, social media monitoring, surveillance drones, and others. That gave the researchers a measure of how much each person actually knew going in.
The core of the study focused on four dimensions of what the researchers call procedural justice concerns. Think of them as four questions the public is asking about any AI adoption:
- Is the system unbiased? (neutrality)
- Did anyone ask us? (voice and input)
- Can we trust why they’re doing this? (trustworthiness)
- Will we still be treated with respect? (dignity and respect)
To analyze how those concerns interact with AI knowledge and support, the researchers used Structural Equation Modeling (SEM). This is worth a brief explanation, because it’s what makes the study more useful than a simple poll. SEM doesn’t just ask whether two things are related. It tests whether a third factor explains the relationship between them. Think of it like this: if you wanted to know whether a longer commute causes people to be less productive at work, you might discover that stress is the thing in the middle making that connection. SEM is the statistical equivalent of finding that middle variable and measuring its effect. In this case, the researchers were testing whether these procedural justice concerns sit between AI knowledge and public support, or whether knowledge has its own direct effect. The model met accepted standards for goodness of fit, meaning the data matched what the model predicted closely enough to be confident in the results.
The research question they were ultimately answering: How do perceptions of fairness, citizen input, trust, and respect in AI applications influence public support for AI use in policing?
What Drives Public Support for Police AI
The model explains 62% of the variation in public support for AI in policing. That figure accounts for fairness concerns, AI knowledge, confidence in police, age, income, employment status, and media consumption. For a survey-based study, 62% explanatory power is high. It tells you this isn’t random. There are concrete, identifiable factors driving public support, and three of the biggest are within an agency’s control.
Is the system unbiased?
Citizens who had greater concern about whether AI-driven decisions in policing are made fairly and without bias were significantly less likely to support AI in law enforcement. This is the neutrality finding.
The concern is well-founded. AI systems in policing are trained on historical crime data, which reflects decades of enforcement patterns. If certain neighborhoods were over-policed in the past, the data shows elevated crime there, and the algorithm sends officers back. The pattern reinforces itself. The Chicago Police Department’s Strategic Subjects List, a predictive tool meant to identify individuals at risk of gun violence, illustrated the problem clearly: a 2016 study found that being placed on the list did not make someone more likely to be a victim of violence, but it did make them more likely to be arrested for a shooting. The tool didn’t predict violence. It predicted police attention.
When communities know stories like that one, or when they’ve watched a nine-year predictive policing program get dismantled under public pressure the way LAPD’s was, concerns about algorithmic bias aren’t abstract. They are specific, grounded, and measurable. And according to this study, they are a direct driver of whether residents support AI at all.
Did anyone ask us?
Citizens who had more concern about whether they have any input into how AI is used in policing were significantly less likely to support it. This is the voice finding, and it’s the most directly actionable result in the study.
A 2022 survey across 30 European Union countries found a useful counterpoint here: even in regions with high skepticism about AI, large majorities supported specific applications when they understood the purpose. Of EU respondents, 89.7% backed using AI to protect children and vulnerable populations, 87.1% supported it for identifying criminal organizations, and 78.6% believed it should be used to predict crimes before they occur. People are not opposed to AI in policing in principle. What erodes support is the feeling that no one asked them, that the technology was acquired quietly and deployed without explanation, and that their concerns wouldn’t matter even if they raised them.
LAPD’s PredPol program wasn’t shut down because the community was opposed to crime prevention. It was shut down, officially for budget reasons, but with years of community organizing against it as the backdrop.
Can we trust why they’re doing this?
Citizens who had greater concern about whether police were implementing AI for legitimate public safety reasons were significantly less likely to support its use. This is the trustworthiness finding, and it’s the most reputation-dependent of the three.
The items used to measure trustworthiness asked whether AI makes police more honest, more trustworthy, and more deserving of community confidence. An agency that already has credibility, that has followed through on commitments, and that has a track record of accountability is starting from a different position than one that doesn’t. Prior confidence in policing was a significant predictor in the model: residents who trusted their department going in were more likely to support AI. That relationship runs in both directions. AI adoption is not a fresh start. It happens inside an existing relationship, for better or worse.
What dignity and respect found
The fourth dimension, dignity and respect, did not produce a statistically significant effect on support. This surprised the researchers, given how central respectful treatment is to the broader procedural justice research. Their explanation is worth considering: people may view AI systems as separate from the interpersonal dynamics that shape how a human police encounter feels. Whether an officer’s tone was respectful, whether they listened, whether they treated someone with dignity during a stop, those judgments don’t map cleanly onto a surveillance algorithm. The public may evaluate AI systems more on whether they seem fair and unbiased as tools, not on whether the officers using them were courteous.
Why more information isn’t enough
Here’s the result that most directly challenges the common assumption that more information equals more support: fairness concerns, input, and trustworthiness fully mediate the relationship between AI knowledge and support. The direct path from knowing more about AI to supporting it was not statistically significant on its own.
What that means in practice: a community member learns exactly how your new gunshot detection system works. Does that knowledge make them more likely to support it? According to this study, not directly. Whether their support increases depends entirely on whether learning about the technology changes their concerns about whether it’s fair, whether they felt included in the decision, and whether they trust the department’s motives for using it.
How to Build Public Support for Police AI
Address community concerns before deployment, not after the backlash. The study is explicit: agencies should engage citizens about specific AI applications both before and after implementation. That means holding community meetings before a contract is signed. It means publishing plain-language documentation of what the technology does, what it can’t do, and what happens when it produces a bad result. Chula Vista didn’t wait for residents to come to them. They built a public dashboard before people had a reason to demand one.
Design the system with fairness built in, not added on. Working with vendors to adopt AI tools without meaningful community involvement can create the impression that citizens are being left out of how policing decisions get made. The study argues AI systems should be designed with fairness requirements baked in: documented human review before any AI-generated lead becomes enforcement action, bias audits on training data, and regular public reporting on actual use. INTERPOL and UNICRI developed a free Toolkit for Responsible AI Innovation in Law Enforcement that walks agencies through procurement, governance, and accountability standards in plain language.
Run public education campaigns that go past the basics. Because information alone doesn’t drive support, the campaigns most likely to move the needle are the ones that address the underlying concerns directly. Explaining how facial recognition works is not enough. Explaining what happens when it produces a false match, who reviews that output before an officer acts on it, and what the policy is when the system gets it wrong, that’s what addresses the fairness and trust concerns that are actually driving the outcome. The Guler study recommends campaigns focused specifically on demystifying AI and highlighting accountability measures. The National Policing Institute’s AI in Policing resource page is a solid starting point for framing those conversations with your community.
Take citizen input seriously as a design requirement. Of the three significant predictors, voice is the most structurally addressable. Establish a community advisory process before the next AI procurement. It doesn’t need veto power. It does need to be real: people need to be able to point to something that changed because of their input. A community meeting where the decision was already made is not community input. It’s notification. The IACP’s AI Resource Hub includes policy templates and peer examples from agencies that have built public input into their AI adoption processes.
Know which fights you are actually walking into. Some communities are going to oppose AI surveillance technology regardless of the process. That opposition isn’t always irrational. Privacy concerns about facial recognition and predictive policing are grounded in documented harms. The goal of community engagement isn’t to neutralize every objection. It’s to separate the objections you can actually address, by being more transparent, more accountable, more inclusive in governance, from the ones that reflect deeper disagreements about the role of surveillance in public safety. Getting ahead of the first category makes the second easier to navigate.
Final Thoughts
Ninety percent of law enforcement officials support using some form of AI in their practices, according to a 2025 survey cited in the study. The technology is not a question of if. For most agencies, it’s already here in some form.
Guler, Kula, and Boke’s research offers something practical: a clear answer to what actually drives community support for AI adoption. It’s not the crime reduction numbers. It’s not the technology performing well in a pilot. It’s whether people believe the process was fair, whether they had a voice in it, and whether they trust the agency holding the controls.
If your agency is heading into an AI adoption and the community conversation hasn’t started yet, this study is a fairly clear signal about what happens next. The departments that got ahead of it built something durable. The ones that didn’t spent years explaining themselves.
Enjoyed this post?
Get more research translated into plain language delivered straight to your inbox.
Subscribe below and let me know which topics you want to hear more about. It only takes a minute.
If you found this helpful, share it with your team or on your socials. It helps more people make sense of research that matters.
English