A researcher set out to ask Florida police about the tools that watch the rest of us. The answer is the place to start.
Scott W. Waddell, D.S.I.
Based on the doctoral research of Trei McMullen, D.G.S., American Public University System, 2025. See “Sources and notes” at the end.
Written with AI assistance under the author’s direction and independently checked against the source dissertation. See “Sources and notes.”
Here is a question worth sitting with for a moment. If you wanted to know how police use the tools that watch the rest of us, who would you ask?
Trei McMullen asked the police.
McMullen is the author of a 2025 doctoral dissertation at American Public University System, and the original plan was simple: sit down with Florida law-enforcement officers, face to face, and hear in their own words how they use the technologies the study is about.
The officers said no.
Not all of them, but enough of them declined that the interviews could not go forward, and the dissertation records why. They did not believe their identities could be protected, and some pointed out that even major corporations cannot keep sensitive information safe from cyber threats. They worried about what would happen to their careers if their answers got out. So the interviews were set aside, McMullen rebuilt the study around an anonymous online survey, and the dissertation admits plainly that the change cost depth.
Stop there for a moment and notice what happened. The officers the study most needed to hear from did not trust that their own words could be kept private. Hold that thought, because we are going to come back to it, and it turns out to be the whole story in miniature.
The question McMullen wanted to answer sounds academic, but it is not. How does bias that nobody intends to have get into the technology police adopt, and what does that do to the privacy and safety of the people on the other side of the lens?
One thing to know before anything else. The dissertation poses four research questions, and the survey it ended up with, as printed, reaches only one of them directly. That is my reading of the published questions, not the dissertation’s own statement, and I will come back to it.
“Unconscious bias” is the study’s term for the stereotypes we carry without choosing them. The dissertation borrows its definition from a Harvard Business Review article: preconceived ideas about people from a particular community, shaped by the world we grew up in and the experiences we have lived through. The difficult part is the word “unconscious,” because you cannot correct what you cannot see in yourself.
McMullen is careful on one point, and so will I be. The study does not say that every officer carries this bias into their work; it says the pattern shows up across institutions often enough to deserve attention, and that the tools can carry it even when the person holding the tool does not.
That is the claim that matters, so read it again. The bias can be in the tool.
Start with the tools you already know: body-worn cameras, dashboard cameras, and drones. Then add the ones you may not: software that reads license plates, software that predicts where crime will happen, and software that matches a face in a crowd against a database of faces.
The face-matching software comes in two kinds, and the difference matters more than most people realize. One kind checks whether your face matches the single face it expects, which is what happens when your phone unlocks. The other kind takes a face from anywhere and searches it against many faces, and that second kind is the one used in policing. The dissertation cites numerous wrongful arrests that came from bad matches.
Then there is the software that predicts. One tool the study spends time on is COMPAS, an algorithm that scores how likely a person is to commit another crime. Judges do not sentence on the score alone, but they see it, and higher scores go with longer sentences. Because the way the score is computed is not public, a person can receive a different sentence from someone else and never learn why. The dissertation cites reporting that the tool is biased against young people.
Another tool, Risk Terrain Modeling, maps where the features of a neighborhood pile up in ways linked to crime, so that an agency can focus its resources there. It does not name people; it names places. But people live in places.
Here is the mechanism the dissertation describes for the predictive tools. They often learn from historical crime data, and police records are history. The dissertation warns that if those records reflect years of heavier policing in some neighborhoods, a tool trained on them can direct more police presence to the same places, generate more records, and set up a feedback loop that deepens the disparity. It presents this as a risk, not a certainty.
The dissertation calls that risk deployment bias, the first of four kinds it names. The others are bias in the algorithm itself, bias in whom surveillance is pointed at, and bias built into an institution’s policies. Cameras and drones raise the second of those, who they are aimed at, and a separate worry the dissertation also raises: who ends up holding what they record. You do not need the labels, but you do need the idea, because nobody has to intend any of it.
This is a Florida study, so look at Florida.
In 2021 the state passed Senate Bill 44, which widened what government agencies may do with drones to include managing traffic, collecting evidence at crime scenes, and running search and rescue. The sheriff of Volusia County, Mike Chitwood, was quoted saying he would like to reach the point where “every supervisor has a drone mounted on their car,” because more information in deputies’ hands means better decisions. A policy analyst at the Electronic Frontier Foundation, Matthew Guariglia, was quoted with the other side of the argument: history shows, he said, that when police get a tool for emergencies, “those emergency circumstance parameters tend to grow and grow in authority and scope.”
Both of those people are right about something, and that is exactly what makes this hard.
Florida later required agencies to buy drones only from approved manufacturers, and in 2023 it banned drones from foreign makers, citing the risk of espionage and data breaches. Notice what that is: the state worrying about who can see the data. Keep that in mind too.
The dissertation also cites reports of what can go wrong. In Pasco County, a program that flagged people as likely future offenders based on their past contact with the justice system drew complaints from residents, who said police visited so often that it felt like harassment. In Volusia County, the same drones praised for faster emergency response were criticized when they were flown over protests.
I want to be exact about what I am telling you here. These are reports the study cites; the study did not investigate them, and neither did I.
Now the survey, and I will give you its size before I give you its numbers, because the size is the point.
McMullen ran two anonymous online surveys. Twenty civilians answered the first, fifteen of them living in Florida and five living elsewhere, and ten current or former Florida officers answered the second. Everyone was screened in by a first question, and nobody was chosen at random. People volunteered, which means, as the dissertation notes, that some kinds of people may not be represented at all.
Everyone also reported on themselves, on a subject where, as the dissertation notes, people may give the answer that sounds best. The dissertation says all of this in its own limitations section. I am not adding it; I am repeating it.
So these are not Florida’s numbers. They are thirty people’s numbers, and here is what those thirty people said.
Ask civilians whether police technology intrudes on privacy and security too much, and they split exactly in half: ten said yes and ten said no. The officers were asked a question on the same subject, worded differently, and the dissertation reports the same even split: five and five.
Read that twice, because the officers were as divided as the public. If you expected the line to run between police and citizens, it does not, at least not in this room. It runs down the middle of both groups.
Now the question about having a say, which the two groups were asked differently. Civilians were asked whether having a voice in the laws that govern them would make them more likely to follow those laws, and fourteen of twenty said yes. Officers were asked whether citizens are more likely to follow laws when they feel heard, and seven of ten said yes.
I should be exact here, because the dissertation is not: its results chapter describes the officer question as if it asked about the officers themselves, while the printed question asks about citizens. Either way, what was measured was a belief about voice, not anyone’s behavior. Seven in ten, on both sides of the badge, held that belief.
Then come the answers I keep thinking about, and all of them come from the officers. Six of the ten had seen technology such as a body camera fail, their own or a colleague’s, while eight of the ten had seen an officer’s technology help in a situation. Both are true at once. The tool that saves you is the tool that fails you.
Five of the ten did not know what happens to the data that police tools capture. Half of them, and these are the people who use the tools. When asked about the training they received on tools like body cameras and facial recognition, the dissertation reports that, of the officers who answered that question, 44 percent called it mediocre, 22 percent a crash course, and 33 percent good or great; it does not say how many answered.
One more number, from the civilians, and I will state it exactly as the study does. Seven of the twenty said they had, at some point, been wrongly identified by some form of police technology or identification system. The printed survey item did not ask what happened, so we do not know what they meant, but seven people volunteered that answer.
McMullen added a third piece to the study: a reading of twenty published statements by law-enforcement executives, drawn from trade magazines and reports, coded for whether they acknowledged bias in these technologies and whether they proposed to do anything about it. The study reports that most of them did acknowledge it, and that about a third offered no concrete plan. Acknowledging a problem in a magazine is not the same as fixing it in a department, and the dissertation knows that. So do you.
Let me be candid, because you deserve it. This is a small study. Its four research questions were ambitious, and they included how agencies decide which tools to buy and how they detect bias in their own officers during hiring, training, and service. As far as the printed survey shows, those questions were not asked; the items reach only the second question, and only through what people perceive. That is my comparison of the questions with the printed instrument, not the dissertation’s own admission, and one caution goes with it: the methods chapter describes open-ended prompts in one place and a multiple-choice-only civilian survey in another, so I can speak only to the questions the dissertation prints.
What the dissertation does say is that losing the interviews limited the depth of what it could learn, and that the face-to-face conversations which might have reached further did not go forward. It also notes that some participants were hesitant to engage because they associated research on predictive policing with fights over diversity, equity, and inclusion programs.
So do not take the percentages to the bank; take the pattern instead. On privacy and on voice, the officers and the public gave the same answers, allowing for the different wording. And on the questions about their own tools, the officers admitted gaps that should concern all of us: the tools fail, the training was thin for most who answered, and half of them said they did not know what happens to the data.
McMullen frames the whole study with two ideas, and if you strip the academic names off them, they are things you already know.
The first idea is about why anyone adopts a new tool. People take up a technology when it seems useful and easy, when the people around them expect it, and when it makes them look current. That last part is where McMullen sees the opening for bias, because an agency that adopts a tool to keep up with the agency next door, or to project an image, may adopt it in a hurry, before anyone asks what it does to the people it is pointed at.
The second idea is about why anyone follows a rule. People comply with rules they see as fair, and they see rules as fair when they had a voice in making them, when the process was neutral, when they were treated with respect, and when the people in charge could be trusted. Voice, neutrality, respect, trust. That is the whole list.
Now put the officers’ refusal next to that list. The officers who held back gave the one reason the dissertation records: they did not trust that their information would be kept safe. The study’s own theory says people cooperate with processes they trust and had a say in. That is my reading rather than the study’s, and the study did not test it, but the fit is hard to miss. And separately, seven of the ten officers who did answer the survey said citizens follow laws more readily when they feel heard.
Put those two things side by side. The officers who held back wanted their own information protected. The officers who answered believed that people cooperate when they have a voice. In this small study, that is what both sides said.
That is not a police problem or a civilian problem. It is the problem.
Here the study ends and I begin, and I want you to see the seam.
McMullen’s dissertation closes with recommendations, and they are sensible ones: independent audits of predictive and facial-recognition tools, public reports on what tools an agency uses and what they can do, algorithms that outside experts can inspect, training that covers ethics and not just buttons, advisory boards and town halls with real influence, feedback from officers and residents alike, and rules for how long data is kept and how it is protected. Those recommendations are the study’s. What follows is mine, and it is the reason I wrote this.
You are not a bystander to this. Whoever you are, you have standing, and you have something to do.
If you are a citizen, ask what happens to the data your city’s police tools collect. It is a fair question, and half the officers in this study said they did not know the answer for their own. When your city or county holds a meeting on a new camera system, a new drone program, or a new piece of software, go to it.
Seven in ten people on both sides of this study believed that people follow rules they had a say in. Test the belief. Help make the rules.
If you are an officer, and the training you got on these tools was a crash course, as some officers in this study said theirs was, ask for more. Ask what happens to the footage your equipment records, because, as the dissertation suggests, some officers worry about how that footage is stored and reviewed too. The officers who held back from this study were right to want their words protected. Extend the same instinct to the people you record.
If you lead an agency or sit on a council, publish the policy before you buy the tool, not after. Put someone from outside the department on the body that reviews it, and get the audit and read it. Fear of a bad headline is not a reason to skip these steps. It is the reason to take them.
That is the pattern, and it is not complicated. The problem is real. You have more standing than you think. The research points to what helps. And you have something to do.
The officers said no. It is tempting to read that as a failure of the study, but read it instead as its clearest lesson. When people cannot see where their information goes, and cannot trust the process that handles it, they hold back.
The officers in this study did. The dissertation cites research suggesting that people in heavily surveilled neighborhoods can become less willing to report crimes or cooperate with police, and elsewhere it links that kind of surveillance to distrust. Whether those are the same reflex is my reading, not the study’s finding. The study’s answer is to give people a say, through audits, disclosure, and a seat at the table, and I think that answer is right.
Being recorded is not the same as being heard, and a camera is not a voice. By the time this study was written, more than three hundred agencies across the country had adopted drones, by one count the dissertation cites, and a Florida sheriff wanted one on every supervisor’s car. The tools were arriving faster than the public conversation about them, as the dissertation observes. So decide. Ask the question the officers asked, and then, unlike the study, do not let it end with no.
This article is independent commentary on: McMullen, Trei. 2025. Unconscious Bias on the Implementation and Utilization of Emerging Technologies by Law Enforcement Agencies, and Effects on the Security and Privacy of Citizens in Florida: A Case Study of Florida. Doctoral dissertation, American Public University System. ProQuest number 31848616. Official record: https://www.proquest.com/docview/3192204026. The dissertation’s author was not involved in writing this article, and the interpretations and the closing charge are the byline author’s own. The honorific and degree letters used on first reference reflect the completed Doctor of Global Security degree, on evidence supplied by the byline author; the dissertation itself records only its submission in April 2025.
How this article was made: the analysis, the draft, and the claim map were produced by Claude Fable 5.1, an Anthropic model, working only from the dissertation and under the byline author’s direction. A model from a different company, GPT-6 Astra from OpenAI, checked every factual statement against the dissertation across three rounds, found gaps, and confirmed each repair. The byline author set the scope, made every editorial decision, and is responsible for what this article says.
Where the key facts come from, by printed page of the dissertation: the abandoned interviews and the reasons officers gave, pp. 121–122; the study’s limitations, p. 52; the two “voice” questions as printed, p. 28; the civilian survey results (twenty respondents, fifteen in Florida), pp. 52–58; the officer survey results (ten respondents), pp. 63–68; officers’ concerns about how footage is stored and reviewed, p. 66; the content analysis of twenty executive statements, pp. 82–92; the four kinds of bias and the Pasco and Volusia County reports, pp. 84–88; Florida Senate Bill 44 and the two quotations, pp. 16 and 19; COMPAS, p. 22; the two theories, pp. 25–26 and 110; the study’s recommendations, pp. 61–62, 72–74, 106, and 113–119; drone adoption and the pace of public discussion, pp. 15, 32, and 39; reduced cooperation under heavy surveillance, pp. 35–36 and 106. Figures attributed to the dissertation’s cited sources (for example, wrongful arrests from facial-recognition mismatches, p. 30) are the dissertation’s citations, not this article’s research. A full claim map with a row for every factual statement is available on request.