Solving Crime With AI: A Look Into AI Research Involving Police Reports

Welcome to this week’s Deep-fried Dive with Fry Guy! In these long-form articles, Fry Guy conducts an in-depth analysis of a cutting-edge AI development or developer. Today, our dive is about cutting-edge research that uses AI to identify crime patterns, specifically those tied to rape cases. We hope you enjoy!

*Notice: We do not gain any monetary compensation from the people and projects we feature in the Sunday Deep-fried Dives with Fry Guy. We explore these projects and developers solely for the purpose of revealing to you interesting and cutting-edge AI projects, developers, and uses.*

Sarah was raped when she was 16. She told her then-boyfriend, who reacted in an extremely unsupportive and hurtful way—blaming her for the incident.

"I was hurt from what happened to me and they were hurt by how I responded for the two years after. So we all pretended that it didn’t happen. If we had talked about it more, it would have been better."

When Sandra was in high school, she was raped by her neighbor whom she had trusted and considered a family friend.

“You are not alone. There are other people going through the same exact thing. I know it’s really hard to do, but talking about it helps.”

Those who have been or know of victims of these damaging acts can attest to the fact that the justice of those afflicted by such things as rape, abuse, theft, and other harming acts deserve consistency in their investigations.

Imagine if AI could help write police reports and show investigators which criminal cases deserve a further look and reveal patterns which might help investigative work … In this way, AI could truly revolutionize the entire criminal justice system.

THE MASTERMINDS BEHIND THE RESEARCH

The criminal justice system has been relatively slow in adopting AI and machine learning technologies compared to other fields. However, Cleveland State criminology professor Dr. Rachel Lovell and a Texas A&M graduate student named Jiaxin Du recognized the potential for AI to address problems within the system, particularly in cases of sexual assault.

Lovell focuses most of her research in gender-based violence, such as sexual assault, human trafficking, intimate partner violence, and more. Since around 2009, she has worked in the applied field, primarily as a methodologist and evaluator for a variety of projects in public health.

Du is currently a Ph.D. candidate at Texas A&M University. His background is in data science and information systems. Du and Lovell got connected through one of Du’s advisors. Given his data-driven technological background combined with Lovell’s background in sexual assault cases and connections within the criminal justice system, they made for a perfect team to conduct this research.

In a world where AI is infiltrating almost every area of our lives, one area that could benefit from this implementation is the criminal justice system. As Lovell remarked, “The criminal justice system is behind the curve in almost all forms of technology, but especially machine learning.” The hope to solve this problem drove this research forward.

DATA ANALYSIS

Through a previous partnership with the Cuyahoga County Prosecutor’s Office, Lovell was able to gain access to over 6,000 rape and police reports from the past two decades. Because they knew the outcomes of these older, closed cases, Lovell wondered if the language in those reports was predicting what happened in those cases—including how certain cases were written for certain types of victims, how long certain cases were open, or how certain cases were closed.

These reports were each put in their own folder by the officials, so this made pulling them out and reading them quite difficult—a process which spoke volumes to the need for improvement of this system. Lovell and Du had to go into each of the files individually, find the report, save it, and transfer them to text via a variety of computer science methods. These methods included hand-writing recognition tools, verbal dictation tools, and more.

From this point, machine learning algorithms were employed to evaluate the language used in these reports and predict which ones led to prosecutions and which ones did not. Because they knew the results of the cases, they were able to train the machine to identifying words, phrases, and patterns within cases which led to prosecutions and those that did not. The algorithms were also designed to identify subjective language, which proved to be a key focus of the study.

THE RESULTS?

The research yielded surprising results. Contrary to some expectations, reports that were more subjective and rich in personalized details were more likely to lead to successful prosecutions. These reports included elements of the crime and highlighted the emotional and human aspects of the cases. This suggested that the way reports were written, with a victim-centric and detailed approach, was more effective in conveying the nature and seriousness of the sexual assault. The more factual, concise, and indifferent reports, on the other hand, were less often looked into.

What Lovell and Du also found were that certain “signaling” words were predictive of certain outcomes. Some of these words were strange at first blush, but had tangential implications. For example, one word they found to be indicative of certain behavioral outcomes was “basketball.” It is not that basketball was related in a direct way to the crime, but tangentially so. The report might have referred to people leaving the basketball court or playing basketball, indicating the deeper correlation: the case involved kids or adolescence outside in certain neighborhoods. This conclusion indicated a risk factor for certain types of rape cases.

INCIDENT REPORT REFINEMENT

Incident reports are the frontlines of investigation. As Lovell pointed, out, “Investigators often don’t see a victim first; they see a report first from someone who responded to the call first.” She continued, “Where there are heavy case loads, these investigators decide what cases to follow up on based on the content of these reports.” This process highlights the importance of report writing, where what the report reads makes a big difference to the investigation.

Sexual assault cases have unique characteristics that can lead to biases in incident reports. In such sensitive cases, victims may not respond emotionally or may have difficulty providing specific details due to the traumatic nature of the crime. For example, some respond to traumatic events by crying and others by laughing—both normal reactions from victims, as there is no typical response to trauma. This can lead to misunderstandings and doubts on the part of law enforcement, which can be reflected in the reports.

With the weight of these police reports in mind, Du said, “We want to help police officers write better reports.” The researchers are working on developing ways to help police officers write reports that are more consistent and better for review and analysis. Lovell adds, “Most officers are taught to write in this perfunctory type of way which removes the reality of rape (or the crime) from the report.” But the research shows that this is not the most effective way to produce a report.

Police officers find themselves in high pressure situations that can put a lot of pressure on their reports. Lovell and Du believe these officers would benefit from having structured prompts rather than an open-ended text box. Lovell envisions the use of AI to develop prompts and utilizing phrasing tools to help officers write reports that are comprehensive and consistent.

SETTING THE TABLE FOR THE FUTURE

Lovell and Du’s commitment to the victims of these rape cases as well as to improving the criminal justice system are clear. AI is a useful tool for society, and they believe that harnessing it in the right ways can help us discover things about crime that we have never known before and give us opportunities as communities to improve upon our current investigative systems and procedures.

Beyond highlighting the potential of AI in addressing biases and improving the quality of incident reports, this study opens up possibilities for further research and development of AI tools in law enforcement. Although Lovell and Du focused primarily on rape cases, this sort of “signaling” approach and improvements to initial reports could be expanded to other types of cases such as theft, homicide, and more. Lovell and Du hope their continued research inspires others to find better ways to implement AI into the criminal justice system to protect the justice of those afflicted by such crimes.

RESOURCES

Do you want to donate to those who have been afflicted by rape and other forms of sexual assault? If so, we encourage you to consider giving to one of the following organizations:

Cleveland Rape Crisis Center

Rape, Abuse & Incest National Network

*If you or anyone you know has been afflicted by some form of sexual assault and needs a resource, help is available: National Sexual Assault Hotline: 1-800-656-4673*