Designing AI for Prosecutorial Governance: Case Prioritization and Statutory Oversight in Mexico
This paper presents a co-developed machine learning system for the Zacatecas State Prosecutor's Office in Mexico that utilizes Random Forest classifiers to predict case resolution timelines, generating prioritized lists to improve prosecutorial efficiency without disrupting existing workflows.
Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer
Imagine a massive library where every book represents a criminal case. In the state of Zacatecas, Mexico, the librarians (prosecutors) are drowning. They have over 234,000 books on their shelves, but they only have enough hands to check a few hundred a week. Many books sit untouched for years, gathering dust, while new ones keep getting added every day. The result? A giant pile of "unread" cases, and some of these books might even be legally expired (like a library book that is so old the library can no longer charge you a fine for it).
This paper describes a project where data scientists teamed up with these librarians to build a smart sorting robot. This robot doesn't make decisions, write reports, or arrest people. Instead, it acts like a very organized assistant that whispers, "Hey, look at these 300 books first; they are the ones most likely to be finished soon."
Here is how the system works, broken down into simple parts:
1. The Problem: The "Endless Pile"
The prosecutors have a specific section of the library called the MAT (Early Attention Module). This is where new cases arrive. Because so many cases come in and so few get finished, the pile grows. The prosecutors need a way to figure out which cases are "ready to go" (likely to be resolved in the next six months) and which ones are "stuck" (likely to sit there forever or have already passed their legal expiration date).
2. The Solution: A "Crystal Ball" for Cases
The team built a computer program (a machine learning model) that looks at the history of every case. Think of it like a weather forecaster, but instead of predicting rain, it predicts case completion.
- What it looks at: The robot doesn't read the whole story of the crime. Instead, it looks at the "footprints" left behind in the digital system. Did the case move from one desk to another recently? Did a lawyer update the file? How long has it been since someone touched the file?
- The Magic Trick: By studying 10 years of history (2014–2024), the robot learned that certain patterns (like a file being transferred between teams) usually mean a case is about to be solved.
3. The Two-Pronged Approach
The robot does two jobs at once, like a traffic cop directing cars in two directions:
Job A: The "Go" List (Top 300)
Every week, the robot picks the top 300 cases that look like they will be finished in the next six months. It hands this list to the prosecutors. The goal is to help them clear the "easy wins" quickly, keeping the workflow moving.- The Result: The robot is surprisingly good at this. When tested, it correctly identified "finishable" cases about 74% of the time in its top 300 picks. That is much better than just guessing or picking cases at random.
Job B: The "Stop" List (Bottom 1,000)
The robot also looks at the bottom of the pile—the 1,000 cases that look like they will never be finished. It checks if these cases have been sitting there so long that they might have passed the legal time limit (called "statutory prescription").- The Result: About 32% of these "stuck" cases appear to have passed their legal expiration date. This gives the prosecutors a "to-do" list for administrative cleanup: "Check these files; they might be legally dead and can be closed."
4. How It Works in Real Life
The authors are careful to say this robot is not a boss. It doesn't tell the prosecutors what to do.
- The Analogy: Imagine a GPS in a car. The GPS says, "Traffic is heavy on this road, try this other route." But the driver still decides whether to take the route. Similarly, the prosecutors still decide which cases to work on. The robot just highlights the ones that look promising.
- The Plan: The team is currently running a real-world test (a "Randomized Controlled Trial"). They will give the "Top 300" list to some prosecutors and see if they finish more cases than the prosecutors who don't get the list.
5. Why This Matters
This project is special because it was built with the prosecutors, not just for them.
- It fits the workflow: It doesn't require them to learn new software or change how they work. It just gives them a weekly list.
- It's transparent: The robot uses simple, understandable rules (like "how many times was the file moved?") rather than a mysterious "black box."
- It solves two problems: It helps clear the backlog (efficiency) and helps find expired cases (legal oversight).
The Bottom Line
The paper shows that by using data to organize a chaotic pile of work, a government office can work smarter without hiring more people. The robot acts as a filter, helping human experts focus their energy on the cases that are most likely to move forward, while also flagging the cases that have been forgotten for too long. It's a tool to help the librarians manage the library, not a replacement for the librarians themselves.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.