Andrew Guthrie Ferguson (American University Washington College of Law) has posted AI-Assisted Police Reports and the Challenge of Generative Suspicion on SSRN. Here is the abstract:
Police reports play a central role in the criminal justice system. Many times, police reports exist as the only official memorialization of what happened during an incident, shaping probable cause determinations, pretrial detention decisions, motions to suppress, plea bargains, and trial strategy. For over a century, human police officers wrote the factual narratives that shaped the trajectory of individual cases and organized the entire legal system.
All that is about to change with the creation of AI-assisted police reports. Today, with the click of a button, generative AI Large Language Models (LLMS) using predictive text capabilities can turn the audio feed of a police-worn body camera into a pre-written draft police report. Police officers then fill-in-the blanks of inserts and details like a “Mad Libs” of suspicion and submit the edited version as the official narrative of an incident.
From the police perspective, AI-assisted police reports offer clear cost savings and efficiencies from dreaded paperwork. From the technology perspective, ChatGPT and similar generative AI models have shown that LLMs are good at predictive text prompts in structured settings, exactly the use case of police reports. But hard technological, theoretical, and practice questions have emerged about how generative AI might infect a foundational building block of the criminal legal system.
This Article is the first law review article to address the challenge of AI-assisted police reports. The Article first interrogates the technology, providing a deep dive into how AI-assisted police reports work. Promises around innovation are countered by concerns around how the models were trained, questions around error, hallucinations, and bias in transcription, and how the final police report will be impacted by the generative prompts. Issues around structure, timing, legal gap-filling, and factual gap-filling are all addressed, with an eye toward comparing this innovation to existing human report writing.
The Article also addresses the bigger theoretical question about the role of the police report. The Article contrasts two visions of a police report: a narrow, instrumental vision of a police report and a broader, accountability vision of a police report. The goal is to show how a change in technology might also change the traditional role of the police report.
Finally, the Article explores how AI-police reports will alter criminal practice especially in misdemeanor and low-level felony cases where investigation and grand jury action is minimal. A police officer’s determination of what happened as a factual and legal matter directly impacts initial prosecutorial charging decisions and judicial pretrial detention decisions. In addition, the police report influences plea bargains, sentencing, discovery obligations, and trial practice. The open question is how reliance on AI-generative suspicion will distort the foundation of a legal system dependent on the humble police report.
