A criminal justice student reviews a fictional disturbance. The scenario states that a witness heard shouting and later saw two people standing outside a building.
The student enters those facts into an AI-assisted writing tool. The draft reads:
“The witness observed the two individuals arguing outside the building.”
The sentence sounds reasonable. But the scenario never says the witness saw the argument. The witness heard shouting and later saw two people outside. It remains unknown whether those people were arguing.
Would the student notice the difference?
Observation comes first
Before students use AI to assist with police documentation, they should practice identifying what they know, how they know it, and what remains uncertain.
That means distinguishing an officer’s observation from a witness’s statement. It also means recognizing the difference between a fact and an inference.
For example, “the window was broken” describes an observation. “The suspect broke the window” requires additional facts connecting the person to that action.
These distinctions are part of clear documentation. The National Institute of Justice identifies recording actions and observations as a component of initial crime scene response and training (National Institute of Justice, 2004).
A polished draft still needs review
Generative AI can produce text that sounds confident but contains inaccurate or unsupported information. The National Institute of Standards and Technology describes this risk as confabulation (Autio et al., 2024).
Students should therefore compare a draft against the original scenario, not just read it for grammar.
An instructor might ask:
- Does each statement match the information provided?
- Are witness statements clearly attributed?
- Has an assumption been presented as an observation?
- Did the draft add a detail or leave one out?
- What remains unknown?
Teach the student to own the report
A classroom exercise could begin with students writing a short narrative themselves. Next, they could use an AI-assisted tool to organize the same fictional facts and compare the two versions.
The goal would be to explain every change, identify unsupported wording, and revise the final draft. Use fictional or synthetic information for the exercise.
AI may help organize language, but the person preparing the report must understand and verify what it says. In a real agency, officers must also follow their department’s policies for approved tools, data handling, and review.
Before students learn what AI can write, they should learn to ask a more important question:
“Can I support every fact in this report?”
Explore a fictional scenario at KLYVOREK Institute and practice making your case.
References
Autio, C., Balunović, M., Baracaldo, N., Beck, A., Brandão, L. T. A. N., Broadway, A., … Ziyada, A. (2024). Artificial intelligence risk management framework: Generative artificial intelligence profile (NIST AI 600-1). National Institute of Standards and Technology. https://doi.org/10.6028/NIST.AI.600-1.
Denning, R., Bennett, S., Mazerolle, L., & Stocks, B. (2009). Serious violent crime investigations: A systematic literature search and technical report. https://core.ac.uk/download/15108253.pdf.
National Institute of Justice. (2004). Crime scene investigation: A reference for law enforcement training. U.S. Department of Justice. https://nij.ojp.gov/library/publications/crime-scene-investigation-reference-law-enforcement-training.