Responses to Police1’s “What Cops Want in 2024” state of the industry survey question, “Is your agency using AI tools?” indicate that the use of AI large language models in law enforcement is off to a slow start.
A large language model (LLM) is an advanced artificial intelligence system designed to process and generate human-like text based on extensive training data, enabling it to assist with tasks such as writing, summarizing, and answering questions. ChatGPT, developed by OpenAI, is one of the most well-known LLMs, using machine learning to understand and analyze human language.
More than 90% of the 2,833 responders answered, “Don’t know” (21.9%) or “Never adopted” (69.4%) to the AI use question. The 242 respondents using or exploring AI tools were asked about specific uses for personnel management, operations analysis, incident documentation, and investigations. Just 12 (5%) of those 242 respondents reported frequently using AI tools to create training or educational materials. Another 29 (16.1%) said they are using AI tools occasionally, and 27 (11.2%) said they are planning to adopt AI tools to create training materials.
Despite the potential benefits, many police trainers are not yet using AI LLMs to create training materials.
Why trainers might not be using AI LLMs
Some of the key reasons police trainers might be reluctant to use ChatGPT, Gemini or other AI LLMs include:
- Lack of awareness or knowledge: Many trainers may not be familiar with what LLMs are or how they can be used effectively for training.
- Skepticism about effectiveness: There may be doubts about the reliability and quality of AI-generated training materials compared to those created manually by experienced trainers.
- Fear of errors: Concerns about inaccuracies or hallucinations in AI outputs could discourage adoption.
- Time investment: Trainers may perceive the learning curve for integrating AI tools into their workflow as too steep or time-consuming.
- Resource constraints: Limited access to necessary tools, technology, or budgets to incorporate AI into their training process.
- Policy and compliance concerns: Uncertainty about whether AI-generated content aligns with department policies, procedures, or legal requirements.
- Cultural resistance: Resistance to change or skepticism about AI within law enforcement agencies can hinder adoption.
- Limited training on AI use: Trainers may lack proper training or support on how to utilize LLMs effectively for specific tasks.
Training and familiarity are often the top reasons that slow the adoption of new technology, especially when time is limited. The persona and task technique is a great way to get started with using an LLM.
AI prompt: Persona and task method
To effectively use an AI LLM, assign it a persona or role. For example: “You are an experienced shift commander with decades of experience supervising patrol officers and delivering impactful and informative shift briefings.” Then provide a specific prompt like writing an outline, answering a question or analyzing data.
The formula is: “You are a [PERSONA], help with [TASK].”
The next step is to assign tasks to your AI persona.
3 police training activities for AI to assist
It’s helpful to think of AI as your virtual assistant. How can this assistant conduct research, and prepare materials or organization information? For example, an AI assistant might be able to help you organize or outline a shift briefing, prepare scenarios to discuss with patrol officers and create questions for a post-shift briefing quiz in your department’s learning management system.
Shift briefing: DUI enforcement
In this example, the persona is the experience shift commander and the task is to write an outline for a 5-minute shifting briefing presentation. This is the LLM response:
5-minute presentation outline for a DUI enforcement shift briefing
- Introduction:
- Briefly highlight the importance of identifying impaired drivers to protect public safety and reduce accidents.
- Emphasize the officer’s role in early detection and intervention.
- Signs of impaired driving:
- Erratic driving behaviors: weaving, sudden stops, or speeding inconsistencies.
- Delayed responses at traffic signals: hesitation or overly cautious movements.
- Physical indicators: slurred speech, smell of alcohol, or impaired motor skills during interaction.
- Three actionable tips for officers:
- Enhance observations: Actively scan for subtle driving irregularities and unusual behaviors at traffic signals.
- Engage effectively: When stopping a vehicle, ask open-ended questions to assess cognitive responses and listen for inconsistencies.
- Document thoroughly: Keep detailed notes of observed behaviors and indicators to support enforcement actions and potential legal proceedings.
- Closing:
- Reinforce the critical impact of proactive DUI enforcement on community safety.
- Encourage collaboration and communication among officers to share strategies and successes.
Scenario: DUI enforcement
After the quick review of signs of impaired driving, engage the patrol officers in a scenario discussion. Either draw on a real incident from your jurisdiction, or ask your AI assistant to develop a scenario. Depending on your LLM you can either continue in the same thread or start a new thread. I recommend continuing the same thread.
Advanced scenario for group discussion:
During a routine patrol on a residential street at dusk, you notice a vehicle traveling unusually slow, frequently braking despite no obstacles, and making abrupt, unnecessary turns into driveways only to reverse back onto the road. When you initiate a stop, the driver appears visibly nervous, avoids eye contact, and speaks rapidly, giving inconsistent answers about their destination. You also notice an open energy drink can. However, their coordination seems intact, and no strong indicators of impairment are immediately present.
Discussion questions:
- How would you balance assessing the subtle signs of impairment with the driver’s seemingly intact coordination? What additional steps could confirm or rule out impairment?
- How should you document and articulate the driver’s behaviors in your report to support any potential charges or further investigation?
- Given the lack of clear signs, what strategies would you use to determine whether to escalate the situation or release the driver with a warning?
Shift briefing quiz or survey
Finish the shift briefing with a quiz or survey using your organization’s learning management system or knowledge management system. Easy quiz questions ask patrol officers to recall information, like the BAC limit for impaired driving. More complex quiz questions require analysis or synthesis.
In the same prompt thread as the scenario, ask the LLM to suggest quiz questions.
Multiple-choice quiz questions:
- A driver is weaving between lanes at night, braking erratically at green lights and appears to be holding a cellphone. During the stop, they exhibit slurred speech and smell of alcohol. What is the best next step?
- Administer a field sobriety test immediately and arrest the driver if they fail.
- Issue a verbal warning and monitor their driving for further violations.
- Request backup to ensure officer safety before proceeding with any tests.
- Document the interaction and request the driver to exit the vehicle for further observation. (Correct Answer: D)
- During a routine stop, a driver displays nervous behavior and inconsistent speech but has no obvious physical signs of impairment. What additional actions could help assess impairment?
- Conduct a field sobriety test and record results.
- Allow the driver to proceed, noting their nervousness in the report.
- Search the vehicle for open containers without probable cause.
- Ask open-ended questions about their recent activities and observe their responses. (Correct Answer: A)
Cautions when using AI for law enforcement training
This article was written with canvas, an interface built into ChatGPT 4o. The prompts in this article were given to the ChatGPT LLM and the responses are from ChatGPT. As a shift commander, patrol officer or law enforcement trainer you are the best person to assess the accuracy and applicability of the outline, discussion questions and quiz questions. How well did the LLM do at creating useful material for a shift briefing?
You might also, because of your experience and knowledge, be able to write better prompts and fine-tune those prompts. For example, I thought the first DUI discussion scenario was too basic and asked ChatGPT to revise the scenario with this follow-up prompt, “Please try again. The scenario and discussion questions are too basic. The patrol officers are experienced and regularly encounter impaired drivers. I need something that will capture their attention and challenge their thinking.”
In addition to applying your own experience and knowledge, here are three other cautions when using AI for developing law enforcement training materials.
- Think of the AI as an assistant or intern: AI often requires multiple instructions and operates best when treated as part of an ongoing conversation rather than a single task assignment without additional explanation or clarification.
- Trust but verify: AI can generate incorrect information, often described as hallucinations. Just as you would with an intern or assistant, it is your responsibility to review all materials, validate accuracy and consider consulting another shift commander for feedback before delivering information to your patrol officers.
- Policy compliance: Ensure all training content aligns with department policies and procedures, as well as applicable local, state and federal laws. Ultimately, you are responsible for preparing the patrol officers on your shift to deliver effective, ethical and legal policing to your community.
Your experience with using AI to create police training
If you are just beginning to experiment with AI in law enforcement, send us your questions. If you have been using AI to develop police officer training, let us know what is working well and your lessons learned you want to share with other law enforcement trainers. Send your questions and experiences to editor@police1.com.
Police1 is using generative AI to create some content that is edited and fact-checked by our editors.
