Generative AI is changing the way we think about writing, and people are taking notice. ChatGPT set multiple new records for how quickly they built their subscriber base — the platform reached 1 million subscribers in just 5 days and over 100 million users within 2 months.
While many civilians use generative AI for fun purposes — like making up rap songs about their dogs or creating digital art — organizations around the world have been jockeying to find innovative applications for AI to solve major customer problems.
As public safety agencies begin to explore the use of generative AI, it is essential to understand the differences between consumer-grade and public-safety grade tools. Not all generative AI is created equally, and there are many solutions on the market that leverage generative AI for specific use cases.
Draft One is one such AI solution, designed to draft police report narratives based on body-worn camera audio. Draft One is a force multiplier for agencies that can save officers hours of paperwork each week.
We’re averaging 67% time savings in our report writing. That is 67% of time the officer is getting back to go out into the community, get back out to do things that they really enjoy... and being engaged with our citizens.
-Sergeant Bob Younger, Fort Collins PD
In this article, we will explore several key differences between Draft One and consumer-grade generative AI solutions. See why Draft One is ideal for public safety use cases, and learn the specific design steps we took to optimize this solution for law enforcement officers.
Consumer-grade vs public safety-grade technology
For decades, public safety agencies have leveraged tools that are purpose built for the specific needs of law enforcement officers.
There is a big difference between a P25 radio used by officers and a walkie talkie that consumers can buy at an electronics store. Similarly, while many civilians leverage dashboard cameras, these consumer cameras don’t have the same features and benefits as public safety-grade camera systems like Axon Fleet 3, which includes ALPR and live-streaming capabilities.
This isn’t just true for hardware — it’s true for software as well. This case study about an agency in Ottawa, Canada, shows why it’s so important for investigators to leverage a public safety-grade video player like Axon Investigate, rather than a consumer-grade video playback tool.
So it’s no surprise that when it comes to generative AI solutions, many officers have expressed concerns about using consumer AI tools to draft report narratives. These consumer tools tend to have less security, create hallucinations and can even demonstrate racial bias. AI can be a powerful tool, but it must be developed in a responsible way, especially when it comes to AI for public safety.
Enter Draft One: a public safety-grade generative AI solution that was specifically designed to assist with police report writing in a way that is secure, efficient and unbiased. Draft One addresses these concerns having its creativity turned off to prevent speculation or embellishments, ensuring that the generated narratives remain factual and reliable. Police officers currently spend up to 3 hours per day on paperwork. With Draft One, officers can cut that time in half. Let’s take a look at some of the key differentiators that make Draft One a good fit for agencies globally.
Data security
Many popular AI tools are data-driven and data-hungry. In fact, there are several lawsuits currently ongoing about how AI companies use their customers’ data. When officers use consumer-grade AI tools, the data they share may be used to influence future iterations of generative AI, and may even be shared with third parties.
Police data is sensitive in nature, so agencies need to protect it.
With Draft One, agency data is securely housed within Axon Evidence. If we need to test with real customer data, we request permission to enroll customers in our voluntary, privacy-centric program — all while working within the confines of our data sharing agreement.
Sticking to the facts
Consumer-grade generative AI tools typically leverage an enormous pool of data when drafting content. These tools can be extremely creative, and can use information that may or may not be true. Hallucinations are a real concern when using large language models.
With Draft One, the large language model’s creativity is turned off and report narratives are drafted strictly from information found in body-worn camera video transcripts.
The underlying model used for Draft One transcription is Open AI’s GPT-4 Turbo. We calibrated the model to prevent speculation or embellishments like those often found in consumer-grade AI output. Draft One report narratives stick to the facts and require officers to review and approve each narrative draft for accuracy
I was concerned that it was going to start making assumptions to fill in the blanks. But the way Axon has worked on this model and turned the creativity down, there are no assumptions, no unusual statements — just the facts, which I really like.
-Sergeant Bob Younger, Fort Collins PD
Keeping humans in the loop
Most generative AI tools do not require users to review and approve the text. Additionally, there are no controls in place to stop users from copying and pasting the generated text and claiming it as their own, or from blaming the AI for any mistakes made.
With Draft One, several safeguards are built into the workflow to keep humans in the loop:
Draft One does not add speculative details into a report narrative. If key information does not appear in a video’s transcript, Draft One requires users to insert that info manually. Officers cannot copy and paste Draft One report narratives into their full report until all sections of the narrative that require additional information have been reviewed and updated.
Once the report narrative is complete, officers must then sign off on the narrative’s accuracy and attest to their ownership of the report before submitting.
Industry-specific standards
Many generative AI tools on the market today lack consistency. It can take time, effort and guesswork to get the AI to deliver a usable final product, and many users give up before acquiring the output they need.
Draft One takes the guesswork out of AI. Draft One is a closed system with industry-specific safeguards. Officers don’t need to come up with directions or questions for the AI, they simply identify the file they would like to use to draft their report narrative and click the ‘Draft One’ button. From there, a high-quality report narrative draft is generated automatically. We trained Draft One to write great police reports using the best training resources available, so officers can focus on what’s important.
Before launch, we put Draft One to the test to ensure we were meeting or exceeding quality expectations. We found that, out of five comparison qualities, Draft One matched the quality of an officer-only report in three categories, and surpassed the quality of an officer-only report in the other two categories.
We make sure we are checking our work when we do AI report writing, but ultimately it is a clean process and some of the reports that we’ve seen are maybe better than what an officer could write.
-Chief Scott Galloway, Lafayette PD
Axon Ecosystem Integration
Today, many police officers use consumer-grade generative AI tools to draft report narratives. These tools pose a variety of risks to agencies, as outlined in the previous section, but they are also less efficient than Draft One. Officers must generate transcripts of video evidence, then copy those transcripts into another platform before finally generating their report narrative. This process can take hours or even days.
Draft One is fully integrated with the Axon Ecosystem, meaning officers can draft report narratives faster than ever before. As soon as an officer stops recording with their Axon Body 3 or Axon Body 4 camera, the audio from the recording begins uploading to Axon Evidence via LTE. Within 5 minutes, the audio has been transcribed, and the transcription is ready to be used to generate a report draft.
With Draft One, officers simply log into Axon Evidence or Axon Records, find available recordings and click to generate a draft report narrative. The entire process is simple and fast.
My experience using Draft One really blew me away. The normal report: your thefts, your shoplifting, your damaged property reports, it really does a good job of breezing through those.
-Officer Nathan Stoneking, Lafayette PD
Responsible AI for modern policing
Draft One is not the only tool leveraging AI to make police work more efficient. To learn more about how your agency can responsibly leverage AI to force multiply your officers, check out our White Paper: Responsible AI for Modern Policing.
To learn more about how to rewrite report writing with Draft One, watch our webinar or contact us for a free demo or trial.
:format(webp))