AI Use in Crime Reporting by Police: Potential Pitfalls Explored

The use of AI chatbots by some US police departments to generate crime reports is a double-edged sword, promising efficiency but fraught with potential pitfalls. Imagine the allure of cutting report-writing time down to mere seconds, as experienced by Oklahoma City police officers using Draft One, an AI tool that leverages OpenAI's GPT-4 model. It sounds revolutionary, doesn't it? But as with any technological leap, there are risks lurking beneath the surface.

Let's dive into what could possibly go wrong with this tech-forward approach. For starters, AI systems, like ChatGPT, have been known to "hallucinate," meaning they can generate information that's not entirely factual. While Axon, the company behind Draft One, claims to have mitigated this issue by adjusting the "creativity dial," can we really trust that every report will be accurate? It's a gamble, and in the world of law enforcement, accuracy is non-negotiable.

The use of Draft One is currently limited to "minor" incidents in Oklahoma City, with no felonies or violent crimes being processed through the AI. But other departments, such as those in Fort Collins, Colorado, and Lafayette, Indiana, are pushing the envelope by using it across the board. This raises a red flag: if AI is used for all cases, how can we ensure the technology doesn't compromise the integrity of these reports?

Concerns from Experts

Legal scholar Andrew Ferguson has voiced concerns about automation leading to less careful report writing by officers. It's a valid point—relying on technology might make us complacent, and in policing, every detail matters. Furthermore, there's the broader issue of AI systems perpetuating systemic biases. Research has shown that AI-driven tools, if not carefully managed, can worsen discrimination in various fields, including hiring. Could the same happen in law enforcement?

Axon insists that every report generated by Draft One must be reviewed and approved by a human officer, a safeguard against errors and biases. But this still leaves room for human error, which is already a known issue in policing. And what about the AI itself? Linguists have found that large language models like GPT-4 can embody covert racism and perpetuate dialect prejudice, particularly with marginalized languages like African American English. This is a serious concern when it comes to ensuring fair and unbiased policing.

Testing and Accountability

Axon has taken steps to address these concerns, conducting internal studies to test for racial biases in Draft One reports. They found no significant differences between AI-generated reports and the original transcripts. But is this enough? The company is also exploring the use of computer vision to summarize video footage, though CEO Rick Smith acknowledges the sensitivities around policing and race, indicating a cautious approach to this technology.

Despite these efforts, the broader impact of AI in policing remains a question mark. Axon's goal is to reduce gun-related deaths between police and civilians by 50%, but statistics from the Washington Post's police shootings database show an increase in police killings since 2020, even with widespread body camera adoption. This suggests that technology alone isn't the solution to complex societal issues.

As more police departments consider adopting tools like Draft One, the potential benefits and risks will continue to be debated. It's a fascinating time, filled with possibilities but also uncertainties. The key will be to balance innovation with accountability, ensuring that AI enhances rather than undermines the integrity of law enforcement.