The rise of ChatGPT has touched nearly every aspect of modern life, and now, law enforcement is experimenting with its capabilities. But the integration hasn't been seamless. A bizarre incident in Heber City, Utah, revealed a startling flaw: a police report inexplicably claimed an officer had shapeshifted into a frog.
This fantastical detail emerged during testing of two AI software programs – Draft One and Code Four. Code Four, the brainchild of 19-year-old MIT dropouts George Cheng and Dylan Nguyen, aims to revolutionize police work by automatically generating reports from body camera footage. The goal is simple: reduce paperwork and get officers back on the streets.
The frog incident, as it turns out, stemmed from background noise. Sergeant Rick Keel explained to a local news station that the AI picked up audio from the Disney film, *The Princess and the Frog*, playing nearby. This highlighted a crucial lesson: AI-generated reports require meticulous review and correction.
Even controlled demonstrations weren't flawless. A mock traffic stop, intended to showcase the software’s potential, required significant editing. Despite generating reports in both English and Spanish, and analyzing tone and sentiment, the AI stumbled. Yet, Keel estimates the tool already saves him six to eight hours each week, praising its user-friendly design.
Draft One, developed by Axon – the company known for Tasers – utilizes OpenAI’s GPT language models to create complete police reports from audio recordings. However, experts caution that this convenience comes with risks. Concerns are growing that crucial details could be overlooked or inaccurately recorded.
The potential for errors extends far beyond comical transformations. While a frog-related claim is easily identified, misheard street names, misinterpreted chaotic scenes, or incorrect details could have serious legal consequences. Critics also fear the technology could create a shield of deniability, diminishing officer accountability. The Heber City police department is currently evaluating whether to continue using Draft One.
The question remains: is the promise of efficiency worth the inherent risks of relying on artificial intelligence in the critical realm of law enforcement?
