ChatGPT as the New Photofit: US Police Explore AI for Suspect Images

AI Replaces Sketch Artists: US Police Embrace ChatGPT for Suspect Identification

In a groundbreaking shift, law enforcement in the United States is beginning to leverage the power of artificial intelligence, specifically Large Language Models like ChatGPT, to aid in suspect identification. This innovative approach moves beyond traditional composite sketches, aiming to create more realistic and potentially more effective visual aids for public appeals.

Phoenix Takes the AI Lead

The Phoenix Police Department in Arizona recently made headlines by publishing an AI-generated image of a suspect in a shooting incident. The image depicted a middle-aged man wearing a hoodie, cap, and beard, based on descriptions provided by a victim and witnesses. Unlike conventional police sketches, this AI rendering was explicitly stated to be an AI creation and not a depiction of a real individual, highlighting the nascent and experimental nature of this technology in law enforcement.

While AI has been a tool in policing for some time – assisting in evidence sorting, facial recognition analysis, and administrative tasks like report writing – Phoenix's adoption of AI for generating suspect likenesses marks a significant departure. This move replaces the age-old practice of hand-drawn sketches with sophisticated AI image generation.

Questions of Reliability and Legal Hurdles

This technological leap, however, is not without its skeptics. Experts voice concerns that relying on AI for suspect identification could introduce new biases, distort already fallible identification processes, and potentially lead to contentious legal battles. The accuracy and ethical implications of using AI-generated images as evidence or in public appeals are subjects of intense debate.

Despite these reservations, proponents within law enforcement see immense potential. Officer Mike Bonasera, who has five years of experience creating traditional sketches and now oversees the use of AI in this capacity, views image generators as a powerful evolution. "We live in a time when people won't pay attention to a pencil drawing posted online," he argues. Bonasera points out that traditional forensic identification requires specialized training in facial features, witness interrogation, and often a certification exam, a demanding process for a task that may only yield a handful of sketches annually.

Experimenting with Generative AI

Bonasera's initial foray into using ChatGPT for sketch generation involved inputting older, existing sketches into the model. The results were reportedly astonishing, producing images that bore a striking resemblance to actual suspects. This success prompted his proposal to integrate AI into what was previously known as the "photofit" or composite sketch process.

The Phoenix Police Department expressed optimism in their announcement: "We hope these new methods and AI technologies will help solve more complex cases in the future, here in Arizona and across the country." Early reactions from witnesses have been positive, with many finding the AI-generated images easier to visualize and allowing for more precise feedback during the refinement process.

No Arrests Yet, but the Future is AI-Driven

While the Phoenix experiment is still in its early stages and has yet to result in any arrests, the images are being disseminated to gather potential leads. However, other instances of AI misuse in policing have emerged, casting a shadow of caution. The Westbrook Police Department faced criticism in July for publishing distorted evidence generated by ChatGPT, even after requesting a simple police seal to be added to an image. This incident underscores the need for careful oversight and understanding of AI's capabilities and limitations.

Meanwhile, in a separate instance, ChatGPT reportedly assisted a car owner in locating a Lamborghini that had been stolen two years prior, showcasing a different, positive application of AI in resolving cases.