Artificial Intelligence and the New Risks of Automated Policing – Saul Roth

Image Credit: MateMedia

By Saul Roth

Artificial intelligence is changing how law enforcement agencies analyze information, identify suspects, allocate resources, and review evidence. These tools may help departments manage large volumes of data, but they also raise difficult questions about accuracy, bias, privacy, and accountability.

Facial-recognition systems, predictive-policing software, automated license-plate readers, and generative AI tools can influence investigations before an officer ever speaks with a suspect. If an algorithm produces an inaccurate match or labels a neighborhood as high risk, that output can affect surveillance, stops, arrests, and sentencing decisions.

The problem is not only whether a system is technically accurate. Accuracy can vary across demographic groups, lighting conditions, image quality, and database composition. A tool that performs well in a laboratory may still produce harmful results in real-world settings.

Predictive systems also learn from historical data. If past enforcement was concentrated in certain neighborhoods, an algorithm may interpret that enforcement history as evidence that those neighborhoods require even more police attention. The result can be a feedback loop that reinforces existing disparities without proving that more crime is occurring.

Generative AI creates additional challenges. Agencies may use it to summarize reports, draft documents, organize evidence, or communicate with the public. Those applications can save time, but fabricated details, missing context, or undisclosed automated changes could undermine criminal cases and public records.

Every agency using AI should maintain a written policy that addresses:

• Which tools may be used and for what purposes
• What human review is required
• How accuracy and bias are tested
• How affected individuals can challenge an automated result
• How long data is retained
• Who is responsible when the system is wrong
• When prosecutors, defense attorneys, judges, and the public must be notified

AI should assist professional judgment, not replace it. No person should lose liberty solely because an opaque system generated a prediction or identification. The basic rule should be simple: the more consequential the decision, the more explanation, human oversight, and independent review it requires.