The Mystery of AI Gunshot-Detection Accuracy Is Finally Unraveling

Liz González’s neighborhood in East San Jose can be loud. Some of her neighbors apparently want the whole block to hear their cars, others like to light fireworks for every occasion, and occasionally there are gunshots.
In February 2023, San Jose began piloting AI-powered gunshot detection technology from Flock Safety. During the first four months, Flock’s system alerted police to 123 shooting incidents. However, data from San Jose’s Digital Privacy Office showed that only 50% of those alerts were confirmed gunfire, while 34% were false positives. After recalibrating the sensors, 81% of alerts were confirmed gunshots, 7% were false alarms, and 12% were indeterminate.
San Jose is unique in requiring its police department to disclose accuracy data for its gunshot detection system. This report, released on May 31, is the first time such data has been published.
False positives are a significant concern for communities of color, who fear unnecessary police presence. González, an organizer with Silicon Valley De-Bug, emphasized the potential danger of any police interaction in their community.
San Jose did not provide data on false negatives but acknowledged that the system does not detect all gunshots. Flock Safety claims its system is 90% accurate, while SoundThinking’s ShotSpotter, the most popular system, claims 97% accuracy. However, data from other cities like Champaign, Illinois, and Chicago show much lower confirmation rates.
New York City’s comptroller recently published an audit showing only 13% of ShotSpotter alerts were confirmed as gunfire. The NYPD has not published accuracy data despite having the necessary information, leading to criticisms of inadequate accountability.
Champaign and Chicago have canceled their contracts with Flock Safety and SoundThinking, respectively. Flock Safety defends its system's accuracy and highlights the benefit of alerts that would otherwise go unnoticed.
Research by Eric Piza at Northeastern University showed faster police responses and more ballistic evidence collection with gunshot detection alerts but no reduction in gun-related crimes or increased crime-solving rates.
San Jose’s Digital Privacy Office recommended improving the system’s accuracy. Despite the issues, the SJPD plans to continue using the technology, citing its usefulness in providing supplementary evidence for gun crimes.
Community advocate Darcie Green and organizer González both emphasize the need to evaluate whether such technology genuinely enhances safety or if resources could be better invested elsewhere.
For the full article, visit WIRED.