By Ethan Gregory Dodge
It was near freezing temperatures in the early hours of the morning when Chicago police received an alert from their gunshot detection technology (GDT) solution, ShotSpotter. Apparently shots had been fired in the city’s Little Village neighborhood, an area predominantly populated by Latinx individuals. Officers rush to the scene to find 21-year-old Ruben Roman and 13-year-old Adam Toledo in an alley. They managed to aprehend Roman, but Toledo slipped away.
An officer caught up to Toledo and ordered him to stop. He complied. The officer demanded Toledo put his hands up. Again, he complied, dropping a gun at the same time. The officer fatally shot Toledo less than a second after he raised his hands.
The story and accompanying footage of the boy’s death caused public outrage across the country. It also resurfaced the debate around the privacy, efficacy and moral concerns of GDT, a technology that has recently come to San Jose.
The San Jose Police Department is currently testing gunshot detection technology in two San Jose neighborhoods: Clemence/Owsley, just south of Little Saigon in East San Jose, and Cadillac/Winchester in West San Jose. The systems, dubbed OnSound by their manufacturer V5 Systems, are mounted to about every other street light. The Fremont-based company claims to leverage artificial intelligence “to detect a broad range of gunshot types,” promising “up to 90% accuracy” without providing any of the data used to reach such a figure.
The tech is also accompanied by V5 Systems’ OnSight automatic license plate reader, effectively logging every car that enters and exits those neighborhoods.
Unfortunately, “artificial intelligence” isn’t a regulated term. Often used as marketing jargon, it can be translated into something akin either to “privacy nightmare” or “disingenuous tech.” ShotSpotter, the GDT used in Chicago, also touts use of AI, yet Dana Delger of the Innocence Project found otherwise in the case of her client Silvon Simmons, who was wrongfully convicted of illegal posession of a firearm based on ShotSpotter recordings. “At the end of the day this is a machine that basically can tell you that there was a loud sound and then a human has to tell you whether it was gunfire or not,” said Delger.
ShotSpotter, being the most widely used gunshot detection solution in the country, has faced similar questions having undergone many third-party audits. Perhaps the most notable one was conducted by New York University Law School’s Policing Project in 2019. The report provided several privacy-preserving recommendations, one being to “improve internal controls and supervision regarding audio access,” which ShotSpotter claims to have implemented.
The report also concluded “that the risk of voice surveillance is extremely low in practice,” quelling concerns that ShotSpotter tech could be used to eavesdrop on conversations.
If V5 Systems has engaged in a similar audit, I was unable to find the results, nor did they respond to my questions attempting to clarify these privacy concerns. SJPD has claimed that only they will have access to the video also captured by the system, but that claim contradicts V5 Systems’ public documentation.
However, privacy is only one side of the coin. The efficacy of these systems could be the difference between life and death. Dave Maass, director of investigations with the Electronic Frontier Foundation, told the Las Vegas Review Journal just two months ago that upon receiving an alert from a gunshot detection system such as ShotSpotter police “might actually assume this is a violent situation with an active shooter, so they are going to go in ready to deal with a shooting situation… we are potentially putting people into a life-or-death situation.”
ShotSpotter’s accuracy and efficacy claims were put on trial in a San Francisco court in 2017. When asked about the company’s guarantee of accurately identifying gunshots 80% of the time, a ShotSpotter engineer stated that the “guarantee was put together by our sales and marketing department, not our engineers.”
Furthermore, research released earlier this year by Northwestern University’s MacArthur Justice Center showed that only 10% of ShotSpotter alerts in Chicago from July 2019 to April 2021 led to police filing an incident report involving a gun. Yet ShotSpotter claims to have a “97% aggregate accuracy rate.” As stated earlier, V5 Systems’ OnSound claims an ambiguous accuracy rate of “up to 90%.”
If research and analysis shows that a decades-old and widely-deployed product is still not living up to its accuracy claims, why is SJPD testing a system that has seemingly faced zero scrutiny? For a police department who has killed 20 people in the past five years, the most recent of which, Demetrius Stanely, being just last week, I would hope they’d be more concerned about verifying their own tech’s accuracy.
Privacy and efficacy aside: does this tech belong in our city?
In the case of Adam Toledo, ShotSpotter got it right. The other man arrested at the scene, Ruben Roman, had fired a gun. But the tech still led to a situation in which an innocent, Black seventh grade student was killed after complying with police. Even if the technology was always 100% accurate and perfectly preserved privacy, it would still lead to deadly situations. It has no place in our city.
To make your voice heard on SJPD’s pilot program of these technologies, please sign this petition demanding more transparency around the technology, its claims and the department’s intentions.
San José Spotlight columnist Ethan Gregory Dodge is the founder of the Citizens Privacy Coalition of Santa Clara County. He is also the creator of Surveillance Today, a weekly newsletter and podcast discussing current events in surveillance. His columns appear every second Wednesday of the month. Contact Ethan at email@example.com or follow @egd_io on Twitter.
The post Dodge: Gunshot detection tech will lead to more police killings in San Jose appeared first on San José Spotlight.
This article was originally published by The San José Spotlight.