AI 'Shot Spotter' that should detect gunshots and reduce crime is confusing the city



In Chicago, Illinois, USA, an AI-driven gunshot detection sensor called

'ShotSpotter' is installed to eradicate gun violence and improve security. ShotSpotter detects gunshots and sends police information as an alert about where the gunshots occurred, but in recent years it has been pointed out that ShotSpotter has not helped reduce gun violence and has caused false accusations. I am.

Police Are Telling ShotSpotter to Alter Evidence From Gunshot-Detecting AI
https://www.vice.com/en/article/qj8xbq/police-are-telling-shotspotter-to-alter-evidence-from-gunshot-detecting-ai

In the case of 25-year-old Safariin Herring being shot dead in the head on May 31, 2020, a 64-year-old man named Michael Williams was arrested as a result of the use of ShotSpotter. At 23:46, when the incident happened, a ShotSpotter on the street detected a 'sound.' Initially, ShotSpotter's algorithm recognized the sound as 'the sound of fireworks.' At that time, protests against George Floyd's death were taking place and were thought to be fireworks related to the demonstration.

However, afterwards, ShotSpotter analysts disabled the algorithm, switched to manual and analyzed, and determined that the sound was a gunshot. A few more months later, when the case was investigated, another ShotSpotter analyst determined that the coordinates where the sound was detected were the plot where Williams was at the time. Williams, who was in the car, was arrested as a suspect from the video of the surveillance camera. Williams claimed that Herring was shot by a car running away.



At the trial, Williams' public defender asked the judge to consider the legitimacy of ShotSpotter as legal evidence. The public defender said, 'By this way of involving people, the output of ShotSpotter is dramatic from'data not related to any criminal case'to'data that is the centerpiece of murder cases related to Mr. Williams'. It has changed. ' Eventually, the prosecutor withdrew all evidence of using ShotSpotter against Williams.

ShotSpotter is used in more than 100 cities in the United States and generates 21,000 alerts annually. Such cases are not uncommon, as ShotSpotter analysts often manually analyze the data detected by the algorithm at the request of police.



However, Sam Klepper, senior vice president of marketing and product strategy at ShotSpotter, said that 190 trials have allowed ShotSpotter's technology and employee testimony, and prosecution in Williams' trial. Argued that withdrawing evidence of using ShotSpotter did not mean that it had lost confidence in its technology.

'More than 90% is mechanical analysis. In a very small number of cases, customers ask us to verify the accuracy of the location,' Klepper acknowledged, admitting that there are cases where analysts analyze coordinates manually rather than algorithms. We have a request for analysis. If there is an error, we will tell you a more accurate location to assist in the investigation. '

ShotSpotter is to have 97% accuracy (PDF file)

will not argue , but this number is not calculated by the engineers, even that is guaranteed by the sales and marketing team, where it is feared. In fact, in May 2021, the MacArthur Judicial Center conducted a data analysis of ShotSpotter, and found that 89% of the alerts that occurred in Chicago in the last 21 months were not evidence of gun crime, and 86% were evidence of any crime. It shows that it did not, and its effectiveness is doubtful.



In addition, research results have increased that 'deploying ShotSpotter does not reduce crime in cities.' A study from Southern Illinois University in Edwardsville found that 'in St. Louis, ShotSpotter has little effect on the control of gun-related crimes, and there is no consistent reduction in police response time, which does not help produce practical results. It is written. In fact, some customers cancel their contracts with ShotSpotter because they have a lot of false alarms and a poor return on investment.

Chicago is one of ShotSpotter's main customers, with 94,313 shooting alerts between 2017 and 2021. This is an average of 29,58 cases per year.

In April 2021, 13-year-old Adam Toledo, who raised his hands in Chicago, was shot dead by police officers, and the Shot Spotter alert was also issued in this case. 'This kind of tool sends more police to black and Latino areas. ShotSpotter alerts black and Latino police officers,' said Alix Goodwin, an action center on race and economics. I'm at risk of getting involved. That's what happened to Adam Toledo. '

In fact, news media Motherboard has shown that despite the fact that gun crimes in Chicago are happening all over the city, ShotSpotter's sensors are only located in areas where blacks and Latinos live. It was said that it was. It seems that there were no sensors in areas where white people live intensively, such as the northern and northwestern parts of the city.

in Software, Posted by darkhorse_log