Michael Williams used to be reminded of the fishing excursions by his wife. The grandchildren. On Sunday mornings, he would braid her hair. At Cook County Jail, she tried to pull him back from whatever edge he was approaching by whispering these things into a telephone receiver.
And he was getting close to one. Williams had accumulated pills in his dorm after spending almost a year in prison on charges of a murder he denies committing. He had devised a scheme.
| Category | Details |
|---|---|
| Subject | Michael Williams — wrongfully jailed Chicago resident |
| Age at Time of Arrest | 63 years old |
| Location | Chicago’s South Side, Cook County, Illinois |
| Incident Date | May 2020, during George Floyd protests |
| Technology Involved | ShotSpotter — AI-powered gunshot detection system |
| Time Spent in Jail | Nearly one year |
| Outcome | Case dismissed — insufficient evidence |
| Key Evidence Used | Noiseless security video + ShotSpotter acoustic alert |
| ShotSpotter Cost to Chicago | $9 million per year — $33 million contract (2018) |
| ShotSpotter Deployment Zone | South and West Sides — predominantly Black and Latinx neighborhoods |
| False Deployment Rate | 89% of ShotSpotter alerts found no gun-related crime (MJC study) |
| Communities Affected | 80% of Black Chicagoans, 65% of Latinx Chicagoans live under ShotSpotter coverage |
There was no eyewitness testimony against him. It wasn’t an admission. It wasn’t even a picture of him with a gun. A network of microphones attached to streetlights and telephone poles throughout Chicago’s South Side picked up the sound, which was then analyzed by an algorithm that the company behind it won’t fully explain to anyone, not even the courts. It’s difficult to ignore that for a little while.
Williams left his apartment to buy cigarettes on a muggy May evening in 2020, as Chicago burned in the days after George Floyd was killed in Minneapolis. There had been looting at the gas station. He pivoted. A young man by the name of Safarian Herring stopped him along the route and requested a ride. Williams concurred, claiming to recognize Herring from the neighborhood.
A car pulled up next to them a few minutes later. From the passenger window, a shot was fired. Herring took a hit. In a panic, Williams yelled at his unresponsive passenger, ran a red light, and drove him to St. Bernard Hospital. Days later, Herring passed away. It was three months later. Then Williams’ door was knocked on by the police.
ShotSpotter, a Silicon Valley company’s product that boasts an AI-driven system that can nearly perfectly distinguish gunshots from fireworks, backfiring cars, and construction noise, is the technology at the heart of his prosecution. Currently operating in about 110 American cities, the company charges up to $95,000 per square mile a year.
Chicago is one of ShotSpotter’s two biggest clients, paying $9 million annually for the service. The company is sizable. After the company went public in 2017, its share price more than doubled. The Biden administration even nominated a former ShotSpotter executive to lead the Bureau of Alcohol, Tobacco, Firearms and Explosives. However, the company views the true dependability of the system as a trade secret.
Serious and particular issues were brought up by an Associated Press investigation that used thousands of internal documents and interviews with defense lawyers nationwide. The algorithm used by ShotSpotter has never undergone external peer review. Church bells, motorcycles, dumpsters, and helicopters all caused false gunshot alerts, according to a 2011 study that the company commissioned.
ShotSpotter evidence has been rejected or seriously questioned by courts in California, Massachusetts, and other states. The system failed to detect all seven shots fired during a downtown murder in Fall River, Massachusetts. Within a year, the city pulled the plug.
What occurs in ShotSpotter’s offices may be more concerning than the technical issues. Workers in a restricted room, located approximately 35 miles south of San Francisco, are able to and do alter the classification of sounds after the fact. At the request of police departments, they have the ability to modify the quantity of shots recorded as well as the location that the system initially identified. This is confirmed by court documents; it is not conjecture.
“We have a constitutional right to confront all witnesses and evidence against us, but in this case the ShotSpotter system is the accuser, and there is no way to determine if it’s accurate, monitored, calibrated or if someone’s added something,” stated defense lawyer Katie Higgins when she challenged this practice.
It is not coincidental that this technology has a racial component. ShotSpotter is only used in Chicago’s South and West Sides, which are the twelve police districts with the largest percentages of Black and Latinx citizens. Compared to 30% of white Chicagoans, 80% of Black Chicagoans reside in areas covered by these microphones. 89% of ShotSpotter deployments in Chicago revealed no gun-related crime at all, according to a MacArthur Justice Center study.
Similar findings were made by the city’s Office of Inspector General, which discovered that the system results in tens of thousands of unnecessary police deployments annually. One of those deployments was Michael Williams.
Eventually, he was set free. Last month, a judge dismissed the case after the prosecution admitted they lacked sufficient evidence. For almost a year, Williams sat behind those walls, reducing the number of calls he made to his wife from three per day to a few per week. At one point, he came to terms with not being able to see her. The charges have been dropped. It’s not the year.
When asked about possible flaws in the system, Ralph Clark, CEO of ShotSpotter, stated that human oversight was what mattered and that the question of artificial intelligence details was “not really relevant.” “Eyes and ears must be on anything that eventually results in a gunshot. human ears and eyes.
In theory, that might be accurate. However, it still requires a 65-year-old man to explain to a judge why he shouldn’t be held accountable for a murder that an algorithm determined occurred close to his car and was later corrected by anonymous employees.
Observing all of this gives me the impression that something significant is being overlooked. It’s not just a question of how well ShotSpotter functions. It’s whether criminal charges should be brought at all for a system this opaque, applied so unevenly, and having such dire consequences. For the time being, that question is conveniently unanswered.