ShotSpotter’s results are iffy, which appears to be a combination of a) ShotSpotter employees overriding the AI and b) weaknesses in the tech. There’s also some potential for reinforcing bias if you only put the sensors in neighborhoods you’re already worried about.
The goal of police is public safety, and nothing to do with preventing reinforcing biases.
If you designed a polar bear detector to help prevent people from being eaten by polar bears, would you also deploy them in Florida for the sake of fairness, even though there are no polar bear attacks there?
The problem is that ShotSpotter doesn’t actually work very well at all.
I mean, if I make a magic box that has a 1% chance per day of calling the police and telling them a crime happened nearby and then only stick that box in areas where poor people live, you could see how that would lead to a whole lot of calls for police in those areas. You can also see how, by having a vastly higher police presence in those areas, more crimes will be detected in those areas, even if the base crime rate is the same as in a different area.
In the past, ShotSpotter employees have gone into the system and changed things after the fact to claim that their system detected shots in a specific area that it never detected. They did this while working with the police to come up with probable cause for an arrest after the arrest happened.
This kind of thing makes it easy for the cops to arrest people they want to arrest without any actual evidence of a crime. This is generally considered a bad thing in the USA.
I still encounter people who think a map of crime is an actual map of crime, like SimCity, and not a map of policing. They really do believe it reflects reality. Most times they snap out of it when you point out it's impossible to map crime without collected data, and then it's a short hop to realize policing is how you get that data. And any intellectually honest person will recognize the biases inherent to that data.
These biases are also present in murder convictions, widely used to benchmark crime rates over time and geography due to the inability of police to ignore dead bodies selectively.
It could be possible that police and police chiefs are in fact, putting this equipment in poorer neighborhoods because that is where the most gunfire already is.
Concerns about the relationship between employees and prosecutors are real, but "reinforcing biases" is the least of our worries in regards to marginalized communities when they are murdered at 5 to 20x the rate of the nonmarginalized communities.
I’d like the public to be safe from the public servants who carry guns as part of their day to day work. If those public servants are biased to believe that my neighborhood is less safe than it is, they’re more likely to use force when it isn’t warranted. Thus, I’m not as safe.
I conclude that it is a matter of public safety to avoid reinforcing biases.
Since the article this thread is about shows that some portion of police are willing to believe any damn thing, I think my concerns are reasonable.
> If you designed a polar bear detector to help prevent people from being eaten by polar bears, would you also deploy them in Florida for the sake of fairness
This is already reinforcing bias, because your question presumes that the places law enforcement deploys this tech are where it needs to be. To use your analogy, cops like deploying polar bear detectors based on how much snow falls in a place, because "everyone" knows that polar bears live in snowy places.
To be clear... the goal of the police _force_ is to enforce the law. The goal of (some) individual police officers is public safety, but certainly not all of them (and some days, it seems like not many of them... but you generally only hear about the bad stuff in the news, so take that with a grain of salt).
The goal of the police is to have secure jobs with a decent salary and benefits, and ideally be meaningful in some way (could be promoting public safety... could be exerting power over others).
Often enforcing the law helps secure budgets (especially when it comes to enforcing drug laws with lucrative civil forfeiture seizures), but more recently many police departments have tried a different tactic when budgets have been threatened: actively refusing to enforce the law, in order to increase crime and prompt political change.
They use microphones to detect gunshots. It works in a controlled environment but in an actual municipality the data needs enough "massaging" by human analysts to get what are still fundamentally low-quality tips it's somewhere between redundant and a waste of resources.
> the data needs enough "massaging" by human analysts to get what are still fundamentally low-quality tips it's somewhere between redundant and a waste of resources.
And those analysts are part of a law enforcement community that is rife with biases about crime and guns, which means most of the places where the system finds gunshots are where the biased people (who by default placed the detectors according to a bias) expect to hear gunshots.
If you've ever shot a gun or been around guns you'd know the rapport of various guns is not always consistent. Often, it is difficult to tell the difference between a gunshot and a firework for example. Even something like dropping a wood pallet far enough away can sound like the echos of a gunshot. Backfiring from a car can sound remarkably close to a small caliber rifle. If a gunshot doesnt happen quite literally within a few dozen feet of you it's often hard to tell. Typically the way you can tell is people who shoot guns generally don't do it once. Several rapports tend to raise the probability of a gun.
ShotSpotter works in isolation because in isolation a certain noise profile is almost certainly a gun shot. Like most ML aiming to change the world it's powered by p-hacking.
https://news.ycombinator.com/item?id=28264686 has previous discussion on this from a number of viewpoints.