Nice, they mentioned the company I founded in college, PredPol. They claim that our algorithm encouraged over policing of minority neighborhoods, but there's no such thing as bad press right?
That sounds like more than "bad press".
I’m not trying to just be ideological here either. I just don’t want the dominant ideology in academia and media to be used to attack and surpress technology that could ultimately benefit the communities they’re allegedly harming.
The closest thing to evidence referenced in the article is this study: http://onlinelibrary.wiley.com/doi/10.1111/j.1740-9713.2016....
It makes some weird comparisons. Why have the heat map of drug users? Police are much more interested in drug dealers, and particularly drug dealers who are likely involved in gangs and violent crime. Unfortunately, those individuals are over represented in minority communities, as are violent criminals more generally.
It also throws around phrases like "historically over-policed communities" without ever actually explaining what that means or why it occurred. There seems to be a general pattern of taking "systemic racism" and equating that to racial discrimination when the two are not the same. There are problems within minority communities which are significant in driving these disproportionate outcomes, and if we just say "racism" and don't look any further that isn't even going to be considered.
If we want to reduce mass incarceration of people for nonviolent crimes, we ought to be reforming the laws, not reducing enforcement just to produce conviction stats we like more. That will almost certainly get people killed.
Importantly, these algorithms should not be secret. These are the kinds of choices a democratic society should make.
Ultimately we have to work with what we have and the police are responsible for maintaining high quality data and ethical standards, but I don't lose any sleep worrying about if my code systematically oppressed minorities for two reasons:
1.First of all, most police departments had separate predictions run per "district" and typically had a fixed number of officers allocated to each district. The idea that we were directing cops away from rich neighborhoods and having them go after the poor or minorities assumes that we had far more power than we actually did. The number of cops assigned to each city district / beat is a political or bureaucratic decision and not something our software is involved in.
2. Furthermore, our app was not directing the cops every move, nor was it trying to. We just told them "when you're not answering a call or doing anything else, when you have a spare minute, check if you're near a prediction box and go figure out why the computer flagged it as a high risk area. Maybe it's a parking lot and the lights have gone out encouraging car burglaries". And even as just a small slice of their time, we really had to push to get them to use the app because the way they saw it was "What does this stupid app know about my city that I, with 20 years of experience, don't know? One of these prediction boxes is in the middle of a lake". Based on what I saw from the analytics and usage logs, I think it would be tough to make a case that anyone was oppressed because the cops barely even used the thing! Not that I take these issues lightly, but I always laugh when people criticize PredPol or someone calls me a murderer on Twitter because it vastly overstates our impact. Maybe I should take it as a compliment that people really think I am powerful enough to direct police to oppress minorities using only a shitty rails app I wrote in my college dorm room. But whatever, you can't do or try anything new in this country today without getting dramatically criticized for it to fuel the internet-rage clickbait economy, so I guess it's just something all of us have to get used to. I haven't had a day to day role at the company for a few years, which makes it a lot easier not to worry about these things as well. But ultimately police using technology and data effectively can REDUCE bias and injustice AND make people safer. I think the questions the Verge presents are certainly worth asking, but based on my experience I do believe that the app I implemented helped more people than it harmed.
If I'm reading your two points correctly, you don't worry about whether or not you made a bad tool because it wasn't ever clear how much it was used?
Depends on how you use the data. If you learn success rates then it shouldn't depend on the absolute number of people searched. If you just feed raw data into machine learning algorithms, then nobody knows.
E.g. you have 100 stop and frisks, 50 in one and 50 in antother neighborhood. Then you might see that you got 1 criminal in one and 10 in the other neighborhood. Then your algorithm will allocate more forces (shouldn't go below a minimum, otherwise no data) in the 20% criminal neighborhood than in the 2% criminal neighborhood.
Obviously this will still lead to more searches in more criminal neighborhoods. But that's what you want. You want to use the police force as efficiently as possible to fight crime.
In one place you can get off with a warning for violating some traffic rule and in other places you can get arrested and hanged. [1]
On body cameras and separate evaluation (not by the police department, but by a separate agency with a separate budget).
Is that a thing that even happens? Most PDs I'm familiar with have a lot of control over their own cams. Even after getting caught using them to plant evidence, many still have some questionable cam practices.
NB: stop and frisk isn't done in the "good" neighborhoods at all. There's not going to be any data about what happens when e.g. stopping and frisking white men in suits in lower Manhattan because it will never be done. The idea of being able to do controls hinges on something for which there is no political will.
> Obviously this will still lead to more searches in more criminal neighborhoods. But that's what you want.
I would think instead you want more searches in places where there is probable cause, you have gotten warrants, etc. How comfortable are we with the notion that people should be targeted on the basis of the neighborhoods they live in? Is this compatible with our guiding documents?
I already addressed that caveat by forcing the algorithm to stop and frisk a minimum amount in the good neighbourhood. Like a minimum of 10% of frisks taking place in good neighbourhoods.
>How comfortable are we with the notion that people should be targeted on the basis of the neighborhoods they live in?
If I live in such a neighbourhood, I'd want more police presence. They will stop and frisk and do that to me, too. At the same time they will have the chance to stop some criminals that might kill or rob me and my family.
That's exactly my point. It's fine to say that, knowing full well that there is absolutely no way this would ever happen. This line of thinking guarantees unequal application of the law based on neighborhood or other signifiers (in the US: ethnicity).
> They will stop and frisk and do that to me, too.
This is not true generally in the US. The way stop and frisk works is the police identify brown men and stop and frisk them. (It's worse if they happen to be in one of the "bad" areas, but in practice the main thing was they are young-looking brown men.) The stop and frisk program would have been stillborn if the NYPD were applying it evenly to people walking around New York City.
Some of this has been addressed by other comments. But in short: there are a lot of ways to break the law other than stabbing or shooting.
Just as an example, I would wager that a stop and frisk in the area around Wall Street could yield lots of drug violations. These crimes are pursued vigorously in other parts of New York City. Why is it okay to stop and frisk some people in some neighborhoods, but not other people in other neighborhoods, when looking for the same violations?
Stop and frisk is sort of a shitty policy in my opinion because even if it were limited to weapons, some people may be carrying them to defend themselves. Laws against this are really shitty if you were to place yourself in the shoes of someone who is at constant risk of violence and just wants to defend themselves. If someone intends to harm you with a weapon, there’s little chance the police are going to intervene in the situation in time.
But outside of stop and frisk and just looking at policing in general, enforcing laws against nonviolent actions while you’re in areas allegedly because you’re trying to prevent violent crime is a huge driver of the disproportionate convictions of minorities. In my opinion, reducing police presence in violent areas is one of the worst solutions to this problem. That shows disregard for the people living there, regardless of how they may feel about the police. Reforming dumb laws that are getting relatively innocent people prison time would be far better. We need to have laws that allow police to be effective in combating violence without harming the entire area they’re policing with frivolous laws.
Exactly
I’m kind of interested as to what the definition of “over-police” is in this context, though. Is it disproportionate with or without the consideration of crime levels?
It seems like over correcting for this problem could be very bad for the communities it’s intended to protect.