Predictive policing algorithms are racist. They need to be dismantled
technologyreview.com
technologyreview.com
Law enforcement officers are charged with preventing crime. They are also human beings who learn quickly and use pattern recognition. They would not be effective if they didn't do this.
Pattern recognition after years of serving in law enforcement tells you that certain cars and driving behaviors are suspicious. That people who are behaving out of place in a certain neighborhood at a certain time of day are suspicious. That there are reliable indicators that someone is deserving of higher scrutiny if their behavior warrants.
You would be a fool not to listen to these accumulated and learned indicators of crime prevention.
But now they are being told (or strongly disincentivized) that they must be unbiased. That appearance alone is not sufficient to raise an alert.
Well, unfortunately, that is not true, and no law enforcement officer would tell you that every car driving in a neighborhood is equally likely to be a candidate if a robbery were reported. Or that if a violent drug deal went down, every person on the street would be equally likely to be involved.
Sure, law enforcement personnel should be trained to not allow uninformed or personal biases to come into play. But the job of policing most definitely is a job that involves identifying patterns and learning biases (in the most technical sense of the word) in order to be effective.
You want totally unbiased policing, yet you want police to be effective? Those two things are in conflict, and we need to make some grown-up choices about what is tolerable. An "unbiased" world is a fantasy.
Personally, I think this line is impossible to draw, and therefore we should pay people who are inconvenienced by the police. Police stopped your car? Receive $15. Police busted into your house with a warrant? Receive $1000.
Now we can let the police use any source of bias in their policing decisions, with the knowledge that those being inconvenienced are being paid well to compensate.
However: there's a big difference between stopping a certain type of car after a robbery and stopping certain kinds of cars every day, without any specific reason, until you can charge a black youth with cannabis possession.
A white person hasn't ever been stopped in a "routine" traffic stop and a black youth has been stopped 4 times in the last month and then searched because a cop claimed he "smelled marijuana" -- that is a bias problem.
Reconsidering everything you think you know in every situation is clearly impossible and can only lead to inaction and indecision.
I've worked a few years on spam filtering. There is no way you can achieve zero false positives and negatives. You can only make it easy to appeal mistakes and processes to minimize their impact. The same is true for the legal system.
Some people think the current system it's not preventing crime well. What with the bias and all that. This is underscored by the police brutality in the media.
> You want totally unbiased policing
I don't think anyone has suggested that's even possible. People have claimed there are gradations somewhere between "totally unbiased" and the current state. And some of us would to try out "somewhat less biased" for a while. Just to see if that's better.
This is your fundamental attribution error.
Making crime prevention a fundamental job of police officers places them directly at odds with the community by putting police officers in too many situations where they are "judge, jury, and executioner."
That's a very fundamental problem that all the current protests about the police are trying to drive home.
The only "prevention" that a police officer should be doing is simply being present. Anything else (checkpoints, traffic pullovers without actual violation, etc.) is just asking for abuse.
The problem with algorithmic based approaches is that they often embed naive correlations that apply population statistics to individual people - this is exactly what you don’t want.
If you are a police officer in a hotspot for crime, you start to believe far more humans are criminals than not.
That isn't an accusation, but it requires effort to realize that you aren't dealing with representative groups of people in your day to day job.
That said, the current tone in that debate is probably too poisoned. There are even accusations that crimes weren't followed up because of the religion of perpetrators in Europe. Or that sentences are significantly lighter if you have an immigrant background. If that is the case, it is certainly not helpful in any way.
Is it the one advanced by Critical Race Theory, which is along the lines of: "The marginalization and/or oppression of people of color based on a socially constructed racial hierarchy that privileges white people."
Or is it the one that most people use: "Racism is the belief that groups of humans possess different behavioral traits corresponding to physical appearance and can be divided based on the superiority of one race over another."
These two definitions are mutually exclusive. You can't be a fence-sitter here!
Furthermore, they so different that I think that the Critical Race Theorists need to create a new word to describe the concept they are trying to slip into the language. Why is this important? Well, for starters, racism is a violation of the Civil Rights Act of 1964, and institutions such as Princeton University run into legal trouble when they make use of the new idiosyncratic concept masquerading as actual racism. https://www.scribd.com/document/476441457/Princeton-Letter-0...
There's a third, simpler definition, for what it's worth: Racism is the (erroneous, falsified) belief that there exist multiple human races/species today. This third definition leads to your first definition, because the false belief in races leads to separating people according to those false beliefs, which leads to groups of people harboring prejudice towards each other, which leads to social constructions which embed those prejudices. It also leads to your second definition, because the false belief in races leads to false emphasis on skin tone, facial features, and other innocuous differences between people.
The point isn't to make "racism" or "racist" bad words. The point is to end racist effects in our society. It's inhumane, wasteful, and at odds with science.
Predictive policing isn't racist by any definition. Given an imaginary perfect predictor, one could detect all the potential criminals. Now, it may happen than a part of society (e.g. the pink people) could be over-represented in this group, and algorithm would target them more often. Would it mean that the predictor is racist? In my opinion, it would only mean, that the society itself has some kind of racial bias and the problem is not the predictor's accuracy, but the environment infected with the racial bias towards the pink people.
One could argue that predictive policing would help to sustain this bias and it's probably true, but it doesn't entail that the policing itself is racist. There are many honest and reasonable actions that may inadvertently keep the bias alive. Unless we create a fourth circular definition, I've seen recently in the internet: "Any action/stance that doesn't actively counteract racism is racist". If that's the case, then sure, racism is everywhere.
Anyhow, predictive policing is still a terrible idea, a perfect dystopian future. It should be illegal even if it wasn't sustaining the racial bias (which it probably does in the current form).
2) your third definition doesn't entail any of the two presented by @maxharris. One can easily believe there are many human races (it's a common error) and still not penalize/marginalize based on it and not believe there is any relation of superiority between those (imaginary) races. And I do not believe, one should call such people racists, as they do not harm anyone, they are just mistaken.
Merely to draw the line between people is already enough of a divisive action.
Answering your objection from the pragmatist point of view: obviously their behavior somehow depends on they belief in human races, but it doesn't have to be a negative effect. I believe that my aunt would be extra nice to a black person, as she is very hospitable especially to people from far away and she would assume a black person in her country has to be some kind of immigrant.
A similar case is sex, believing there are two different sexes doesn't entail sexism. Per analogiam: believing there are different races doesn't entail racism.
However, it's still incorrect to stereotype people based on skin color, even if it's polite stereotyping. To think of people with different skin color as different is missing two points: First, that everybody is different from everybody else, and second, that everybody is human! We shouldn't promulgate or tolerate worldviews which get this wrong, because they're deeply prone to fundamental attribution error: Your aunt may well believe that her dark-skinned neighbor does things a certain way because he is an immigrant!
I would also be careful about prosecuting (not tolerating?) worldviews based on their ideas alone, freedom of speech is a very important thing. A thought police on the other thing is a terrible idea and there are scary precedents (e.g. cultural revolution in China). I would rather educate to spread "better" worldviews and prosecute wrongful actions to minimize impact of the "wrong" worldviews.
"Species"-level differences are off the table. Everybody can have babies with everybody else. All the babies can grow up to have babies with all the other ones. Clearly, humans are 1 species.
With "species" addressed the 3rd proposed definition becomes "Racism is the (...) belief that there exist multiple human races".
(Is the existence of race contentious? What is being used for the definition of "race"?)
I disagree that this 3rd definition leads to the GP's 1st or 2nd. Existence of differences does not always imply separating per those differences. Nor does it require placing false emphasis on anything. It can, certainly, but it need not.
The GP's point remains open.
If you can show me a society that is racist but that nonetheless ensures equal rights for all races, then I'll retract the point. However, from what I've seen, every government which thinks that race is real has enacted racist policies, so I think that the burden of proof will be hard to meet. CIMO, caste, aboriginals, reservations, blood quanta, and more; governments enshrine whatever hateful bigotry their people demand.
It is worth pointing out that, even if race did not exist, the Civil Rights Act of 1964 also bans discrimination based on national origin. This covers Central American foreign nationals being neglected and experimented on in our concentration camps; it covers Indian-Americans being rejected for jobs because of the caste of their last name; it even covers African-Americans being pulled over for driving-while-black. Ultimately, most racist practices are netted by the national-origin clause and the concept of race does not need to be invoked.
Why is black being capitalised and not white? Reads very strangely
Not to say that something is good and acceptable if it disadvantages one race over another without intent. It just seems the word racist is being diluted to mean something less.
The article goes on to make the case that the entire process is inherently unsalvageable, which I find less than convincing, but I also am not sure that I trust US police to get it right
This article is also in the context of recent rioting and calls to defund the police, versus data that indicate the basis for these calls is perhaps not so clear-cut
It is a conundrum
It seems to be taken as a given that arrest data is “biased”. I don’t think there’s solid evidence for this. What if this data isn’t biased?
I thought the algorithms were touted as being less racist than the system that was in place. Isn't ripping these algorithmic systems out just reverting back to the old system?
The best part is that they don't even have to run the algorithm, they can just claim they did so[2] to justify their biases.
[1]https://www.youtube.com/watch?v=AJQ3TM-p2QI
[2]https://www.politico.com/news/magazine/2020/08/30/black-us-d...
What people don't seem to mention in these discussions is that these algorithms give cops plausible deniability to harass anyone they wish.
"Sorry, ma'am, the computer told me to strip-search you. Yes, again."
This is the good old "the dog barked, I'm searching your car for weed" trick, except nobody can even see the dog.
You think this is far fetched? It's already happening[1]:
You have been selected for a random computer-generated inspection.
Yes, surely the computer randomly generates an inspection request for a black diplomat every other time she crossed the border. What are the chances, huh. What logs? There isn't a problem.
In a lengthy statement, CBP spokesman Matthew Dyman disputed Spears‘ account of her border crossings, saying that she was referred to secondary inspection during 12 of 43 crossings, and that “all referrals were system generated.“ The language of the statement did not make clear whether CBP keeps records of referrals to secondary inspection made for other reasons, such as an officer’s discretion.
[1]https://www.politico.com/news/magazine/2020/08/30/black-us-d...