An artificial intelligence system will only make a model based on training data. If there is a bias reflected in this training data, it will be reflected in any predictive model based on the...
An artificial intelligence system will only make a model based on training data. If there is a bias reflected in this training data, it will be reflected in any predictive model based on the training data. The AI system in question (PredPol) inherits the bias reflected in the training data.
The problem is people view these systems as an infallible crystal ball when all they do is advanced mathematical guesswork. In analytics the heavy lifting is done in the interpretation, and work should be done to determine biases in the data, it is literally the first step.
This is, IMO, a clash of hard and soft problems. "Hard" being the analytics/math stuff, "soft" being the human stuff, like what causes increased crime/reporting in some areas (not gonna start listing things because it's a mess). The soft problems need to be fixed first, and new data accumulated based on this new system, or we will use what amount to dumb pipes to reinforce biases and systems of oppression by using data that reflects the result of these bad systems.
At risk of sounding like a broken record: in predictive analytics you can only see the past and guess the future, which means a poorly chosen application of predictive models will only perpetuate the environment the data came from.
This is just a reflection of conversations that have been happening in AI ethics for a few years now. Nearly all publicly available data sets are deeply flawed. GPT-3 is incredibly racist towards...
This is just a reflection of conversations that have been happening in AI ethics for a few years now. Nearly all publicly available data sets are deeply flawed. GPT-3 is incredibly racist towards muslims. Healthcare algorithms are racist towards blacks and other minorities. Image searching technology is racist and sexist. Many large bodies have tried to weigh in on how to correct and accommodate these biases, but we are many years away from dealing with these problems on a systematic scale and especially through the lens of currently-available large AI offerings. The problem, in short, is that capitalism has found a lot of reasons to employ AI rapidly, without fully understanding what the scientists have to say about it. A broken algorithm can still make you money and make you more efficient, while simultaneously amplifying existing biases that come from the data set.
As someone who works in data science, I've been watching this conversation evolve for some time now and I'm becoming increasingly more worried about the state of affairs. Luckily there are plenty of researchers refining the techniques of algorithms to allow 'fixing' of these AI algorithms by incorporating additional weighted steps and curated data sets, but these are not easy, nor cheap fixes. We need to be ever vigilant about the AI that we see, and perhaps the most important question we should be asking about them is the data on which they are trained. What biases may lie in the data that we see that are obvious, and what biases may be hidden within cultural values and internalizations we create from our exposure to the world and to large data sets which reflect these values?
It seems like there's an underlying assumption that more police presence is bad. I think we all know reasons why that assumption is made, and the article provides a few more. But if you think...
Exemplary
It seems like there's an underlying assumption that more police presence is bad. I think we all know reasons why that assumption is made, and the article provides a few more.
But if you think about it, it's kind of odd. Why should providing more of a government service be bad? Isn't that weird? Shouldn't we fix that? If people in certain neighborhoods call the police more often, they should get better service.
If we thought more policing were good, the results of using these crime prediction tools would be interpreted differently. They would be biased in favor of certain minority neighborhoods that are more in need, so they get better service.
If sending the police is bad on average, no crime prediction tool is going to help. I guess the idea behind getting rid of the prediction tools is that we can't fix the police, so better to limit the damage by spreading police presence evenly across more neighborhoods, so everyone is has an equal chance of being punished? It seems kind of bleak.
I'm thinking that the people who came up with the crime prediction tools probably didn't make this bleak assumption. They probably thought more police presence is helpful, so the kind of bias described in this article isn't a problem. This is... optimistic. But most people don't assume they're making the world worse. Despite all the protests, I would guess that most people in law enforcement still don't think that doing their jobs makes things worse?
The article gives us a couple stories explaining mechanisms for how more police presence can be bad. They describe how a "crime-free housing ordinance" can get someone evicted. In another case, someone is stopped for smoking in a non-smoking area, and the police officer finds they have an unpaid fine.
These sound like things the police did according to laws that were passed on purpose, but had bad effects. Maybe that's more important to fix than the crime prediction tools?
Yes! But we can't. Politically, it is much more difficult to remove or significantly reform an existing system than it is to prevent a new one from being put in place. This is absolutely true, and...
Why should providing more of a government service be bad? Isn't that weird? Shouldn't we fix that?
Yes! But we can't. Politically, it is much more difficult to remove or significantly reform an existing system than it is to prevent a new one from being put in place.
If sending the police is bad on average, no crime prediction tool is going to help.
This is absolutely true, and that's the position the majority of reformists are coming from these days. We have been trying to reform the police, in a broad sense, since the early 2010s; a large number of specific measures have failed to make it through local, state, and federal processes over that time. I'm sure that reform effort will continue, but when something like this shows up, we have to fight it.
We can't just say, "oh well, this would be really bad because it'll increase the punishment load on the underprivelaged, but we won't fight that because we want to spend all our resources on reforming the police as whole." We know that real police reform, getting to a point where sending the police is more likely to produce a positive outcome than a negative one, is going to take a very long time, if it's even possible. Therefore, we have no choice at all but to fight this at the same time; to prevent short term regressions as we work towards long-term progress.
EDIT: I just wanted to address this.
It seems kind of bleak.
Yeah, man. It sucks. I hate it here. I think this is where a lot of the Left-doomers come from; examining systems like this and looking at the history of attempts to change them, since the 1870s, is... really, really, really, depressing. There have been victories - even recent victories! - but the general trend is that, with a few, mostly local blips over time, the police have gotten more money, more power, and more immunity. The police regularly defy the mayors of the cities they supposedly serve, and their unions - which I would argue don't really count as unions, since they don't actually have bosses to unionize against - consistently block any kind of accountability or reform.
Read the first half and skimmed the rest. But these findings seem.. obvious? It would seem intuitive that poor areas have higher crime levels. Don’t know if this was buried in the body, but were...
Read the first half and skimmed the rest. But these findings seem.. obvious? It would seem intuitive that poor areas have higher crime levels.
Don’t know if this was buried in the body, but were the predictions wrong? I would be more interested if they could show that certain groups get a greater number of false positives than others. If the system is pretty close to predicting the truth, then I sounds like it’s working well.
TBH, a response exactly like yours, basically a "TL;DR but here's what I think about this subject anyways" was what I feared most when posting this. I would highly recommend you don't just skim...
TBH, a response exactly like yours, basically a "TL;DR but here's what I think about this subject anyways" was what I feared most when posting this. I would highly recommend you don't just skim the article, as it's really not as simple as you're making it seem, and the article goes into why. And it also talks about why "accuracy" of the predictions is not the only issue, as the predictions are often self-fulfilling prophecies, and they're being used to justify even more heavy-handed enforcement in areas where there are already problems with discriminatory policing practices.
Ok, I read it, but the article still makes no fucking sense. Now I'm not terribly sympathetic towards "crime prediction software" and it pleases me to read that it seems to fall out of fashion (it...
Ok, I read it, but the article still makes no fucking sense.
Now I'm not terribly sympathetic towards "crime prediction software" and it pleases me to read that it seems to fall out of fashion (it looks like usage more than halved since 2018) and I get the argument that self-fulfilling prophecies can be a thing. But the "Steep Consequences" section has exactly one anecdote. A man was banned from a neighborhood with a zero-tolerance policy for crime (now that is questionable but a different topic). This was for shoplifting, over a decade earlier. Now he gets caught moving in said neighborhood with his girlfriend, gets arrested for trespassing and gun possession (totally not his, though!) and as a result, his girlfriend gets evicted. All kinds of things are fucked up about this situation but if this is literally the only anecdote they could find (used to counter statistics software, btw), that's just not a very coherent argument against police patrols.
The headline deliberately chooses to make this about "bias" and starts with numbers that suggest it being plain racist. That's the whole angle that hangs over the piece. And it doesn't seem to be the case. The software just uses time, location and type of crime for its predictions. That's probably why it doesn't seem to work very well. The result might illustrate prevalence of crime in poor or Black neighborhoods and there's something to learn from this other than sending more police there. But it doesn't necessarily make the software "biased". That's just a weird way of phrasing it.
What? No it doesn't. The headline very specifically say the software "perpetuates" the bias (already present in the data), which is even later referred to "bias by proxy" by Elgin's deputy chief....
The headline deliberately chooses to make this about "bias" and starts with numbers that suggest it being plain racist.
What? No it doesn't. The headline very specifically say the software "perpetuates" the bias (already present in the data), which is even later referred to "bias by proxy" by Elgin's deputy chief.
The software just uses time, location and type of crime for its predictions.
Yes, and the article spends an entire section (which you seem to have missed?) discussing why that overly simplistic model leads to problems (such as feedback loops), since the crime data itself is biased due to which groups tend to actually report crimes, and which don't. Which are the conclusions drawn directly from studies released by federal Bureau of Justice Statistics.
And all of which is perfectly summarized by this quote from that section:
“There’s no such thing as crime data,” said Phillip Goff, co-founder of the nonprofit Center for Policing Equity, which focuses on bias in policing. “There is only reported crime data. And the difference between the two is huge.”
That section on reporting was one that really took me by surprise - I'd have expected the opposite, that trust in police (and thus crime reporting) would be higher among the white and wealthy. I...
That section on reporting was one that really took me by surprise - I'd have expected the opposite, that trust in police (and thus crime reporting) would be higher among the white and wealthy. I clicked through to the BJS report but sadly I don't have the time right now to dive into a fairly dense 22 page paper, so if anyone happens to know the tl;dr on that I'd appreciate it for curiosity's sake.
This is entirely speculation, and based on anecdotal experience, but I suspect wealthier people tend not to report crimes as frequently because they value their time more, so don't want to have to...
This is entirely speculation, and based on anecdotal experience, but I suspect wealthier people tend not to report crimes as frequently because they value their time more, so don't want to have to waste that time dealing with the police. And they can also more easily afford to replace anything that gets damaged/stolen, and/or they have good enough insurance (and/or lawyers) to recover any losses.
E.g. When I woke up to find my car alarm having been torn out, passenger window smashed, stereo, subwoofer, tires and rims missing, and the entire vehicle resting on milk crates, I didn't report it to the police because when I phoned my insurance and asked if they needed a police report, they said no. The rep said they already knew the value of everything that was taken, so I just needed to send them pictures of all the damage, a copy of the tow truck receipt, and the estimate from my mechanic for fixing everything.
Yeah, that makes some intuitive sense if insurance isn't necessarily tied to a police report (in my personal experience has been, but I can quite believe it varies widely). To an extent I'd still...
Yeah, that makes some intuitive sense if insurance isn't necessarily tied to a police report (in my personal experience has been, but I can quite believe it varies widely). To an extent I'd still feel an almost... righteous? desire to record that I'd been wronged, even expecting nothing to come of it, but again I can see that varying a lot from person to person if the financial motive is less in the picture.
My righteous indignation at having been stolen from is far outweighed by my intense dislike of having to waste time waiting around for the police to arrive, spend even more time waiting for them...
My righteous indignation at having been stolen from is far outweighed by my intense dislike of having to waste time waiting around for the police to arrive, spend even more time waiting for them to "investigate", only to ultimately get nothing back in return but a piece of paper.
The first time I had my car broken into when I was living in Miami, it took over an hour for an officer to arrive, they spent another hour dusting the car for prints and doing various other BS... but not even bothering to check my apartment building's security cams because that was a "waste of time" (irony!), since they would "likely never be able to catch the thief anyways" or recover my property. So yeah, that's why I simply don't bother reporting most thefts anymore. Unless my insurance company demands it, it's not worth the headache.
But following that logic, aren't lower income and people of color even more distrustful of police? It's fascinating to me that white victims would actually report crime at a lower rate. Maybe the...
But following that logic, aren't lower income and people of color even more distrustful of police? It's fascinating to me that white victims would actually report crime at a lower rate. Maybe the wealthier victims are mostly facing property damage and they don't bother to report it, like you say. (Whereas more of the crime in poorer neighborhoods is violent and therefore gets reported?). It confuses me and I can't find where they get the percentages in the little bar graph. The special report they cite does support their argument but has different figures as far as I can tell.
People of lower income, and people of color are possibly more distrustful of police, but they're also probably far less able to afford to lose their property (esp if their livelihood depends on...
People of lower income, and people of color are possibly more distrustful of police, but they're also probably far less able to afford to lose their property (esp if their livelihood depends on that property), and are also probably less likely to have good insurance and/or lawyers that can aid them in those situations too. So when push comes to shove, despite more mistrust, they still call the police in desperation and the hopes of having their property eventually returned to them. But again that's just me spitballing, and also only relates to property crime. However, the varying natures of the different crimes more prevalent in each demographic, and the likelihood of police presence in different areas to begin with, is also probably a big factor in the disparity of crime reporting as well.
I can't find where they get the percentages in the little bar graph
A 12% difference is significant but it doesn't explain that 17,000%+ difference in crimes predicted in Plainfield from the intro. Nor does it control that number for total population per...
People earning $50,000 or more a year reported crimes to the police 12 percent less often than those earning $25,000 a year or less.
A 12% difference is significant but it doesn't explain that 17,000%+ difference in crimes predicted in Plainfield from the intro. Nor does it control that number for total population per neighborhood, which makes the numbers near impossible to compare.
Obviously, this is an article about large-scale politics masquerading as a critique of statistics software and that just seems disingenuous. It muddies discussion of either. The relevant question is why. Why do those numbers exist? What's the best way to improve them? I won't pretend like I know the answer but I doubt that arguing over accuracy in 12% ranges is relevant. What I do get is that this software probably is a distraction from real, impactful measures and sending more people with guns isn't a good way to solve the problem. If that is the point, I agree. But why then spend so much time arguing about "bias"?
The article attacks the messenger (crime prediction software which is basically just crime report statistics) when it should be attacking policy. The attacks on the former are flawed, the attacks on the latter are sensible. Mixing both just makes the conclusion less impactful.
It does when you realize that 12% is an averaged national statistic, but the software predictions are localized and based on that area's reported crime data, which likely varies wildly, especially...
A 12% difference is significant but it doesn't explain that 17,000%+ difference in crimes predicted in Plainfield from the intro
It does when you realize that 12% is an averaged national statistic, but the software predictions are localized and based on that area's reported crime data, which likely varies wildly, especially since the wealth gap in some areas is far larger than that $25k -> $50k comparison. And elsewhere in the article they also talk about other sources of bias in the data too, such as feedback loops.
Obviously, this is an article about large-scale politics masquerading as a critique of statistics software and that just seems disingenuous.
Oh, is that what it obviously is? Because, IMO, it sounds like that is simply the conclusion you came to before even having read the article, and no matter how much more evidence to the contrary gets provided to you, I suspect you're just going to keep shifting the goalposts and never change your opinion on this. So this will be my last reply. Have a nice day.
Why do you think this is intuitive? Well you see, if an officer is told to go somewhere because a crime is likely to be committed there, they will be primed to find a crime. For example, someone...
It would seem intuitive that poor areas have higher crime levels.
Why do you think this is intuitive?
were the predictions wrong?
Well you see, if an officer is told to go somewhere because a crime is likely to be committed there, they will be primed to find a crime. For example, someone smoking in a non-smoking area. It predicted a crime, and the cop found one.
certain groups get a greater number of false positives than others.
Given that white people are significantly less likely to be arrested in their lifetimes for any crime than someone who is colored, areas with a high density of white people would have more 'false positives'.
If the system is pretty close to predicting the truth, then I sounds like it’s working well.
It's pretty good at predicting the behavior of cops, that's for sure.
I have a few questions for you - how much do you know about policing or crime? How much do you know about artificial intelligence? What about ethics in either of those fields? Is there a reason your post fails to touch on any of these? The author distinctly points out several arrests which seem frivolous at best, all of which being targeted at minority individuals when cops were in an area which the software predicted a crime would happen. How do you square that away with an increasing public awareness as well as ever-reinforced scientific background consisting of countless studies on policing disparity and AI ethical issues?
An artificial intelligence system will only make a model based on training data. If there is a bias reflected in this training data, it will be reflected in any predictive model based on the training data. The AI system in question (PredPol) inherits the bias reflected in the training data.
The problem is people view these systems as an infallible crystal ball when all they do is advanced mathematical guesswork. In analytics the heavy lifting is done in the interpretation, and work should be done to determine biases in the data, it is literally the first step.
This is, IMO, a clash of hard and soft problems. "Hard" being the analytics/math stuff, "soft" being the human stuff, like what causes increased crime/reporting in some areas (not gonna start listing things because it's a mess). The soft problems need to be fixed first, and new data accumulated based on this new system, or we will use what amount to dumb pipes to reinforce biases and systems of oppression by using data that reflects the result of these bad systems.
At risk of sounding like a broken record: in predictive analytics you can only see the past and guess the future, which means a poorly chosen application of predictive models will only perpetuate the environment the data came from.
This is just a reflection of conversations that have been happening in AI ethics for a few years now. Nearly all publicly available data sets are deeply flawed. GPT-3 is incredibly racist towards muslims. Healthcare algorithms are racist towards blacks and other minorities. Image searching technology is racist and sexist. Many large bodies have tried to weigh in on how to correct and accommodate these biases, but we are many years away from dealing with these problems on a systematic scale and especially through the lens of currently-available large AI offerings. The problem, in short, is that capitalism has found a lot of reasons to employ AI rapidly, without fully understanding what the scientists have to say about it. A broken algorithm can still make you money and make you more efficient, while simultaneously amplifying existing biases that come from the data set.
As someone who works in data science, I've been watching this conversation evolve for some time now and I'm becoming increasingly more worried about the state of affairs. Luckily there are plenty of researchers refining the techniques of algorithms to allow 'fixing' of these AI algorithms by incorporating additional weighted steps and curated data sets, but these are not easy, nor cheap fixes. We need to be ever vigilant about the AI that we see, and perhaps the most important question we should be asking about them is the data on which they are trained. What biases may lie in the data that we see that are obvious, and what biases may be hidden within cultural values and internalizations we create from our exposure to the world and to large data sets which reflect these values?
It seems like there's an underlying assumption that more police presence is bad. I think we all know reasons why that assumption is made, and the article provides a few more.
But if you think about it, it's kind of odd. Why should providing more of a government service be bad? Isn't that weird? Shouldn't we fix that? If people in certain neighborhoods call the police more often, they should get better service.
If we thought more policing were good, the results of using these crime prediction tools would be interpreted differently. They would be biased in favor of certain minority neighborhoods that are more in need, so they get better service.
If sending the police is bad on average, no crime prediction tool is going to help. I guess the idea behind getting rid of the prediction tools is that we can't fix the police, so better to limit the damage by spreading police presence evenly across more neighborhoods, so everyone is has an equal chance of being punished? It seems kind of bleak.
I'm thinking that the people who came up with the crime prediction tools probably didn't make this bleak assumption. They probably thought more police presence is helpful, so the kind of bias described in this article isn't a problem. This is... optimistic. But most people don't assume they're making the world worse. Despite all the protests, I would guess that most people in law enforcement still don't think that doing their jobs makes things worse?
The article gives us a couple stories explaining mechanisms for how more police presence can be bad. They describe how a "crime-free housing ordinance" can get someone evicted. In another case, someone is stopped for smoking in a non-smoking area, and the police officer finds they have an unpaid fine.
These sound like things the police did according to laws that were passed on purpose, but had bad effects. Maybe that's more important to fix than the crime prediction tools?
Yes! But we can't. Politically, it is much more difficult to remove or significantly reform an existing system than it is to prevent a new one from being put in place.
This is absolutely true, and that's the position the majority of reformists are coming from these days. We have been trying to reform the police, in a broad sense, since the early 2010s; a large number of specific measures have failed to make it through local, state, and federal processes over that time. I'm sure that reform effort will continue, but when something like this shows up, we have to fight it.
We can't just say, "oh well, this would be really bad because it'll increase the punishment load on the underprivelaged, but we won't fight that because we want to spend all our resources on reforming the police as whole." We know that real police reform, getting to a point where sending the police is more likely to produce a positive outcome than a negative one, is going to take a very long time, if it's even possible. Therefore, we have no choice at all but to fight this at the same time; to prevent short term regressions as we work towards long-term progress.
EDIT: I just wanted to address this.
Yeah, man. It sucks. I hate it here. I think this is where a lot of the Left-doomers come from; examining systems like this and looking at the history of attempts to change them, since the 1870s, is... really, really, really, depressing. There have been victories - even recent victories! - but the general trend is that, with a few, mostly local blips over time, the police have gotten more money, more power, and more immunity. The police regularly defy the mayors of the cities they supposedly serve, and their unions - which I would argue don't really count as unions, since they don't actually have bosses to unionize against - consistently block any kind of accountability or reform.
Read the first half and skimmed the rest. But these findings seem.. obvious? It would seem intuitive that poor areas have higher crime levels.
Don’t know if this was buried in the body, but were the predictions wrong? I would be more interested if they could show that certain groups get a greater number of false positives than others. If the system is pretty close to predicting the truth, then I sounds like it’s working well.
TBH, a response exactly like yours, basically a "TL;DR but here's what I think about this subject anyways" was what I feared most when posting this. I would highly recommend you don't just skim the article, as it's really not as simple as you're making it seem, and the article goes into why. And it also talks about why "accuracy" of the predictions is not the only issue, as the predictions are often self-fulfilling prophecies, and they're being used to justify even more heavy-handed enforcement in areas where there are already problems with discriminatory policing practices.
Ok, I read it, but the article still makes no fucking sense.
Now I'm not terribly sympathetic towards "crime prediction software" and it pleases me to read that it seems to fall out of fashion (it looks like usage more than halved since 2018) and I get the argument that self-fulfilling prophecies can be a thing. But the "Steep Consequences" section has exactly one anecdote. A man was banned from a neighborhood with a zero-tolerance policy for crime (now that is questionable but a different topic). This was for shoplifting, over a decade earlier. Now he gets caught moving in said neighborhood with his girlfriend, gets arrested for trespassing and gun possession (totally not his, though!) and as a result, his girlfriend gets evicted. All kinds of things are fucked up about this situation but if this is literally the only anecdote they could find (used to counter statistics software, btw), that's just not a very coherent argument against police patrols.
The headline deliberately chooses to make this about "bias" and starts with numbers that suggest it being plain racist. That's the whole angle that hangs over the piece. And it doesn't seem to be the case. The software just uses time, location and type of crime for its predictions. That's probably why it doesn't seem to work very well. The result might illustrate prevalence of crime in poor or Black neighborhoods and there's something to learn from this other than sending more police there. But it doesn't necessarily make the software "biased". That's just a weird way of phrasing it.
What? No it doesn't. The headline very specifically say the software "perpetuates" the bias (already present in the data), which is even later referred to "bias by proxy" by Elgin's deputy chief.
Yes, and the article spends an entire section (which you seem to have missed?) discussing why that overly simplistic model leads to problems (such as feedback loops), since the crime data itself is biased due to which groups tend to actually report crimes, and which don't. Which are the conclusions drawn directly from studies released by federal Bureau of Justice Statistics.
See here: https://themarkup.org/prediction-bias/2021/12/02/crime-prediction-software-promised-to-be-free-of-biases-new-data-shows-it-perpetuates-them#who-reports-crime
And all of which is perfectly summarized by this quote from that section:
That section on reporting was one that really took me by surprise - I'd have expected the opposite, that trust in police (and thus crime reporting) would be higher among the white and wealthy. I clicked through to the BJS report but sadly I don't have the time right now to dive into a fairly dense 22 page paper, so if anyone happens to know the tl;dr on that I'd appreciate it for curiosity's sake.
This is entirely speculation, and based on anecdotal experience, but I suspect wealthier people tend not to report crimes as frequently because they value their time more, so don't want to have to waste that time dealing with the police. And they can also more easily afford to replace anything that gets damaged/stolen, and/or they have good enough insurance (and/or lawyers) to recover any losses.
E.g. When I woke up to find my car alarm having been torn out, passenger window smashed, stereo, subwoofer, tires and rims missing, and the entire vehicle resting on milk crates, I didn't report it to the police because when I phoned my insurance and asked if they needed a police report, they said no. The rep said they already knew the value of everything that was taken, so I just needed to send them pictures of all the damage, a copy of the tow truck receipt, and the estimate from my mechanic for fixing everything.
Yeah, that makes some intuitive sense if insurance isn't necessarily tied to a police report (in my personal experience has been, but I can quite believe it varies widely). To an extent I'd still feel an almost... righteous? desire to record that I'd been wronged, even expecting nothing to come of it, but again I can see that varying a lot from person to person if the financial motive is less in the picture.
My righteous indignation at having been stolen from is far outweighed by my intense dislike of having to waste time waiting around for the police to arrive, spend even more time waiting for them to "investigate", only to ultimately get nothing back in return but a piece of paper.
The first time I had my car broken into when I was living in Miami, it took over an hour for an officer to arrive, they spent another hour dusting the car for prints and doing various other BS... but not even bothering to check my apartment building's security cams because that was a "waste of time" (irony!), since they would "likely never be able to catch the thief anyways" or recover my property. So yeah, that's why I simply don't bother reporting most thefts anymore. Unless my insurance company demands it, it's not worth the headache.
But following that logic, aren't lower income and people of color even more distrustful of police? It's fascinating to me that white victims would actually report crime at a lower rate. Maybe the wealthier victims are mostly facing property damage and they don't bother to report it, like you say. (Whereas more of the crime in poorer neighborhoods is violent and therefore gets reported?). It confuses me and I can't find where they get the percentages in the little bar graph. The special report they cite does support their argument but has different figures as far as I can tell.
People of lower income, and people of color are possibly more distrustful of police, but they're also probably far less able to afford to lose their property (esp if their livelihood depends on that property), and are also probably less likely to have good insurance and/or lawyers that can aid them in those situations too. So when push comes to shove, despite more mistrust, they still call the police in desperation and the hopes of having their property eventually returned to them. But again that's just me spitballing, and also only relates to property crime. However, the varying natures of the different crimes more prevalent in each demographic, and the likelihood of police presence in different areas to begin with, is also probably a big factor in the disparity of crime reporting as well.
AFAICT it's a combination of the data reported in both BJS studies linked in the article:
Criminal Victimization, 2020 - has race/ethnicity data of victims, and a section on unreported violent crime
Victimizations Not Reported to the Police 2006-2010 - has crime victim reporting rates for race and income levels
A 12% difference is significant but it doesn't explain that 17,000%+ difference in crimes predicted in Plainfield from the intro. Nor does it control that number for total population per neighborhood, which makes the numbers near impossible to compare.
Obviously, this is an article about large-scale politics masquerading as a critique of statistics software and that just seems disingenuous. It muddies discussion of either. The relevant question is why. Why do those numbers exist? What's the best way to improve them? I won't pretend like I know the answer but I doubt that arguing over accuracy in 12% ranges is relevant. What I do get is that this software probably is a distraction from real, impactful measures and sending more people with guns isn't a good way to solve the problem. If that is the point, I agree. But why then spend so much time arguing about "bias"?
The article attacks the messenger (crime prediction software which is basically just crime report statistics) when it should be attacking policy. The attacks on the former are flawed, the attacks on the latter are sensible. Mixing both just makes the conclusion less impactful.
It does when you realize that 12% is an averaged national statistic, but the software predictions are localized and based on that area's reported crime data, which likely varies wildly, especially since the wealth gap in some areas is far larger than that $25k -> $50k comparison. And elsewhere in the article they also talk about other sources of bias in the data too, such as feedback loops.
Oh, is that what it obviously is? Because, IMO, it sounds like that is simply the conclusion you came to before even having read the article, and no matter how much more evidence to the contrary gets provided to you, I suspect you're just going to keep shifting the goalposts and never change your opinion on this. So this will be my last reply. Have a nice day.
Why do you think this is intuitive?
Well you see, if an officer is told to go somewhere because a crime is likely to be committed there, they will be primed to find a crime. For example, someone smoking in a non-smoking area. It predicted a crime, and the cop found one.
Given that white people are significantly less likely to be arrested in their lifetimes for any crime than someone who is colored, areas with a high density of white people would have more 'false positives'.
It's pretty good at predicting the behavior of cops, that's for sure.
I have a few questions for you - how much do you know about policing or crime? How much do you know about artificial intelligence? What about ethics in either of those fields? Is there a reason your post fails to touch on any of these? The author distinctly points out several arrests which seem frivolous at best, all of which being targeted at minority individuals when cops were in an area which the software predicted a crime would happen. How do you square that away with an increasing public awareness as well as ever-reinforced scientific background consisting of countless studies on policing disparity and AI ethical issues?