6 ms·

AI predicts crime a week in advance with 90 per cent accuracy

24 points by Tozen 4y ago

itisit 4y ago
> Details of the algorithm and the list were initially kept secret, but when the list was finally released, it turned out that 56 per cent of Black men in the city aged between 20 to 29 featured on it. That’s politically inconvenient, but it’s not bias. [0, 1] [0] https://home.chicagopolice.org/wp-content/uploads/2014/12/2011-Murder-Report.pdf https://home.chicagopolice.org/wp-content/uploads/2014/12/20... [1] https://www.chicagotribune.com/news/criminal-justice/ct-2021-homicides-final-20220103-lrpzuh5nsjhspmos3edrzxu2ei-story.html https://www.chicagotribune.com/news/criminal-justice/ct-2021...
- bergenty 4y ago
  It’s not politically inconvenient, it’s socially inconvenient. It’s a straightforward equation where “bad” elements in the black community need to be ruthlessly put away while not being hostile to every black person the cops come across. This seems to be something American cops are incapable of doing. The current black community isn’t helping either so it’s kind of a stalemate.
- nicbou 4y ago
  If you make it your policy to treat one group as suspicious, you'll catch a lot more people from that group. Isn't that what the police has been doing for decades? So what kind of AI would you train with the resulting dataset? The AI doesn't know about the crimes you let slide or didn't investigate. It will reproduce whatever bias you based your policing on.
pfisherman 4y ago
I am very skeptical. I suspect that the accuracy - metric undisclosed in the article - is dominated by “easy” cases. For example, I can tell you where protestation / solicitation will occur next week with 100 percent positive predictive value. Same with car break ins and drug dealing. You don’t need an algorithm for that. You can figure it out by just driving around a few neighborhoods for a couple hours. I suspect that the algorithm does not do well on the non-obvious stuff.
- erichocean 4y ago
  Is it important for the algorithm to "do well" on the non-obvious stuff? Why not handle low-hanging fruit first, and then try progressively (heh) harder situations?
  - zoover2020 4y ago
    It's the risk of bias slipping into the algorithms. See news coverage on Dutch cities using algorithms to predict financial crime/fraud. Ethnic profiling quickly becomes a thing.
  - dunefox 4y ago
    If the dataset is unbalanced a model that predicts the majority class only can have a very high accuracy, while being useless.
  - dagw 4y ago
    Why not handle low-hanging fruit first Because we don't need help on the obvious stuff. Cops already know where the low level drug dealers hang out and where the prostitutes work, they don't need a computer to tell them that. Also things like 'there will probably be a shooting in this violent neighborhood sometime within the next 7 days' is completely useless information, even if correct. There is no new actionable information that can improve on the current policing situation.
  - pfisherman 4y ago
    It is important because otherwise it has negative utility. There is a cost associated with building, deploying, operating and maintaining it. What value is this adding if it only tells me stuff I already know?
- daniel-cussen 4y ago
  Yeah like Plutonium Fountain on the edge of Lastarria in downtown Santiago. I have witnessed, defended against, or interceded in 6 crimes at this one intersection. One at the fountain itself, which is technically a beautiful statue of Neptune and Amphitrite. Like urchins bathe there, dudes washing their clothes. It's a dangerous intersection because there's five streets radiating from it. The traffic system is a little anti-pedestrian, or pedestrianophile, out to fuck pedestrians. So in Santiago there's three crosswalks per residential intersection, figuring if a pedestrian wanted to cross on the missing crosswalk all she'd have to do is cross all the other crosswalks. More efficient. So Plutonium Fountain has a really short light, just really timings, so you have to walk across on a red all the fucking time. And like motorized bicycles crossing reds, cars almost running you over...several traffic violations, just rough. It's also the point where the grid of the city gets warped by the Cerro Santa Lucía, beyond there it is all squares, this is where the Manhattan geometry meets the actual geography, the Cerro Huelén as its also called and the road to the river. Like a bottleneck. This is where the shit hits the fan. I can predict with 1 likelihood there'll be twenty crimes at that fountain in the next month. Not with machine learning! Just spinal. And of course I'm not going to waste the sight I have on figuring out how many more crimes there will be, but there will be twenty and there will be more. Like twenty and six, twenty and seven, there are twenty and there are also seven. Twenty and some other amount.
aeternum 4y ago
Wasn't it known in the 70s that something like 90% of crime occurs on a select streets? Not sure we need AI to make this prediction and achieve this accuracy.
1270018080 4y ago
Perpetuating bias, observing reality. Tomato tomato? Crime statistics aren't racist, people's explanations for "why" can be. If it's 90% accurate, then great? Do we need the algorithm to perform worse to make people feel better?
- nicbou 4y ago
  Crime statistics can absolutely be racist. The police as a force and as individuals choose which crimes to respond to, and with what intensity. Say for example that a white guy smoking a joint gets a slap on the wrist, and the black guy gets the book thrown at him. What would an AI derive from what it's given? Crime stats aren't collected in a scientific vacuum. They reflect the previous strategies the police used to fight crimes.
  - dtx1 4y ago
    > Crime stats aren't collected in a scientific vacuum. They reflect the previous strategies the police used to fight crimes. It may also just reflect reality. Let's say you have a group of people that commit a specific kind of crime significantly more often than another group, i.e. white male investment bankers are more likely to commit tax fraud than female black nurses. Any reasonable policy fighting this kind of crime would have to look biased against white men when it comes to enforcing tax fraud. I think that no one would reasonably call such enforcement policies bad or racist. But when black men are significantly more likely to be prosecuted for violent crime, suddenly it's a racist policy and must be the racist polices fault, because that's the only acceptable answer.
    - nicbou 4y ago
      Profiling people on what they do is very different from profiling people based on what they are. Conflating both is a perfect definition of a racist policy. If you keep targeting the same people, and you punish their crimes harder than any other people, you will get a data set that will bias any algorithm. The AI won't know that the police stops a disproportionate number of black drivers, or that they are more likely to prosecute a black criminal than a white one. It might not understand the biases that leads to harsher sentences. In other words, an algorithm trained on our biases will just try to reproduce our biased outcomes.
anonygler 4y ago
- bergenty 4y ago
  A long unbroken history of the cops being racist is what got us into a situation where black communities are less likely to report those crimes in the first place. Those activists may have a point.
higeorge13 4y ago
Minority report? :-)
- mdp2021 4y ago
  From TFA: > Chattopadhyay concedes that the data used by his model will also be biased, but says that efforts have been taken to reduce the effect of bias and the AI doesn’t identify suspects, only potential sites of crime. “It’s not Minority Report” he says
  - motoxpro 4y ago
    If it doesn't identify suspects then how do they know this information? "Details of the algorithm and the list were initially kept secret, but when the list was finally released, it turned out that 56 per cent of Black men in the city aged between 20 to 29 featured on it." So it is Minority Report... Literally.
    - mdp2021 4y ago
      No, that is a different AI: > Previous efforts to use AIs to predict crime have been controversial because they can perpetuate racial bias. In recent years, Chicago Police Department has trialled an algorithm that created a list of people deemed most at risk of being involved in a shooting, either as a victim or as a perpetrator. Details of the algorithm and the list were initially kept secret, but when the list was finally released, it turned out that 56 per cent of Black men in the city aged between 20 to 29 featured on it «Previous efforts», «In recent years, Chicago Police Department has trialled an algorithm that created a list of people». The work of Chattopadhyay is meant to reduce the bias and to avoid outputting names but just places. RTFA...
_l7dh 4y ago
So many concerns about this one... - what is the metric exactly? - what does 90% mean, is it bad or good? That depends on the baseline and the prior distributions. - on what dataset is the 90% computed? On past crimes? What about sampling bias in this dataset? What if the data focuses on some specific neighborhoods (oh wait...)? What about missing crimes in the data? In ML, data should fairly well represent a phenomenon and in this case, it does not represent crime occurrence, it only represents crimes detected and documented.
hulitu 4y ago
I've seen this one in Beavis & Butthead.
mdp2021 4y ago
The submitted divulgative article did not quite clarify what the AI is doing - it seems like weather forecasting on crime related events, but I expected more content about "how" and "how well"; for example, you do not need AI to determine "how much", and it takes some cleverness to instead predict "when", which is one of the interesting points you want explicit. So I found another divulgative article at Phys.org : https://phys.org/news/2022-06-algorithm-crime-week-advance-reveals.html https://phys.org/news/2022-06-algorithm-crime-week-advance-r... and a news piece / press release at UChicago.edu : https://biologicalsciences.uchicago.edu/news/algorithm-predicts-crime-police-bias https://biologicalsciences.uchicago.edu/news/algorithm-predi... and finally, the original research article, at: https://www.nature.com/articles/s41562-022-01372-0 https://www.nature.com/articles/s41562-022-01372-0 I am unfortunately lacking the time now to examine the material (hopefully I will later). For now, it seems clear that the effort is not really about crime prediction but to show that some methodology can produce unbalanced police planning: > We introduce a stochastic inference algorithm that forecasts crime by learning spatio-temporal dependencies from event reports, with a mean area under the receiver operating characteristic curve of ~90% in Chicago for crimes predicted per week within ~1,000 ft. Such predictions enable us to study perturbations of crime patterns that suggest that the response to increased crime is biased by neighbourhood socio-economic status, draining policy resources from socio-economically disadvantaged areas, as demonstrated in eight major US cities
wooque 4y ago
You can predict crime with 90% accuracy without AI
motohagiography 4y ago
A.I. should not be used for law enforcement because it just creates an infinite pretext for whatever they want to do that would otherwise require probable cause, reasonable suspicion, presenting a case to a prosecutor or a judge, and all the other inconvieniences your inalieable rights create for them. Law enforcement use cases for machine learning are nothing more than a laundering of accountability for individual decisions. The whole notion of algorithms being somehow autonomous is the most pernicious and magically anthropomorphic falsehood supposedly smart people believe, imo. I did a privacy impact assessment many years ago for a government agency who planned to outsource their benefits fraud detection to an "AI" as a service, and the service just happen to be run by a credit bureau company, effectively giving complete benefits and other personal information "for analysis" to a literal scumbag consumer credit reporting company. The project managers were trying to convince me and everyone else that somehow this data would not get re-used and integrated into the credit agencies other data sets and resold among the others - because "they're using AI, nobody there is going to see this data." I think similar law enforcement use cases for "AI" are just as magically stupid.

24 points by Tozen 4y ago

> Details of the algorithm and the list were initially kept secret, but when the list was finally released, it turned out that 56 per cent of Black men in the city aged between 20 to 29 featured on it. That’s politically inconvenient, but it’s not bias. [0, 1] [0] https://home.chicagopolice.org/wp-content/uploads/2014/12/2011-Murder-Report.pdf https://home.chicagopolice.org/wp-content/uploads/2014/12/20... [1] https://www.chicagotribune.com/news/criminal-justice/ct-2021-homicides-final-20220103-lrpzuh5nsjhspmos3edrzxu2ei-story.html https://www.chicagotribune.com/news/criminal-justice/ct-2021...

bergenty 4y ago
It’s not politically inconvenient, it’s socially inconvenient. It’s a straightforward equation where “bad” elements in the black community need to be ruthlessly put away while not being hostile to every black person the cops come across. This seems to be something American cops are incapable of doing. The current black community isn’t helping either so it’s kind of a stalemate.
nicbou 4y ago
If you make it your policy to treat one group as suspicious, you'll catch a lot more people from that group. Isn't that what the police has been doing for decades? So what kind of AI would you train with the resulting dataset? The AI doesn't know about the crimes you let slide or didn't investigate. It will reproduce whatever bias you based your policing on.

pfisherman 4y ago

I am very skeptical. I suspect that the accuracy - metric undisclosed in the article - is dominated by “easy” cases. For example, I can tell you where protestation / solicitation will occur next week with 100 percent positive predictive value. Same with car break ins and drug dealing. You don’t need an algorithm for that. You can figure it out by just driving around a few neighborhoods for a couple hours. I suspect that the algorithm does not do well on the non-obvious stuff.

erichocean 4y ago
Is it important for the algorithm to "do well" on the non-obvious stuff? Why not handle low-hanging fruit first, and then try progressively (heh) harder situations?
- zoover2020 4y ago
  It's the risk of bias slipping into the algorithms. See news coverage on Dutch cities using algorithms to predict financial crime/fraud. Ethnic profiling quickly becomes a thing.
- dunefox 4y ago
  If the dataset is unbalanced a model that predicts the majority class only can have a very high accuracy, while being useless.
- dagw 4y ago
  Why not handle low-hanging fruit first Because we don't need help on the obvious stuff. Cops already know where the low level drug dealers hang out and where the prostitutes work, they don't need a computer to tell them that. Also things like 'there will probably be a shooting in this violent neighborhood sometime within the next 7 days' is completely useless information, even if correct. There is no new actionable information that can improve on the current policing situation.
- pfisherman 4y ago
  It is important because otherwise it has negative utility. There is a cost associated with building, deploying, operating and maintaining it. What value is this adding if it only tells me stuff I already know?
daniel-cussen 4y ago
Yeah like Plutonium Fountain on the edge of Lastarria in downtown Santiago. I have witnessed, defended against, or interceded in 6 crimes at this one intersection. One at the fountain itself, which is technically a beautiful statue of Neptune and Amphitrite. Like urchins bathe there, dudes washing their clothes. It's a dangerous intersection because there's five streets radiating from it. The traffic system is a little anti-pedestrian, or pedestrianophile, out to fuck pedestrians. So in Santiago there's three crosswalks per residential intersection, figuring if a pedestrian wanted to cross on the missing crosswalk all she'd have to do is cross all the other crosswalks. More efficient. So Plutonium Fountain has a really short light, just really timings, so you have to walk across on a red all the fucking time. And like motorized bicycles crossing reds, cars almost running you over...several traffic violations, just rough. It's also the point where the grid of the city gets warped by the Cerro Santa Lucía, beyond there it is all squares, this is where the Manhattan geometry meets the actual geography, the Cerro Huelén as its also called and the road to the river. Like a bottleneck. This is where the shit hits the fan. I can predict with 1 likelihood there'll be twenty crimes at that fountain in the next month. Not with machine learning! Just spinal. And of course I'm not going to waste the sight I have on figuring out how many more crimes there will be, but there will be twenty and there will be more. Like twenty and six, twenty and seven, there are twenty and there are also seven. Twenty and some other amount.

aeternum 4y ago

Wasn't it known in the 70s that something like 90% of crime occurs on a select streets? Not sure we need AI to make this prediction and achieve this accuracy.

1270018080 4y ago

Perpetuating bias, observing reality. Tomato tomato? Crime statistics aren't racist, people's explanations for "why" can be. If it's 90% accurate, then great? Do we need the algorithm to perform worse to make people feel better?

nicbou 4y ago
Crime statistics can absolutely be racist. The police as a force and as individuals choose which crimes to respond to, and with what intensity. Say for example that a white guy smoking a joint gets a slap on the wrist, and the black guy gets the book thrown at him. What would an AI derive from what it's given? Crime stats aren't collected in a scientific vacuum. They reflect the previous strategies the police used to fight crimes.
- dtx1 4y ago
  > Crime stats aren't collected in a scientific vacuum. They reflect the previous strategies the police used to fight crimes. It may also just reflect reality. Let's say you have a group of people that commit a specific kind of crime significantly more often than another group, i.e. white male investment bankers are more likely to commit tax fraud than female black nurses. Any reasonable policy fighting this kind of crime would have to look biased against white men when it comes to enforcing tax fraud. I think that no one would reasonably call such enforcement policies bad or racist. But when black men are significantly more likely to be prosecuted for violent crime, suddenly it's a racist policy and must be the racist polices fault, because that's the only acceptable answer.
  - nicbou 4y ago
    Profiling people on what they do is very different from profiling people based on what they are. Conflating both is a perfect definition of a racist policy. If you keep targeting the same people, and you punish their crimes harder than any other people, you will get a data set that will bias any algorithm. The AI won't know that the police stops a disproportionate number of black drivers, or that they are more likely to prosecute a black criminal than a white one. It might not understand the biases that leads to harsher sentences. In other words, an algorithm trained on our biases will just try to reproduce our biased outcomes.

anonygler 4y ago

bergenty 4y ago
A long unbroken history of the cops being racist is what got us into a situation where black communities are less likely to report those crimes in the first place. Those activists may have a point.

higeorge13 4y ago

Minority report? :-)

mdp2021 4y ago
From TFA: > Chattopadhyay concedes that the data used by his model will also be biased, but says that efforts have been taken to reduce the effect of bias and the AI doesn’t identify suspects, only potential sites of crime. “It’s not Minority Report” he says
- motoxpro 4y ago
  If it doesn't identify suspects then how do they know this information? "Details of the algorithm and the list were initially kept secret, but when the list was finally released, it turned out that 56 per cent of Black men in the city aged between 20 to 29 featured on it." So it is Minority Report... Literally.
  - mdp2021 4y ago
    No, that is a different AI: > Previous efforts to use AIs to predict crime have been controversial because they can perpetuate racial bias. In recent years, Chicago Police Department has trialled an algorithm that created a list of people deemed most at risk of being involved in a shooting, either as a victim or as a perpetrator. Details of the algorithm and the list were initially kept secret, but when the list was finally released, it turned out that 56 per cent of Black men in the city aged between 20 to 29 featured on it «Previous efforts», «In recent years, Chicago Police Department has trialled an algorithm that created a list of people». The work of Chattopadhyay is meant to reduce the bias and to avoid outputting names but just places. RTFA...

_l7dh 4y ago

So many concerns about this one... - what is the metric exactly? - what does 90% mean, is it bad or good? That depends on the baseline and the prior distributions. - on what dataset is the 90% computed? On past crimes? What about sampling bias in this dataset? What if the data focuses on some specific neighborhoods (oh wait...)? What about missing crimes in the data? In ML, data should fairly well represent a phenomenon and in this case, it does not represent crime occurrence, it only represents crimes detected and documented.

hulitu 4y ago

I've seen this one in Beavis & Butthead.

mdp2021 4y ago

The submitted divulgative article did not quite clarify what the AI is doing - it seems like weather forecasting on crime related events, but I expected more content about "how" and "how well"; for example, you do not need AI to determine "how much", and it takes some cleverness to instead predict "when", which is one of the interesting points you want explicit. So I found another divulgative article at Phys.org : https://phys.org/news/2022-06-algorithm-crime-week-advance-reveals.html https://phys.org/news/2022-06-algorithm-crime-week-advance-r... and a news piece / press release at UChicago.edu : https://biologicalsciences.uchicago.edu/news/algorithm-predicts-crime-police-bias https://biologicalsciences.uchicago.edu/news/algorithm-predi... and finally, the original research article, at: https://www.nature.com/articles/s41562-022-01372-0 https://www.nature.com/articles/s41562-022-01372-0 I am unfortunately lacking the time now to examine the material (hopefully I will later). For now, it seems clear that the effort is not really about crime prediction but to show that some methodology can produce unbalanced police planning: > We introduce a stochastic inference algorithm that forecasts crime by learning spatio-temporal dependencies from event reports, with a mean area under the receiver operating characteristic curve of ~90% in Chicago for crimes predicted per week within ~1,000 ft. Such predictions enable us to study perturbations of crime patterns that suggest that the response to increased crime is biased by neighbourhood socio-economic status, draining policy resources from socio-economically disadvantaged areas, as demonstrated in eight major US cities

wooque 4y ago

You can predict crime with 90% accuracy without AI

motohagiography 4y ago

A.I. should not be used for law enforcement because it just creates an infinite pretext for whatever they want to do that would otherwise require probable cause, reasonable suspicion, presenting a case to a prosecutor or a judge, and all the other inconvieniences your inalieable rights create for them. Law enforcement use cases for machine learning are nothing more than a laundering of accountability for individual decisions. The whole notion of algorithms being somehow autonomous is the most pernicious and magically anthropomorphic falsehood supposedly smart people believe, imo. I did a privacy impact assessment many years ago for a government agency who planned to outsource their benefits fraud detection to an "AI" as a service, and the service just happen to be run by a credit bureau company, effectively giving complete benefits and other personal information "for analysis" to a literal scumbag consumer credit reporting company. The project managers were trying to convince me and everyone else that somehow this data would not get re-used and integrated into the credit agencies other data sets and resold among the others - because "they're using AI, nobody there is going to see this data." I think similar law enforcement use cases for "AI" are just as magically stupid.