19 ms·
Police across Canada are using predictive policing algorithms, report finds
- jbob2000 6y agoThis is no different than the human profiling that Police engage in. Except now we have logs and we can track the decisions that are being made. We can adjust the algorithms rather than re-train humans. We can replace an algorithm quickly if it doesn't give us the results we want, whereas removing bad cops takes years, if it's even possible. Secondly, Canadian police operate differently than American ones. They aren't perfect, but don't read this article with the lens that American media has shaped for you about their own police forces.
- colejohnson66 6y agoYou’re right: there is no difference; they both don’t work and are unethical. The idea that someone may commit a crime based on things they are doing is abhorrent. It runs against the idea of innocent until proven guilty. It’s the same idea as “violent video games cause violent people;” It’s not true. If you play an online FPS like COD, should we be watching you more closely because you may end up shooting up some place? EDIT: previously said “guilty until proven innocent”
- xyzzy_plugh 6y agoWhere in TFA do they mention persecution of anyone who hasn't yet committed a crime? I don't necessarily agree with mass-surveillance, but how does this differ significantly from "putting more patrol cars in high-crime neighbourhoods"?
- colejohnson66 6y agoI’m not referencing persecution, but surveillance. I’ll admit that it is a gray area. But dubious or outright false criteria are used all the time until a court orders the police to stop. As for “putting more patrol cars in high-crime neighborhoods,” that is somewhat different IMO because a high-crime neighborhood actually commits more crime (assuming accurate reporting). But, for example, “plays violent video games,” is a meaningless statistic with no correlation to reality.
- stinkytaco 6y agoPart of the argument of the "abolish the police" movement is that putting cars in high-crime neighborhoods becomes as self fulfilling prophecy. More police mean more crimes, even for slight infractions that someone in a lower crime neighborhood (read: white and middle class) would likely get away with. I think the issues here are much more nuanced than rhetoric can encompass, but the problem with preventative policing strategies is that they seem to find a way to justify themselves, making them ultimately self defeating and that perhaps preventative policing should be replaced with other preventative measures.
- themaninthedark 6y agoSomehow a neighborhood gets the label "High Crime". If the label is created due to the sum of reports from community and reports from officers, we could remove the officer generated reports to create a less biased label.
- ghostpepper 6y agoI think the policing problems of the US may not be applicable to Canadian cities such as Vancouver (referenced in the article as being tracked using https://geodash.vpd.ca https://geodash.vpd.ca) since the vast majority of crimes in Vancouver are property crimes, and thus likely citizen-reported. That being said it does feel like violent crime is increasing here, as the COVID pandemic has exacerbated the already-ongoing opioid public health emergency.
- pbhjpbhj 6y ago>More police mean more crimes, even for slight infractions // That's not a given, unless you send the police in to arrest people for 'slight infractions'. It's not a necessary outcome.
- unishark 6y agoI would assume the algorithm predicts the actual type of crime, or at least category (e.g. violent versus theft versus moving violation). Then it's still up to the department which crimes to target (which I really hope would be the crimes with victims).
- 6y ago
- stinkytaco 6y agoI would not normally be pedantic, but: > It runs against the idea of guilty until proven innocent. I think this is not what you mean?
- colejohnson66 6y agoYou’re right; I had a brain fart. Fixed. Thank you.
- bluntfang 6y ago>If you play an online FPS like COD, should we be watching you more closely because you may end up shooting up some place? This ship has kinda sailed, no? how many lists do you think you're on?
- colejohnson66 6y agoThe ship may have sailed, but that doesn’t mean we shouldn’t push back against further abuses of power. It’s easy to give more power to them, but much harder to remove it.
- bluntfang 6y agoWe've been trying to push back on cops murdering black people for like 6 months now. The ones with the monopoly on violence aren't really interested.
- jbob2000 6y agoYou aren't contributing honestly. The police systems work, but they have problems. And again, don't view Canadian policing with an American lens. Toronto, for example, has an independent division of police called the SIU, Special Investigations Unit. Any time an officer is involved in a violent altercation, the SIU is onsite and investigating. Not perfect, but better. They also have a mental health unit that responds to non-violent calls where people are in distress. Again, not perfect, but better. Toronto and Vancouver both operate safe injection sites. Toronto recently put up hundreds of homeless people into vacant apartments to get them off the street. All this to say, I don't see these programs in the US. You guys aren't even trying to improve your Police. Don't view canadian police with an American lens.
- refurb 6y agoAs someone who experience with both Canadian and US police, I think the differences are less than they appear. Especially when you take into account Canada has 1/10th the US population (so you’d expect 1/10th of the incidents). I’ve had several friends who have been tuned up by either the RCMP or provincial police. Sure, you’re less likely to get shot than in the US, but my anecdotal experience suggests there are plenty of bad apples in Canada. Hell, they are still trying to clean up the RCMP. You see some pretty egregious behavior in the small town detatchments.
- manbackharry 6y agoI'm not sure you want to look at the SIU as some paragon of justice holding police to account. The SIU is not a division of the police, being under the purview of the provincial government, but it sure is staffed by a whole lot of former police (nearly all white too) who seem to have no problem finding other police innocent of any wrongdoing. The most recent stats I read were that 97% of cases are cleared without charges and you can count on 1 hand the number of cops who actually went to jail over their 30 year existence. Their "mental health unit" didn't do much to save Regis Korchinski-Paquet, and the feckless SIU wasn't much better in investigating. They also don't deserve any credit for the current investigation of Ejaz Choudry's murder, another person with mental health problems being "assisted" by police while in distress. Absolutely police systems work, as long as you're not on the side that's being policed.
- luckylion 6y ago> The idea that someone may commit a crime based on things they are doing is abhorrent. It runs against the idea of innocent until proven guilty. That's nonsense, it has nothing to do with the presumption of innocence, that only comes into play at a trial. Law enforcement often looks at things that aren't crimes to predict crimes. An obvious example: people grooming minors online. It's totally legal to talk to a twelve-year-old on the internet. It's totally legal to meet twelve-year-olds you've talked to on the internet. But it's highly suspicious when you're 30, and if the police hears about it, they'll very much be interested in checking that out, even though you haven't committed any crimes yet. It's a different question whether these systems look at things that are strong predictors of criminal activity or not.
- chard_slicks_ 6y agoI agree with the unethical part but I'm not so sure about the statement that they don't work. Check out this video [0] which talks about profiling at Ben Gurion airport. [0] https://www.youtube.com/watch?v=1Y1kJpHBn50 https://www.youtube.com/watch?v=1Y1kJpHBn50
- pbhjpbhj 6y agoSo, car insurance should be the same price for everyone as no one is more likely to be an at-fault cause of an accident?
- kwhitefoot 6y agoDo they work? Are they cost effective?
- save_ferris 6y agoName one commercial ML algorithm that works with 100% accuracy all the time.
- luckylion 6y agoNobody needs 100% accuracy though, they just need better-then-current accuracy.
- Spooky23 6y agoThey do and they are -- alot of the tech/methodology was initially developed for identifying and proactively intervening high-risk individuals for counter-extremist purposes. Governments do things like buy search engine ads to counter misinformation, etc. This type of capability can be applied to things like opioid intervention. The problem is that the issues are being "attacked" from the perspective of criminality vs. public health or other scenarios. When you look at correlating factors with say, opioid abuse, things like transportation, access to healthcare, etc are prominent.
- ideashower 6y agoThis is dumb. Predictive policing algorithms predict where the police will be, not where crime takes place: as crime data is simply a measurement of policing activity. If there is bias in policing, it would therefore be amplified. Ban police from using this technology.
- mc32 6y agoAs departments get fewer funds and simultaneously try to keep a lid on crime, they will resort to more of these “more bang for the buck” strategies.
- ideashower 6y agoIndeed. Especially as companies like Palantir serve the tools on a silver platter.
- Ensorceled 6y agoFewer funds? Metro Toronto policing budget went up 8% last year.
- creddit 6y agoWhat if the model data is citizen reported crimes?
- ghostpepper 6y agoThis is my first thought as well. If police go to a place, there is more likely to be a recorded instance of a crime investigated in that place, and thus the algorithm will be more likely to send the police there in the future. The crime map for Vancouver is interesting - I'm glad to see that the tool reported in the article is at least exposed to the public. https://geodash.vpd.ca/ https://geodash.vpd.ca/
- at_a_remove 6y agoIn some sense, if you did not send police anywhere at all, you would have no crime to report. The solution to end crime, then, is to never send the police anywhere.
- jdc 6y agoSource article: https://citizenlab.ca/2020/09/to-surveil-and-predict-a-human-rights-analysis-of-algorithmic-policing-in-canada https://citizenlab.ca/2020/09/to-surveil-and-predict-a-human...
- motohagiography 6y agoThere are lots of good arguments against the use of these technologies, but disproportionate effect on marginalized communities may be ironically a less powerful one to people who believe in them, because the intent is to find indicators independent of the experiences of individual enforcers. Anyone with experience in marginalized communities knows it is a good bet to predict the self-sustaining effects of marginalization, which are violence, drugs, mental illness and exploitation by people there to take advantage of it. Doesn't take rocket science to predict where that's going to happen and to be there when it does. The real solutions are way upstream of a police encounter, where these predictive models would have a way more viable impact in elementary and high schools where people can be nudged and diverted from police encounters in a more subtle and constructive way. But even then the tech still reinforces and systematize the aggregate bias (or experience) of the data analysts, with the real danger it just becomes a justification, the same way statistics are always used, which is the way a drunk uses a lamp post - for support and not illumination. Whether it's teachers using it to mark and isolate kids from positive peers, or police using it to divert them from a bigger incident, it's the same underlying belief in social engineering that problematizes "other" people and tries to "fix," them. We should begin by recognizing that these ML systems are weapons, not "solutions," and regulate the use of them in the context of weapons against an enemy other, then ensure someone holds personal individual accountability for that decision, including the social and political aspects of it. The second order effect is that more-focused enforcement methods will necessarily mean more confrontations with people prone to violence, which creates a selection bias toward violent encounters with police, which makes these systems all the more dystopian.
- 3pt14159 6y agoEh, I'm not sure I agree. I was contracted by The Government of Canada to help them with a recommendation engine that was built to help people. The problem? Some criminals had reverse engineered the algorithm and were using it for really, really horrible ends. What would you have had me do? Not reverse engineer the criminal's reverse engineering? Did I help GC make a "predictive policing" app because they could now alert the RCMP to these bastards? It's a balance. We need privacy, but we also need security. My take on it is that we should start with good faith and only surveil where it makes sense (embassies, airports, etc) unless there is some sort of signal (in the ML sense) of bad faith. From that point on it should go through the crown's lawyers (like it does in Canada), but ultimately we need to stop murderers, human traffickers, and other people that make the world worse. I agree that some ML systems are weapons, but I don't think they all are. Some are benign. Some start off off as benign then get turned into dual use, like the one I worked on. But not all weapons are necessarily bad. We need a government that can actually protect us. There is real evil in the world.
- icegreentea2 6y agoHonestly, better just to skip straight to the linked report that Vice is just lifting from - (https://citizenlab.ca/2020/09/to-surveil-and-predict-a-human-rights-analysis-of-algorithmic-policing-in-canada/ https://citizenlab.ca/2020/09/to-surveil-and-predict-a-human...) The executive summary does a far better job at articulating the situation, its problems, its negative implications, and proposed mitigations than most of us will be able to manage in an HN comment. Note that the report suggestions are not to ban predictive approaches. I think the report recommendations are reasonable under the assumption of competence and good faith among all parties. If you don't think that the limitations the report suggests are sufficient, I think that's a reasonable sign that you're not assuming good faith and competence - which is fair, but does totally shift the argument. If you're just assuming that the police will abuse this regardless of safe guards (that's fair), you can just make that argument up front, and cut through all the shit, and then you can get down to arguing over real conflict points.
- ramshorns 6y agoThe police have shown us that they don't act in good faith. But maybe if we decide to keep them around, these recommendations, along with better civilian oversight to ensure they're actually implemented, will mitigate the dangers of predictive policing.
- mbostleman 6y ago>>The police have shown us that they don't act in good faith.>> There are ~650k police officers in the US employed by ~18k agencies. Can you provide more detail on how it is that that group has shown us a lack of good faith?
- adamsea 6y agoBy not publicly and consistently condemning and/or holding their fellow officers accountable when they should be. The blue wall of silence. To be fair after the murder of George Floyd you did see some officers and police chiefs doing this in the initial round of protests. Generally speaking though police departments have proven incredibly difficult to reform - there’s a lot of entrenched behavior, such that even if one officer isn’t engaging in anything illegal themself, the police culture brings huge pressure to bear against anyone who might condemn another officer for breaking the law.
- dghughes 6y agoI recall a former RCMP officer who created something similar only it was for historical data not used to try to predict anything. This "Person of Interest" (sci-fi TV show) level of prediction seems impossible. Although giant powerful computers are used by stock analysts to predict trends. Stocks aren't people but there is a direct link between human emotion and stock buying and selling.
- Ensorceled 6y agoThere are two entirely different matters being conflated here: "Where will crime happen" prediction, which is based on crime reporting data for the most part and, as a sibling by comment by motohagiography articulates very well, the solution is upstream. This tool is useful, but problematic. Especially with "more encounters with police lead to more charges of all types" issues. "Who will commit crime" predictions are just plain problematic; they discard the entire premise of innocent until proven guilty that is supposed to underlie Canadian enforcement and jurisprudence as well as tend to be self fulfilling prophecies.
- oh_sigh 6y agoIn what manner is it problematic? No one is being punished by the system predicting that they will commit a crime.
- core-questions 6y agoIt's problematic because of the demographics of the areas and of the people who are predicted to commit crimes, of course. If the system doesn't spit out a perfectly egalitarian, proportional representation of the population down to an infinite level of granularity, it must be wrong; just like all the other machine learning systems that have been revealed to have incredible bias by similarly pointing out things that agree with observable, measured reality.
- mLuby 6y agoSo you would be okay with police driving behind you for years, waiting for you to make a mistake? Uneven application of even a just law is still unjust, and also undermines trust in the justice system.
- oh_sigh 6y agoAgain... is tracking and following potential future-criminals actually what this tech is used for?
- 6y ago
- michaelcampbell 6y ago"Weapons of Math Destruction" by Cathy O'Neil has a section devoted to this sort of thing. Good read, IMO.
- mabbo 6y agoIsn't this just computerization of what already existed? "Hey, I need to get my arrest numbers up so let's head on down to <area of town> where we know we'll pick up at least a couple people for <mostly harmless activity>". Now the police are just optimizing the algorithm to maximize their goal. The problem isn't the means, it's the goal itself. Where the police can predict they're going to be arresting a lot of people, maybe instead we should be focusing on how we can reduce the problems in those areas that lead to crimes and arrests. Send some social workers, some youth outreach, open a food bank, get health workers in there. "Tough on crime" policies gets expensive when you have to house and jail all those "criminals". It's a lot cheaper to just prevent future crime by helping people.
- notatoad 6y agothe report discusses multiple different types of predictive analysis, including the type you refer to (location and time based) but also personal: predicting which people might commit crimes, based on social media surveillance. that's a long ways from what police have always done.
- mabbo 6y ago> predicting which people might commit crimes, based on social media surveillance Well that's terrifying and more reason than ever to get off social media. But even still- if some system can say "This person has a strong likelihood of committing a crime", maybe we should get right down to the personal level, and send someone who can help prevent the crime.
- freeone3000 6y agoWhat makes you think this system is in any way accurate for individuals? The system can say it, and it'll even retroactively be true - if the police start watching someone for a crime, they'll find one. It's arbitrary policing outsourced to a third-party tool under the cover of "algorithmic correctness".
- 6y ago
- gentleman11 6y agoBeing profiled so deeply by government on an individual level using automated systems is the holy grail of the police state and is the most advanced and worrisome form of mass surveillance. It is possibly fine today, but over the course of 50-100 years there will surely be severe and scary problems with this
- core-questions 6y agoPerhaps you should stop simply giving all of your data away, then? Nobody's making you post all your thoughts and pictures and location and plans.
- VLM 6y agoThe concept of criminality as a protected career category is a political litmus test of no other content. May as well just ask if people are "D" or "R".
- mLuby 6y agoLet's algorithmically predict which police officers are most likely to commit misconduct, and then follow them to catch them in the act if and when they do. Seems only fair…
- ciarannolan 6y agoUnfortunately non-white police are more likely to use deadly force [1], so you'd get the same social justice outrage. Good idea in principle though. [1] https://www.pnas.org/content/pnas/116/32/15877.full.pdf https://www.pnas.org/content/pnas/116/32/15877.full.pdf
- mLuby 6y agoThe second page of that link shows the authors retracted the paper, and they also issued a correction. > we want to correct a sentence in our significance statement that has been quoted by others stating ‘White officers are not more likely to shoot minority civilians than non-White officers.’ This sentence refers to estimating Pr(shotjrace, X). As we estimated Pr(racejshot, X), this sentence should read: ‘As the proportion of White officers in a fatal officer-involved shooting increased, a person fatally shot was not more likely to be of a racial minority.’
- ciarannolan 6y agoThey retracted the paper because of political pressure, not because of any factual errors (this is pretty clear in both the timing of the retraction and the retraction statement). > While our data and statistical approach were appropriate for investigating whether officer characteristics are related to the race of civilians fatally shot by police, they are inadequate to address racial disparities in the probability of being shot. That's not what my comment addressed. It addressed the probability of non-white officers shooting to kill being higher than that of white officers. I think that conclusion is still supported by the paper (as in the last line of your quote).
- AtHeartEngineer 6y agoPalantir is just getting more and more scary everytime I read it hear anything about them
- LatteLazy 6y agoI don't know whether PP algos work or whether they're fair or what the trade off between fairness and effectiveness should be. But I am amazed the number of commentors here who think the devs behind the algos are not smart enough to normalise data for police numbers in an area!?
- lawnchair_larry 6y agoThis is HN, where everyone knows better than everyone else, especially in areas where they have no experience whatsoever.
- wdevanny 6y agoIt's a difficult problem area. I would bet naively normalizing the data would introduce many new concerns. Does the amount of crimes reported scale linearly with police presence? If not, your data is now going to be skewed. How do you count police numbers in an area? Man hours, individual officers, or something else? Keep in mind that not all officers are going to be responding to the same type of crime and so police hours are not interchangeable. IMO machine learning algorithms are brittle and prone to breaking in unexpected ways. In the past, organizations who should have known better and should have been smart enough have produced bad machine learning systems. Policing is a highly sensitive area. It's okay to be very concerned.
- just_myles 6y agoOf course understanding that this is Canada: I don't recommend this kind of approach. It might solve the immediate problem but, does not address the cause. Having police visibility during "peak" times will resolve the issue of crime in a specific area and time but, will not solve the crime from not happening somewhere else. Also, I don't know who is working on these reports and whether or not they apply their own biases. The US already has an issue with over policing. This would definitely exacerbate it.
- jariel 6y agoSocial scientists have wanted to predict when victimization, abuse, suicide is going to happen for a long time. This is the other side of the coin. I suggest that as long as 'individuals are not targeted for culpability' - ie they are looking in 'general terms' and especially for 'victims' then maybe this might work? 'Jane Doe is at very high risk of abuse from a family member' -> maybe we should just check in. '123 ABC St. is at very high likelihood of break-in - maybe just cruise on by to 'show a presence'. All of this said, I often wonder if these 'algorithms' just reproduce with cops on the ground, managing relationships already know. 'Yes, thank you computer, Jane Doe who's been beaten by her husband 3 times is 'at risk''. 'Yes, thank you computer, 6th and Finch, aka 'crack corner' is at 'high risk' of crime'.
- dleslie 6y agoPart of the problem is that police are being tasked with the impossible: to increase closure rates on crimes for which they have little evidence. This is part of us in Canada both underfunding social services _and_ offloading the defence of person and property on to the state. If we properly funded mental health, low income housing and food banks and also allowed citizens to defend their property and their family... Well I suspect that police would have less to do.