TE
科技回声
首页24小时热榜最新最佳问答展示工作
GitHubTwitter
首页

科技回声

基于 Next.js 构建的科技新闻平台,提供全球科技新闻和讨论内容。

GitHubTwitter

首页

首页最新最佳问答展示工作

资源链接

HackerNews API原版 HackerNewsNext.js

© 2025 科技回声. 版权所有。

Predictive policing software terrible at predicting crimes

301 点作者 AndrewDucker超过 1 年前

25 条评论

tech_ken超过 1 年前
I saw a talk by researcher Kristian Lum a few years back that I think made this case far more effectively. Her point was somewhat limited to drug crimes, but she pointed out that if you look at medical data (where people tend to be fairly honest about their drug usage) pretty much everyone in the metro area under study (SF) used drugs, or at least drug use was equally prevalent in pretty much every geographical area and among all demographics. Therefore, when the cops went to a location to make drug arrests they typically succeeded, because it&#x27;s not hard to find drug crimes in the Bay Area.<p>The problem was that they then used that arrest data to make decisions about where to perform future searches and arrests. Because they found drugs where they had looked previously, they looked there again and found more drugs. This creates a bad feedback loop where they were basically busting the same neighborhoods and demographics over and over again, despite the fact that drug crime was prevalent everywhere. In effect it was an insufficiently explorative learning strategy, just hitting the same lever over and over. Dr. Lum&#x27;s point was that predictive policing software merely hides this dynamic under a layer of black-box ML crap. Because the training data is itself the result of this type of bad policing, the resulting model can only further engrain these practices, it can&#x27;t offer truly novel solutions.<p>Crime and criminology is complicated, but at the end of the day not that complicated. On the whole people commit crime because they are desperate (for money, for drugs, etc.), occasionally because they have an anti-social personality disorder. Applying all these abstract epidemic&#x2F;broken-windows type models which pretend like the root causes of crime are unknowable allows police to appear like they&#x27;re operating efficiently, while at the same time just responding to the symptoms rather than facing the sickness itself. Until we actually look at why crime occurs (mainly because poor people need money badly, secondly because people in sufficiently dire poverty stop caring about the social norms of the middle class) we won&#x27;t be able to make a meaningful difference.
评论 #37756697 未加载
评论 #37754459 未加载
评论 #37754266 未加载
评论 #37757318 未加载
评论 #37754382 未加载
评论 #37756779 未加载
评论 #37754170 未加载
评论 #37757746 未加载
评论 #37756614 未加载
评论 #37754682 未加载
评论 #37756025 未加载
评论 #37758970 未加载
评论 #37754327 未加载
评论 #37757473 未加载
评论 #37757266 未加载
评论 #37762403 未加载
评论 #37761576 未加载
评论 #37757328 未加载
评论 #37762302 未加载
评论 #37760846 未加载
评论 #37758563 未加载
评论 #37764430 未加载
评论 #37759401 未加载
评论 #37756622 未加载
评论 #37761907 未加载
评论 #37758135 未加载
评论 #37758339 未加载
评论 #37756713 未加载
评论 #37761604 未加载
评论 #37757143 未加载
评论 #37757772 未加载
评论 #37758687 未加载
评论 #37757211 未加载
评论 #37758726 未加载
评论 #37757643 未加载
评论 #37760246 未加载
评论 #37758601 未加载
评论 #37754235 未加载
评论 #37754165 未加载
dionidium超过 1 年前
They say that:<p>&gt; <i>In 2021, The Markup published an investigation in partnership with Gizmodo showing that Geolitica’s software tended to disproportionately target low-income, Black, and Latino neighborhoods in 38 cities across the country.</i><p>But you should understand what is meant by &quot;disproportionately&quot; in this context. It <i>does not</i> appear to mean &quot;disproportionate to the amount of crime in those areas.&quot; It seems, as with most accusations of disparate impact, to mean just quite literally that more crimes are predicted in those neighborhoods <i>without reference to disparities in crime rates</i>.<p>They then imply (without direct reference to the enormous offending disparities) that this is explained by race and class differences in crime reporting: [0]<p>&gt; <i>The agency has found repeatedly that White crime victims are less likely to report violent crime to police than Black or Latino victims.<p>&gt; In a special report looking at five years of data, BJS found an income pattern as well. 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.<p>&gt; This disparity in crime reporting would naturally be reflected in predictions.</i><p>It&#x27;s <i>possible</i> this is having some effect, but, again, because there is no reference to the (often very large) baseline differences in crime <i>rates</i>, we can&#x27;t see what&#x27;s true, which is that this probably accounts for only a small amount of that difference.<p>[0] <a href="https:&#x2F;&#x2F;themarkup.org&#x2F;prediction-bias&#x2F;2021&#x2F;12&#x2F;02&#x2F;crime-prediction-software-promised-to-be-free-of-biases-new-data-shows-it-perpetuates-them" rel="nofollow noreferrer">https:&#x2F;&#x2F;themarkup.org&#x2F;prediction-bias&#x2F;2021&#x2F;12&#x2F;02&#x2F;crime-predi...</a>
评论 #37765699 未加载
shadowtree超过 1 年前
This is such an interesting topic, where the conclusions are so politically charged that no ones wants to risk their careers.<p>See this article in the NYT: <a href="https:&#x2F;&#x2F;www.nytimes.com&#x2F;2023&#x2F;04&#x2F;15&#x2F;nyregion&#x2F;shoplifting-arrests-nyc.html" rel="nofollow noreferrer">https:&#x2F;&#x2F;www.nytimes.com&#x2F;2023&#x2F;04&#x2F;15&#x2F;nyregion&#x2F;shoplifting-arre...</a><p>&quot;Nearly a third of all shoplifting arrests in New York City last year involved just 327 people, the police said. Collectively, they were arrested and rearrested more than 6,000 times, Police Commissioner Keechant Sewell said.&quot;<p>Can you predict crimes? Of course you can, the math is clear. The solution to <i>reduce</i> crime is also logical and clear, but politically unfeasible: extended incarceration.<p>All the handwringing about various side topics, like race, gender, class are just distractions. See El Salvador&#x27;s murder rate drop this year too. WSJ, with handwringing: <a href="https:&#x2F;&#x2F;www.wsj.com&#x2F;articles&#x2F;the-country-with-the-highest-murder-rate-now-has-the-highest-incarceration-rate-b5401da7" rel="nofollow noreferrer">https:&#x2F;&#x2F;www.wsj.com&#x2F;articles&#x2F;the-country-with-the-highest-mu...</a><p>Criminals commit crimes, a lot. Once you have identified a criminal, you can safely predict more crimes. No AI needed, a simple spreadsheet will suffice.<p>And here, another article for a balanced world view - Ireland: <a href="https:&#x2F;&#x2F;www.sundayworld.com&#x2F;crime&#x2F;irish-crime&#x2F;decrease-in-leinster-burglaries-attributed-by-gardai-to-deaths-of-three-criminals-in-n7-crash&#x2F;40858796.html#:~:text=A%20significant%20decrease%20in%20burglaries,a%20number%20of%20arrest%20operations" rel="nofollow noreferrer">https:&#x2F;&#x2F;www.sundayworld.com&#x2F;crime&#x2F;irish-crime&#x2F;decrease-in-le...</a>.<p>&quot;A significant decrease in burglaries in most Leinster counties is being attributed by senior gardaí to the deaths of three prolific criminals as well as a number of arrest operations.&quot;
评论 #37757814 未加载
评论 #37757320 未加载
评论 #37755703 未加载
评论 #37763793 未加载
评论 #37775478 未加载
quadrifoliate超过 1 年前
I am a little confused by this. Firstly, Gizmodo is reporting on somebody else&#x27;s investigation:<p>&gt; A new joint investigation by The Markup and Wired...<p>And when I go to the page about actual investigation by The Markup [1]<p>&gt; Our investigation <i>stopped short of analyzing precisely how effective Geolitica’s software was at predicting crimes</i> because only 2 out of 38 police departments provided data on when officers patrolled the predicted areas. Geolitica claims that sending officers to a prediction location would dissuade crimes through police presence alone. It would be impossible to accurately determine how effective the program is without knowing which predictions officers responded to and which ones they did not respond to.<p>Also, later in the article<p>&gt; Plainfield officials said they never used the system to direct patrols.<p>Given all this, it&#x27;s somewhat simplistic to say it&#x27;s &quot;pretty terrible at predicting crimes&quot;, even though that makes for a good clickbait headline. It seems that the software was intended to identify high-crime areas that to target for patrolling, which doesn&#x27;t seem like a <i>huge</i> problem to me -- but it seems like the software was never actually used as intended in the first place.<p>----------------------------------------<p>[1] <a href="https:&#x2F;&#x2F;themarkup.org&#x2F;prediction-bias&#x2F;2023&#x2F;10&#x2F;02&#x2F;predictive-policing-software-terrible-at-predicting-crimes" rel="nofollow noreferrer">https:&#x2F;&#x2F;themarkup.org&#x2F;prediction-bias&#x2F;2023&#x2F;10&#x2F;02&#x2F;predictive-...</a>
评论 #37755033 未加载
评论 #37754425 未加载
评论 #37753918 未加载
评论 #37753436 未加载
评论 #37753930 未加载
评论 #37756211 未加载
评论 #37753851 未加载
评论 #37754024 未加载
评论 #37753640 未加载
bunderbunder超过 1 年前
If we really believe in the basic principles of individual liberty that are hypothetically core American values (or even just the idea of &quot;first, do no harm&quot;), then it would imply that these tools should be optimized for 100% precision, even at the cost of terrible recall.<p>But if you&#x27;re trying to market your stuff to law enforcement agencies, I&#x27;m guessing you&#x27;re instead incentivized to optimize for recall, even at the cost of terrible precision. Because, probably with the best of intentions, that&#x27;s what they think they should be doing. But we&#x27;ve got basically the entire history of forensic &quot;science&quot; to demonstrate just how poorly the police tend to understand some of these basic statistical principles.
fnordpiglet超过 1 年前
I would be surprised if there are more predictive variables than “places where crime has happened repeatedly over time” that are significant enough or uncorrelated and noncausal to ever be more useful than patrol planners do today using standard statistics.
评论 #37754952 未加载
firebat45超过 1 年前
&gt; the software accurately predicted where crimes would occur with a “less than half a percent” success rate. &gt; [previous investigation found] that cops used it to disproportionately targeted low-income communities of color.<p>Ironically, this is exactly what I predicted.
评论 #37758076 未加载
评论 #37830116 未加载
Nevermark超过 1 年前
How many security measures would be unnecessary if the proper authorities would just show up just in time?<p>A big problem is we will soon have large models planning Turing unpredictable, devilishly elaborate, infallible heists.<p>(Cue short balding man staring into a safe full of Hershey’s, in lieu of missing gold bars. “Inconceivable!”.)<p>It’s an arms race.<p>Completely serious: Large scale multi-target generative social engineering (with no trace of North Korean accents), surreptitious access problem solving. Somehow this is going to be a real thing.<p>We are going to need better and more layers of security.
评论 #37754030 未加载
thexumaker超过 1 年前
I want more details. Say they predicted X spot would have more crime and so the PD sent more patrols there. Wouldn’t that affect the amount of crime?
评论 #37753785 未加载
评论 #37754612 未加载
评论 #37753681 未加载
chimeracoder超过 1 年前
Predictive policing isn&#x27;t about predicting crimes. Like other trends before it, like &quot;police psychics&quot;[0], it&#x27;s about manufacturing probable cause.<p>It&#x27;s a tool for police departments looking to meet their quotas and justify their ever-increasing budgets. Public safety is nowhere on the list of priorities.<p>[0] Yes, these are a thing, and they&#x27;re actually more horrifying than you are probably imagining.
评论 #37757069 未加载
daoboy超过 1 年前
These programs are an enormous waste of money. I took a geographic profiling course during undergrad expecting an interesting application of GIS and statistics.<p>It turned out to be a lot of handwaving to come to the obvious conclusion that most crimes are more likely to occur in some areas than others, along with the offender&#x27;s residence.<p>It was taught by a former FBI agent, and his academic acumen was very...disillusioning.
评论 #37760373 未加载
hsbauauvhabzb超过 1 年前
Is this any surprise? Tools like this are just a political tool to allow LE to say ‘we’re not racist, the computer made us do it’<p>That is unless they don’t agree with what the computer tells them to do, at which point it becomes ‘computers can’t be right ALL the time’
DoctorOetker超过 1 年前
I always assumed predictive policing was about whitewashing parallel construction?
评论 #37754034 未加载
评论 #37754316 未加载
ltbarcly3超过 1 年前
I don&#x27;t know the details of this system, and it may well be complete trash, so don&#x27;t take this as an apology for it.<p>It&#x27;s challenging to measure the utility of a system for predicting very low likelihood events, especially when the thing being measured is very heavily influenced by the method of measurement. For example, in Minority Report, the accuracy of the crime predictions is assumed to be perfect, and the result is that they are always incorrect. No crimes are committed because the intervention is always successful (at least up to the point where the story starts). So you might consider &#x27;never have to make an arrest&#x27; to be a perfect record for such a system, as in minority report. You might equally logically take &#x27;always makes an arrest&#x27; as proof a system like this is working.<p>You also have to worry about crime moving around to avoid the police. Maybe the bank robbery was going to happen at BoA, but they drove by and saw a bunch of police and robbed the Wells Fargo 20 minutes away instead.<p>As silly as it probably sounds, probably the best heuristic to get a good idea of whether a system like this works is the feedback police give after spending time following it&#x27;s leads. If they say this thing is predicting where they need to be better than whatever they did before, and there isn&#x27;t an obvious correlation to the predictions (as in, it doesn&#x27;t always send them to the black neighborhood) then it probably warrants more careful analysis. In this case the police thought the system was crap.
UIUC_06超过 1 年前
NYC is a little bigger sample than Plainfield, NJ:<p><a href="https:&#x2F;&#x2F;ccnmtl.columbia.edu&#x2F;projects&#x2F;caseconsortium&#x2F;casestudies&#x2F;127&#x2F;casestudy&#x2F;www&#x2F;layout&#x2F;case_id_127_id_885.html" rel="nofollow noreferrer">https:&#x2F;&#x2F;ccnmtl.columbia.edu&#x2F;projects&#x2F;caseconsortium&#x2F;casestud...</a>
mannyv超过 1 年前
Like everything, 20% of the population are responsible for 80% of the crime.<p>That&#x27;s true in the US, that&#x27;s true in Europe, Asia, etc. That&#x27;s even true in healthcare, where 20% of the people make up 80% of the spend.<p>There are people who just don&#x27;t obey the law and&#x2F;or excel at making bad decisions.<p>They don&#x27;t need to &quot;predict&quot; that they&#x27;ll commit a crime, they just need to predict &quot;when.&quot;
mainpassathome超过 1 年前
if net_worth &gt;= broke_lmao: return criminal
unhammer超过 1 年前
Predicting Financial Crime <a href="https:&#x2F;&#x2F;whitecollar.thenewinquiry.com&#x2F;static&#x2F;whitepaper.pdf" rel="nofollow noreferrer">https:&#x2F;&#x2F;whitecollar.thenewinquiry.com&#x2F;static&#x2F;whitepaper.pdf</a><p>&gt; We are confident that our model matches or exceeds industry standards for predictive policing tools.<p>;)
JoeAltmaier超过 1 年前
But if you could be 100% accurate at predicting crime, then you&#x27;d go stop it. Then the algorithm would be shit at predicting crime because there wasn&#x27;t any anymore?<p>A degenerate case of &#x27;predicting crime&#x27; could be, predict zero crime. Because you should have stopped all the crime I was going to predict!<p>&lt;&#x2F;humor&gt;
cratermoon超过 1 年前
It&#x27;s well known that poor and minority neighborhoods are overpoliced and profiled. All the software does is take data from where police have made arrests <i>for street crime</i> in the past and project that to the future, so of course it&#x27;s just going to send police to where cops think crime happens – in the poor and minority areas.<p>Look what happens when you model white collar crime instead: <a href="https:&#x2F;&#x2F;whitecollarcrime.zone&#x2F;" rel="nofollow noreferrer">https:&#x2F;&#x2F;whitecollarcrime.zone&#x2F;</a><p>Imagine the outcry if cops started to suddenly show up all the time in places where that kind of crime happens most.
评论 #37767792 未加载
Guvante超过 1 年前
Can&#x27;t even tell what the article is about when 2&#x2F;3 of my phone screen is ads...
AnadaP超过 1 年前
You have to choose between:<p>- accurate and politically incorrect<p>- inaccurate and politically correct
lynx23超过 1 年前
Oh my, Philip K. Dick predicted this, what, 50 years ago?
repelsteeltje超过 1 年前
&gt; the software accurately predicted where crimes would occur with a “less than half a percent” success rate.<p>&gt; [previous investigation found] that cops used it to disproportionately targeted low-income communities of color.<p>TL;DR not good at predicting, pretty useful at confirming bias.
epivosism超过 1 年前
This story is so far down the argument about crime, it&#x27;s ridiculous. This is my interpretation of what&#x27;s going on:<p>stage 1: Most people know where the dangerous neighborhoods are. Police know this, locals know this, and police send patrols there, or they avoid the area and leave poor&#x2F;minorities to fend for themselves.<p>stage 2: Someone runs the query: &quot;select race_distribution from patrol_history&quot; and finds police patrol certain racial areas more or less, and sues the department. Media simultaneously writes stories like &quot;Police are harassing the population from group X&quot; or &quot;Police are abandoning protecting X areas&quot;. So the police look for &quot;scientific&quot; methods to choose where to dispatch officers since they can&#x27;t put down on paper what everyone knows - that the rent-controlled&#x2F;low wage&#x2F;etc areas are dangerous. After all, most victims of crime here are locals - random kids, storekeepers, innocent people, people mistaken for someone else, or killed to show bravado.<p>stage 3: The software looks at obvious things like past history of crime, arrests, poverty, drug use detected in water, low academic achievement, gunshot detection devices<i>, etc. It&#x27;s </i>all* correlated. So it finds the thing everyone knew - that certain areas of cities have much more open crime, and it would actually do good to increase the feeling of security around there, scare off drug dealers and people with warrants etc. So they send police to those areas; Either to harass the people, exploit them, or to help protect them. Both cases happen - i.e. in Ferguson, the police were using street crime as an excuse to harass and ticket local minorities in a disproportionate way (according to the justice dep&#x27;t investigation). But, there also actually was a lot more genuine violent crime there (which they police may not have even been helping out with). And in other cities, police are correctly going to areas which need protection, and are wanted by the majority of the local population. See surveys on high rates of minorities saying they actually want more police protection, referenced in the book Ghettoside by Jill Levoy; white liberals typically are more anti-police than black people actually living in the dangerous areas. But either way, the media can spin it as a negative - both under- and over- policing.<p>stage 4: activists &quot;debunk&quot; the crime prediction software, but if you read the debunking, it&#x27;s obvious motivated BS. None of this is necessary. Take a video camera to the tenderloin and look at the state of the people. I don&#x27;t need a PhD to know that this is a dangerous area for theft, violence, disease, etc.<p>In the end, nobody can admit that some areas need police more than others right now. It doesn&#x27;t have to be that way forever, but it is the case now. Same way a high school needs at least one or two security officers, but an elementary school doesn&#x27;t. Rather than fight to deny reality, how about we figure out how to stop lying to ourselves about what&#x27;s going on, and then get to work helping and protecting the kids who are trying to make it out of there, and immigrants who have no other place to live? The book Ghettoside is highly recommended. It&#x27;s the story of a liberal journalist who works with a right-wing coded white detective in LA who nevertheless passionately works against the police&#x27;s internal system, and the local black population&#x27;s reasonable reluctance to trust him and testify, to find the black killer of the child of his fellow detective, a black man. I learned a lot from this - things aren&#x27;t just a case of &quot;evil police &amp; good locals&quot; or the right-wing stereotyped &quot;evil poor and good police&quot; view. Both views miss the more realistic description: that the police abandoned protecting black people for a long time; black communities started to hate &amp; distrust the police for this &amp; other reasons. And so now, they are left without a good means to protect themselves except via local cultural behavior (bravado, vigilantism, etc.) So the book is a call to greatly <i>improve</i> the protection black communities receive, with their own involvement, so that they do not have to do their own self-defense anymore.<p>* yes, there are lots of articles which claim &quot;gunshot detection devices&quot; are racist. It&#x27;s hard for me to see that view, but if you refuse to admit that some areas are actually more dangerous than others, your only way out is to attack all reports and data that suggests it.<p>** I mention blacks but I&#x27;m actually making a cultural argument; if you look at culture in Appalachia, you see the same thing in whites. Groups which don&#x27;t feel like part of the majority and are left to their own internal justice systems tend to have more violence, because the systems are underground and covert. This applies regardless of race. e.g. look at crime differences between Appalachian areas and the rest of Virginia. <a href="https:&#x2F;&#x2F;www.cityrating.com&#x2F;crime-statistics&#x2F;virginia&#x2F;appalachia.html" rel="nofollow noreferrer">https:&#x2F;&#x2F;www.cityrating.com&#x2F;crime-statistics&#x2F;virginia&#x2F;appalac...</a>