Catch up on stories from the past week (and beyond) at the Slashdot story archive

 


Forgot your password?
Close
typodupeerror
×
AI Crime

New Algorithm Can Predict Future Crime a Week In Advance, With 90% Accuracy (psychnewsdaily.com) 114

Posted by BeauHD from the predicting-the-future dept.
An anonymous reader quotes a report from PsychNewsDaily: Scientists from the University of Chicago have developed a new algorithm that can predict future crime a week in advance with about 90% accuracy, and within a range of about 1000 feet. It does so by learning patterns from public data on violent and property crimes. The tool was tested and validated using historical data from the City of Chicago around two broad categories of reported events: violent crimes (homicides, assaults, and batteries) and property crimes (burglaries, thefts, and motor vehicle thefts). These data were used because they were most likely to be reported to police in urban areas where there is historical distrust and lack of cooperation with law enforcement. Such crimes are also less prone to enforcement bias, unlike drug crimes, traffic stops, and other misdemeanor infractions.

The new model isolates crime by looking at the time and spatial coordinates of discrete events, and detecting patterns to predict future events. It divides the city into "spatial tiles" roughly 1,000 feet across, and predicts crime within these areas. Previous models relied more on traditional neighborhood or political boundaries, which are subject to bias. The model performed just as well with data from seven other U.S. cities: Atlanta, Austin, Detroit, Los Angeles, Philadelphia, Portland, and San Francisco.

Lead author Ishanu Chattopadhyay is careful to note that the tool's accuracy does not mean it should be used to direct law enforcement policy; police departments, for example, should not use it to swarm neighborhoods proactively to prevent crime, Chattopadhyay said. Instead, it should be added to a toolbox of urban policies and policing strategies to address crime. "We created a digital twin of urban environments. If you feed it data from what happened in the past, it will tell you what's going to happen in the future," he said. "It's not magical; there are limitations, but we validated it and it works really well," Chattopadhyay added. "Now you can use this as a simulation tool to see what happens if crime goes up in one area of the city, or there is increased enforcement in another area. If you apply all these different variables, you can see how the systems evolve in response." The findings have been published in the journal Nature Human Behavior.
This discussion has been archived. No new comments can be posted.

New Algorithm Can Predict Future Crime a Week In Advance, With 90% Accuracy More

New Algorithm Can Predict Future Crime a Week In Advance, With 90% Accuracy

Comments Filter:

  • They discovered high crime areas (Score:5, Funny)

    by Kernel Kurtz ( 182424 ) on Thursday June 30, 2022 @05:07PM (#62663900)
    Groundbreaking stuff!

    • IF ( Area.Crime > MEDIANCRIME) THEN {return True;}

    • This kind of AI is very similar to the age old moving average which is probably at the 90% also
    • The algorithm: crime_probability = number_of_broken_windows / number_windows
      Ref: Broken windows theory [wikipedia.org]
    • I can do it even better. But that involves organizing the crime I predict. So better predictions are not always something to go for.
    • Sure, but the next version will allow police to make preemptive arrests and jail the perps before they even thought about doing a bad deed.

      • Re: (Score:2)

        by mark-t ( 151149 )

        I have no problem with a technology that has been proven being used to proactively prevent crime, but there is no permutation of this reality's future in which punishing a person for a crime that they have not yet committed is justifiable in any moral sense.

        It follows logically that if you know enough details about some particular upcoming crime sufficiently in advance of its occurrence, the worst outcomes can probably be prevented from happening if you are willing to expend the resources.

        It may possib

        • More accurately if you have enough information to determine there may be a crime in the future, you also know the reason why the crime will be committed and can work to remove that reason from consideration. I.e. If someone is about to rob a grocery store because they lack money to pay for food, you can give them food and work to give them a job so that they won't have a need to rob the store in the first place. No police required.

          Of course, that's not the real reason for these pre-crime systems to exist.

          • Re: (Score:2)

            by mark-t ( 151149 )
            Caught perhaps, stopped definitely, but as I said above there is no permutation of this reality in which it is acceptable to punish a person for committing a crime they did not actually do, even if the certainty they were otherwise going to do it was a fucking 100%. Like I said, if enough details of it were actually known sufficiently far in advance I can see no reason that any otherwise most dire consequences could not be averted, as long as one were willing to expend the resources to do so.

  • “You see the dilemma don't you?" (Score:5, Insightful)

    by rmdingler ( 1955220 ) on Thursday June 30, 2022 @05:08PM (#62663906) Journal

    Predicting a crime will occur within a window of time and opportunity manifests as law enforcement searching these parameters for a crime until they locate one... whether or not a crime may have been intended.

    • Exactly.

      Nothing like waiting for someone to inadvertently walk into a crime sce, er I mean "zone". Between that and deepfake corruption, this should put an end to all that work required to find a suspect.

      Meh. Maybe it's not that bad. After all, most people have $100K in legal funds to defend themselves, right?

    • So now we can add "area most likely to have a crime happening in the near future" to the already existing (oh, wait, no, that doesn't exist whatsoever, no way...) "people who match a profile that fits the average criminal we like to arrest".

      What could possibly go wrong?

    • Perhaps this is why the author clearly states that this technology should not be used for law enforcement, but for modeling urban policies. Maybe, just maybe, I dunno.

  • Pre-Crime Division (Score:5, Funny)

    by nospam007 ( 722110 ) * on Thursday June 30, 2022 @05:09PM (#62663908)

    Here they come.

  • Genius. (Score:5, Funny)

    by RightSaidFred99 ( 874576 ) on Thursday June 30, 2022 @05:09PM (#62663914)

    Wow, you mean you can predict crime in crime-riddled areas? Guess what? I can predict with better than 90% certainty that a raindrop will hit any given area of exposed outdoor land during a rain storm, too.

    Predicting gang shootings, muggings, and robberies in Chicago is probably actually easier. You don't need an AI to do that, you need a wall, a map, and some darts.

    • Not unless you're a pretty good dart player. Not that many crime ridden areas in Chicago. Instead it's an issue of magnitude.

    • Honestly Chicago's crime isn't that high for a city of its size and it's been going down for decades. There's a lot of sites that like to pile onto Chicago because it's run by democrats and the right wing media likes to point out crime there. They're currently pointing to statistics that show Chicago's crime has spiked through the roof ignoring the fact that the statistics only go back 2 years when we had major lockdowns. It's kind of hard to commit crimes when everybody's stuck indoors.

      Chicago is sligh
  • Will settle for Tom Cruise.

  • tool to justify more police (Score:3)

    by awwshit ( 6214476 ) on Thursday June 30, 2022 @05:18PM (#62663944)

    > police departments, for example, should not use it to swarm neighborhoods proactively to prevent crime
    > Now you can use this as a simulation tool to see what happens if crime goes up in one area of the city, or there is increased enforcement in another area. If you apply all these different variables, you can see how the systems evolve in response

    What happens when the simulation tells them that a cop every half-block covering the entire city prevents crime?

    • Re: (Score:2)

      by splutty ( 43475 )

      What happens when the simulation tells them that a cop every half-block covering the entire city prevents crime?

      Don't need a simulation to tell them that...

      • My point is that this tool will lead to absurd conclusions and be used to justify bad decisions.

        • I think this tool has the potential to be a more accurate, less biased simulator of policies than a committee of politicians if that's how you're using it. We can use it to do this, or the opposite of this, it all depends on us.

          For example such a model would have a test set, something like software testing, that can be used to certify a certain level of neutrality and accuracy. This testset can be large and complex, as detailed as you like. On the other hand humans decide on policy with much less justifi

          • Guns are tools too. Cops kill people with guns every single day. If the cops can do something stupid with a tool then they will, especially if they can point to a tool that justifies a bigger budget.

        • Re: (Score:2)

          by jythie ( 914043 )
          is that really any worse than the traditional 'do what the guts of people with no domain knowledge and do not live there think will work' that usually drives urban policy?.

          Seriously, what is people's problem with actually having a tool to explore problems? Are people just worried it might indicate something that doesn't mesh well with their personal beliefs? The same complaint people seem to have with all crime related research that doesn't give them the conclusion they want?

          • You are talking about a tool that makes up information, a tool that is wrong at least 10% of the time. The future is not always predicated on the past. This tool will produce nonsense, which will waste police time and implicate innocent people. Not all tools are good tools.

            • Re: (Score:2)

              by jythie ( 914043 )
              From the sounds of it, the tool is intended for experiments to inform policy decisions, not implicating specific people. The null prediction is just a validation tool, the point seems to be to input changes to policy and see how it impacts crime rates. You are confusing a pretty straight forward modeling system with some kind of sci-fi dystopian thing.

      • Yes you do. Now it's not just a crackpot "hard-on-crime" fantasy, it's scientific.

  • Stock prediction (Score:5, Interesting)

    by BytePusher ( 209961 ) on Thursday June 30, 2022 @05:20PM (#62663954) Homepage
    I was able to make extremely predictive and profitable stock trading models using machine learning, but the problem was I had many variables and poor discipline about unsealing my out of sample. So in effect, my out of sample became in sample as I tried more and more parameter variations. My model essentially memorized the testing dataset. I would want to know their methods and validation procedures before trusting that this is somehow a better model than a much simpler model like SVM.
    • Predicting crime is more like predicting the weather than predicting the market - which is to say, fundamentally easier. The weather does not respond to a correct prediction by doing something else instead, like markets do. (Unless your prediction is somehow far better than everybody else's and you're a fairly small actor). The weather is chaotic (butterfly effect etc), but markets are worse - accurate prediction is literally paradoxical - self-falsifying.

      Admittedly if you had a good crime-prediction mo

      • Re:Stock prediction (Score:4, Informative)

        by sphealey ( 2855 ) on Thursday June 30, 2022 @09:25PM (#62664536)

        "Admittedly if you had a good crime-prediction model and acted heavily on it by allocating police forces"

        Can't say I agree with your police work there, so to speak. For example if you have a crime-prediction model developed at the deeply right wing University of Chicago, an institution that has been working to seize control of the entire Hyde Park area of Chicago for 60 years and repopulate it with upper class U of C types, and you provide that model to police who uses its output to arrest every black male in sight, that could very well lead to the black community becoming enraged over selective, biased, and unfair policing that doesn't even lower the crime rate, and that rage could lead to increased levels of apparent and even actual crime.

        Not that this scenario has been playing out around the U of C, sans the AI, since 1965. Unpossible.

        • Fair point. I suppose treating people like criminals even if they aren't could cause them to actually become so.
    • I'm more concerned with feedback effects. Making economic predictions results in human behavior that makes those predictions less accurate. The same principle likely applies here. The model output drives police behavior, and results in the prediction being inaccurate. That's likely why they don't want the model used as an input to law enforcement policy.

  • and the input data ... then I would run it, look for an area without crimes predicted and go there to be evil. BooHaha!

  • Like weather predictions (Score:5, Insightful)

    by Tony Isaac ( 1301187 ) on Thursday June 30, 2022 @05:32PM (#62664022) Homepage

    If you always predict that Los Angeles will be sunny tomorrow, you'll be right 80% of the time. I've seen TV weather forecasters brag about 80% accuracy rates. Not exactly difficult!

  • there are roads where the speed limit is broken 99% of the time.

  • Oh goody, high tech broken windows policing (Score:4, Insightful)

    by rsilvergun ( 571051 ) on Thursday June 30, 2022 @05:44PM (#62664062)
    So here's the thing with what they're doing. It's called "Broken Window Policing". It's been thoroughly debunked.

    The way it works is you over police poor and/or minority neighborhoods. You then find more crime because you're looking for it. When income levels are accounted for you're not really finding any more crime, but you are giving a bunch of people criminal records for minor offenses. This in turn leads to fewer job opportunities, lower property values, underfunded schools because we use property tax to fund them and a death spiral for the entire neighborhood.

    It was started in the 60s to enforce segregation without running afoul of the Federal Courts. It continued even afterwards because cops like easy work, and they have stats and defacto arrest quotas. Going into a neighborhood where you could hassle people who didn't want you there in the first place is a good way to do that.

    We know exactly what reduces crime: clean air and water and jobs & education. Anything else is just an excuse to shit on people you don't like or somebody's brother in law getting a juicy gov't contract for phony new tech.

    • Broken window policing doesn't apply to murder rates. Or motor vehicle theft. Or the other categories that the AI uses. It's almost like they explicitly excluded broken window crimes or something.

      • because it increase murder and vehicle theft rates by saddling the community with a ton of people with convictions for petty crime and in turn cratering the community.

        And if you think this'll only be used for murder & vehicle theft you're too naive. Cops have quotas. Anything goes when you've got a quota to meet.

        • Huh, I didn't even notice it was your stupid ass I replied to. Why am I not surprised. Also you're incorrect. The one thing police themselves can do to consistently drop crime rates is concentrate on street patrols. Bullshit like stop and frisk is not needed. Mere regular police presence and attentive response to calls has an obvious effect on crime rate. Of course, more street patrols means less overhead for administrative graft. This is especially bad in Chicago because about half of Chicago 'police offic

      • Broken window policing creates neighborhoods filled with poor people with criminal records. Poor people with criminal records steal more cars because 1. they are poor, and as such need money; and 2. they have far less stake in a society that has decided they are no longer a valued part of it, because of that one time they were caught spray-painting their name on a building while black by a cop who really hates that. They also value the lives of others a lot less for many of the same reasons. This is somethi

    • Re: (Score:2)

      by jythie ( 914043 )
      Is it though? It mentions allocation of enforcement resources, but also other factors. While I am not familiar with this project, I work on another simulation designed to answer these kinds of questions and 'allocation of security forces' makes up only a few of the dozens of levers that one can pull to see what effect it might have.... including the ones you mention.

      I am a bit bothered by a bunch of 'nerds' jumping to the 'more knowledge is bad, we should just do what feels right!' stance every time rese

  • Reported Crime* (Score:4, Insightful)

    by Gravis Zero ( 934156 ) on Thursday June 30, 2022 @05:50PM (#62664070)

    There is a shitload of white collar crime that goes on but doesn't get reported for various reasons. You cannot track (and therefore predict) crimes which are never reported.

    • Restrict your attention only to clear-cut crimes like murder, and you still see all the same patterns you are trying so hard to explain away.

      • "Clear-cut"? Embezzlement is a clear-cut crime and yet a lot of times corporations will just fire the person and not report it because dealing with the police costs them even more money.

    • Re: (Score:2)

      by jythie ( 914043 )
      Yeah, but urban crime is something that scares suburban taxpayers, so that is the only kind people really care about.
  • It's AI, so the government will just throw tax dollars at it.

  • You are being watched. The government has a secret system, a machine that spies on you every hour of every day. I know because I built it. I designed the machine to detect acts of terror but it sees everything. Violent crimes involving ordinary people, people like you. Crimes the government considered "irrelevant." They wouldn't act, so I decided I would. But I needed a partner, someone with the skills to intervene. Hunted by the authorities, we work in secret. You'll never find us, but victim or perpetrato

    • I developed a brief episode of addiction to Person of Interest. Living in a disused library sounded interesting, but I then realized that I might already do so.

  • Maybe there's more to it? (Score:3)

    by King_TJ ( 85913 ) on Thursday June 30, 2022 @06:48PM (#62664182) Journal

    Initially, I thought the same thing most of you commented... This is just more "Minority Report" type nonsense. If you take a high crime area like parts of Chicago, of course you can show a 90% chance a crime will happen again in the right spot.

    But .... I think I can see how this might have more usefulness if it's good at predicting the way crime spreads and moves about? In other words, no ... this isn't a tool to try to find out where the next crime is supposed to occur. But maybe it can help build a current "heat map" of the areas in a city that are higher vs lower crime. Right now, you can view these types of maps on a number of web pages if you're interested in knowing what parts of town might be safer to buy a home in or rent a place in. But they're based on static crime data that's probably a year out of date. I'm not sure it's a great assumption that these high crime areas are all that static in nature? If an A.I. predictive model can make some of them shrink and make others grow in a pretty accurate way based on time that's passed since the last set of actual stats were fed in to build the initial map, that would be impressive.

    Still, I'm not sold that such a thing can really be done. How does the modeling predict that a given "nice neighborhood" is due to have higher crime just because crime settles down in another?

    • Re: (Score:2)

      by Junta ( 36770 )

      I'm just skeptical the 'forecasting' adds anything more than just looking at the prior week's data. For seeing the situation evolve, reviewing the last few weeks' data over time.

      I wonder if a program that just labeled the map of last week's data as the 'forecast' for next week without any production, would it also achieve a 90% accuracy given the same requirements.

      Part of me wants to think that they would know more than a casual observer, but also must recognize this is a commercial endeavor and thus at ri

    • Re: (Score:2)

      by jythie ( 914043 )
      So, I can't speak for this particular research, but I work on another simulation that has a similar kind of goal to it so I might be able to partially speak to its utility.

      For systems like this, the ability to predict is more of a validation method than a primary function. If it can predict moving forward that means the model is working, which means you can change inputs, meaning you can run experiments to try to answer the question 'what policies help, what policies hurt?'.

      At least in my domain, policy

  • The paper is not free, so all we can read is the abstract, which doesn't say much about key details. For example, what exactly is being predicted. From the abstract it seems like the city is divided into 1000-ft x 1000-ft areas, and then crime of some categorization is predicted within that area. It's nearly impossible to predict the exact time of any crime, so obviously there is some allowable time window within which a prediction is considered a match. Is that time window on the order of minutes or da

  • Let's abuse some statistics.

    If we assume that crime events are independent of eachother, then you can roughly say that the number of crimes on a given day (or week, or month) obeys Poisson statistics.

    And you can back out what 90% accuracy actually means. To keep the math easy in my head, if N crimes is predicted as a 90% confidence interval, then roughly speaking that's about 1 sigma (and change) above the mean number of crimes*.

    One sigma is sqrt(mean estimate) so for example if the mean estimate is 100 cri

  • Person of Interest? https://www.imdb.com/title/tt1... [imdb.com]

  • Give it some time, and — like those before it [archive.org] — this technology will be denounced as racist too.

    • There's now two possible reasons. First would be that we feed the algo with racist information, then it will create a racist output. Garbage in, garbage out.

      The second is that there is a certain, let's call it "bias", when building places for the poor and stuffing them into bad neighborhoods, with poor people also being mostly people of color.

      So which one is it?

      • Re: (Score:2)

        by mi ( 197448 )

        There's now two possible reasons

        The one and only reason is that an accusation of "racism" is so dirty and toxic, that it is impossible to debunk — and is used by the unscrupulous to reject any idea, and any approach, that they don't like for whatever reason.

        when building places for the poor and stuffing them into bad neighborhoods

        You got it backwards. A neighborhood goes bad, when the kind of poor, who couldn't afford a home before, move in with government's assistance.

        One can see this first hand

        • And here's the thing: We don't stuff our poor into projects on the outskirts of the town. Where they get to wallow in the cesspool or crappy infrastructure, crappy housing, crappy environment and crappy utilities. We spread them out across town. Yeah, that means that these ... funny looking kinda-humans, ya know what I'm talking about, that they may even live next door to you. Teh horrorz! Funny enough, that doesn't cause any trouble.

          People generally don't try to be assholes, given the choice. Apparently th

          • Re: (Score:2)

            by mi ( 197448 )

            We don't stuff our poor into projects on the outskirts of the town

            You missed most of what I said in order to make some jingoistic point of your own — which attempt fell flat, because I don't even know, which hole in the globe your particular kind of asshole dominates :)

            • It's called Vienna. According to Mercer and The Economist the most livable city in the world.

              • Re: (Score:2)

                by mi ( 197448 )

                It's called Vienna. According to Mercer and The Economist the most livable city in the world.

                And also the hub of Russian spying [bbc.com], khmm...

                Nice confections, though, I'll grant you that :)

                • What do you expect if you have a conservative party in the government for over 35 years?

                  • Re: (Score:2)

                    by mi ( 197448 )

                    What do you expect if you have a conservative party in the government for over 35 years?

                    Vienna was a spying hub through the entire Cold War [thedailybeast.com].

                    We're far enough away from the topic of artificial intelligence, its supposed racism — and even the government-provided housing — that I can stop responding.

                    • Vienna was one of the few places in Europe that was neutral and still accessible for both sides of the cold war. Of course it was spy-central.

                      But we can certainly return to the point of where and why AI gets racist. Which actually opens up an interesting question: Is racism inherent in any kind of examination? Is it even possible to get an unbiased evaluation of a person or a situation? It's in my opinion coming down to the question nature vs. nurture. Are people the way they are because of their ancestry o

  • Subject says it all
  • I'm sure this tool will make it very easy for the racist criminal justice system in this Country to fill up all those for-profit prisons with people of color as free labor.

  • Tom Cruise's opinion (Score:3)

    by Walt Dismal ( 534799 ) on Thursday June 30, 2022 @08:07PM (#62664386)
    I asked Tom Cruise which is his favorite AI for prediction and he said 'Get stuffed, you mutant jerk". Then he roared off on a motorcycle and parachuted it off a building.

  • I saw this movie (Score:4, Funny)

    by nucrash ( 549705 ) on Thursday June 30, 2022 @10:05PM (#62664584)

    It didn't end well.

  • Comment removed based on user account deletion

  • Any crime prediction algorith is one of two things:

    A) bound to be called racist
    B) bound to deliver very inaccurate predictions

    A crime prediction mechanism, whether it is people or systems must always look at what IS, and disregard the WHY. What IS is that crime is highly predicated on racial and ethnic factors (in addition to a lot of other factors, who are also confounded with racial and ethnic factors, so much that their linkage is inextricable).

    This is seen as "racist" in today's society, because society regards the WHY of these differences to be rooted in racism and therefore prohibits the observation and attribution of observations and patterns.

    This is a fundamental problem in law enforcement, because law enforcement cannot and should not care about the events, systems and developments years before a crime that caused the perpetrator to become criminal. When a system is leading people into different paths according to their racial background, with one becoming criminal and the other not, it is a societal problem per se, but at the time law enforcement becomes involved, it is far too late to care WHY the person became a criminal. And society somehow tiptoes around this distinction and demands law enforcement, when the crime has already happened, to somehow rectify all problem society as a whole couldn't resolve in the last two hundred years.

  • Fundamental flaw: The model is based on historical data so the model is only as accurate as the real world is similar to the model. If the real world changes, e.g. effective crime reduction interventions, then the model is no longer predictive. In other words, the model only works as long as nothing changes. Also, as we've seen of these kinds of algorithms before, they tend to entrench existing cultural biases & prejudices. In other words, culturally & prejudiced law enforcement practices produce hi

  • Here's what you do.

    1) There are areas known for a high incidence of crime. Withdraw police from those areas, if they have not already run away to avoid being killed by the ruling gangs. This will concentrate crime in areas already known for it, and drive up your crime prediction stats.

    2) Increase the number of crimes, such as "Walking while brown in the city of Detroit", and "Talking funny to an officer of the law". If your crime prediction stats are looking a bit low, go and round up the usual suspects. If

  • Lead author Ishanu Chattopadhyay is careful to note that the tool's accuracy does not mean it should be used to direct law enforcement policy; police departments, for example, should not use it to swarm neighborhoods proactively to prevent crime,

    If you did that, it would drive down its accuracy rating.

    It is obvious areas that are less policed have higher rates of crime. How is this news?

  • Total nonsense. Anyone who believes this is a moron. You want to predict future crime? Check their arrest record.

Slashdot Top Deals

Chemist who falls in acid is absorbed in work.

Close