Slashdot Mirror
Academics Confirm Major Predictive Policing Algorithm Is Fundamentally Flawed (vice.com)
An anonymous reader quotes a report from Motherboard: Last week, Motherboard published an investigation which revealed that law enforcement agencies around the country are using PredPol -- a predictive policing software that once cited the controversial, unproven "broken windows" policing theory as a part of its best practices. Our report showed that local police in Kansas, Washington, South Carolina, California, Georgia, Utah, and Michigan are using or have used the software. In a 2014 presentation to police departments obtained by Motherboard, the company says that the software is "based on nearly seven years of detailed academic research into the causes of crime pattern formation the mathematics looks complicated -- and it is complicated for normal mortal humans -- but the behaviors upon which the math is based are very understandable."
The company says those behaviors are "repeat victimization" of an address, "near-repeat victimization" (the proximity of other addresses to previously reported crimes), and "local search" (criminals are likely to commit crimes near their homes or near other crimes they've committed, PredPol says.) But academics Motherboard spoke to say that the mathematical theory that is used to power PredPol is flawed, and that its algorithm -- at least as pitched to police -- is far too simplistic to actually predict crime. Kristian Lum, who co-wrote a 2016 paper that tested the algorithmic mechanisms of PredPol with real crime data, told Motherboard in a phone call that although PredPol is powered by complicated-looking mathematical formulas, its actual function can be summarized as a moving average -- or an average of subsets within a data set. "The academic foundation for PredPol's software takes a statistical modeling method used to predict earthquakes and apply it to crime," reports Motherboard. "Much like how earthquakes are likely to appear in similar places, the papers argue, crimes are also likely to occur in similar places. Suresh Venkatasubramanian, a professor of computing at the University of Utah and a member of the board of directors for ACLU Utah, told Motherboard that earthquake data and crime data are, naturally, collected in different ways."
"I would say in our mind, the key difference is that in earthquake models, you have seismographs everywhere -- wherever an earthquake happens, you'll find it," Venkatasubramanian said. "The crux of the issue really is that to what extent are you able to get data about what you're observing that is not also totally on the model itself." "If you build predictive policing, you are essentially sending police to certain neighborhoods based on what what they told you -- but that also means you're not sending police to other neighborhoods because the system didn't tell you to go there," Venkatasubramanian said. "If you assume that the data collection for your system is generated by police whom you sent to certain neighborhoods, then essentially your model is controlling the next round of data you get."
The company says those behaviors are "repeat victimization" of an address, "near-repeat victimization" (the proximity of other addresses to previously reported crimes), and "local search" (criminals are likely to commit crimes near their homes or near other crimes they've committed, PredPol says.) But academics Motherboard spoke to say that the mathematical theory that is used to power PredPol is flawed, and that its algorithm -- at least as pitched to police -- is far too simplistic to actually predict crime. Kristian Lum, who co-wrote a 2016 paper that tested the algorithmic mechanisms of PredPol with real crime data, told Motherboard in a phone call that although PredPol is powered by complicated-looking mathematical formulas, its actual function can be summarized as a moving average -- or an average of subsets within a data set. "The academic foundation for PredPol's software takes a statistical modeling method used to predict earthquakes and apply it to crime," reports Motherboard. "Much like how earthquakes are likely to appear in similar places, the papers argue, crimes are also likely to occur in similar places. Suresh Venkatasubramanian, a professor of computing at the University of Utah and a member of the board of directors for ACLU Utah, told Motherboard that earthquake data and crime data are, naturally, collected in different ways."
"I would say in our mind, the key difference is that in earthquake models, you have seismographs everywhere -- wherever an earthquake happens, you'll find it," Venkatasubramanian said. "The crux of the issue really is that to what extent are you able to get data about what you're observing that is not also totally on the model itself." "If you build predictive policing, you are essentially sending police to certain neighborhoods based on what what they told you -- but that also means you're not sending police to other neighborhoods because the system didn't tell you to go there," Venkatasubramanian said. "If you assume that the data collection for your system is generated by police whom you sent to certain neighborhoods, then essentially your model is controlling the next round of data you get."
bonjour, mesdemoiselles
(tips beret seductively)
Only has so many police to send to any neighborhood.
Have all the skilled police waiting around a low crime neighborhood is going to add to support times in crime filled neighborhoods.
The ability to pack a lot of police into the "neighborhood" that is full of criminals allows a rapid response to crime and criminals.
Call for help get paid police action.
More police are in the area to support the number of calls.
Don't have skilled police waiting around in the better low crime neighborhoods.
The "system" quickly works out what neighborhoods have the crime.
To map out the better neighborhoods with normal people who don't do crime all day and night.
The system is working perfectly and allows a city to place police jwre needed, in the most crime filled neighborhoods.
Domestic spying is now "Benign Information Gathering"
By targeting a portion of a city or neighborhood with more LEOs, you will create more arrests there.
There are undoubtedly several areas in virtually every city that are predisposed to criminal activity... areas that have a high concentration of poor people or others that are resigned to accept more overt peddling of the drug and skin trades, for instance. If you're looking to prove an algorithm's efficacy, deploying additional police personnel and assets to any of these areas will get it done.
Happiness in intelligent people is the rarest thing I know.
Ernest Hemingway
when you do everything in your power to deny what is undoubtedly obvious: black people commit a lot of crimes.
If it claims to use AI for something that cannot be programmed, then it is fundamentally flawed at least for the next 25 years or so.
Laws are rules for the court, but merely a bottom bar to hit for life. Think beyond laws in your actions always.
Today it is taboo to discuss that few minority populations responsible for 90%+ of all urban crime. Without these models, police would be faced with wasted effort of policing no-crime white suburban neighborhoods instead of trying to stop black gangs from killing each other and selling drugs to black children.
Software used for predicting the future shown to be poor at predicting future.
Chief of police: "Their salesman said it worked and he was a real nice guy"
Science man: "What the fuck where you expecting, next weeks lottery numbers?"
Clairvoyants association spokesperson: "nuh uh, we legit"
more at 11.
Clearly the police crime prediction algorithm failed because it wasn't granular enough. Collect data on everybody. Absolutely everything everyone does, where they go, who they talk to, what they buy, what they read, what they look up on the internet. Mash that into a ML algorithm. After a few criminals are found, more and more will be sifted out as the algorithm learns and becomes better at identifying criminal thought patterns and behaviours. What a glorious future awaits when crime is impossible! :D
The data fed to the model is not just from police sent to neighbourhoods the model sends them to as suggested. The data is gathered from _all_ police dispatches. So the suggestion that the model is reinforcing its own biases is wrong.
"Consensus" in science is _always_ a political construct.
I have mixed feelings about this.
First, the idea that algorithms alone can 'predict' something as subjective, human, and impulse-based as crime is ridiculous, and (I believe) born of a Utopian idea that taking people out of the equation can somehow remove bias, racism, and subjectivity from the process leaving some sort of idealistically mechanical, sterile system. For anyone who's worked in policing, crime prevention, or law enforcement fields, this should be a staggeringly stupid idea. Who's writing such algorithms but other people? On top of that, I expect there are now encrusted layers of ideology, in which results that don't conform to some utopian ideal of demographics are claimed to be 'racist' and formulaically 'corrected' to suit political goals, regardless of the facts of reality.
OTOH there is ABUNDANT work that shows that recidivism, particularly in the worst crimes, is concentrated in a surprisingly small number of individuals. I worked for a police dept where the longest serving officers maintained that 80%+ of the crimes were committed by a handful of families in the 50,000+ person city.
https://www.politifact.com/tex... lists some examples:
So it's clear that if we could identify this small percent and aggressively police them, we could make a sizable impact on crime.
-Styopa
a predictive policing software that once cited the controversial, unproven "broken windows" policing theory as a part of its best practices.
The only thing "controversial" about broken window theory was that it worked.
The fact that it worked really bothered a certain class of people (people who wanted crime to be the fault of something other than criminals).
You don't need police to report a crime, anyone who is a victim of a crime can report what and where it happened, just like an earthquake sensor.
This is not "predictive policing", this is Profiling. It already exists, there is already a term for it, no need to try and coin a new term.
https://law.justia.com/cases/c...
So.. while not an academic, this is pretty close to my field of research. Looking at their model, I am not surprised they sold this product but deeply disappointed. This is the type of model that is REALLY easy to sell to people, both law enforcement and the military (our customer) are enamored with them for their near magic ability to 'predict' things. Only they don't, they tend to fail in unpredictable ways. They are not bad in multi-model systems where you take a dozen or so different systems built by different teams, run them in parallel, then have subject matter experts ponder the conflicting results. But actual police out of a single model? Madness... or hubris.. or stupidity... or simply being enamored with a slick sales pitch from 'one of your own' offering to solve problems in the way you want them solved.
Oddly enough, we actually DID do a LEO model years back, which was actually pretty effective, but it encouraged things like community outreach and police/citizen interaction which worked really well for officers on the ground but pissed off lawmakers and 'police unions', so it was largely dropped.
Which gets back to this story and one of the fundamental flaws in such attempts. The decision makers are not interested in solutions that make things better for high crime areas in the first place, the people in those areas are not part of their power block. They want solutions that 'sound right' to people who live elsewhere and confirm what they already believe. Which is exactly what models like this are good at producing. They are kinda like torture... useless for prediction or information gathering, but an excellent political tool for confirming the story your career depends on being 'true'.
Police typically respond to calls, not just cruise around and "spot" crime. The fact they're in a neighborhood cruising already does not increase the likelihood of a call coming in; it may shorten response time, but it does not cause the call in the first place. If anything, it would tend to depress illegal activity in that neighborhood...
Browsing at +1 - no ACs, I ignore their posts. So refreshing!
No. You did not hate to use Trump as an example. I am sure you also conveniently forgot or more likely never knew you you only watch socialist news sources that Trump is the only national level candidate to ever openly declare the DNC ghetto crime bullshit has to stop and he wants to help.
Your socialist masters love seeing blacks stay in ghettos with DNC created poverty, crime, and anti-education ghetto culture.
We didn't have the hospitalization data, but we had everything else (and it was admittedly patchy, humans are more unpredictable than they are predictable in longer time frames of observation), so all of the models 'predictions' were often a guess based on... moving averages and baselines.
I would propose PredPol should be shut down as immediately as our investors pulled out when they found out the truth: You won't fix this problem in PredPol, it would be too expensive for PredPol and the product would likely fail.
That and this is unethical just like telling patience they are in the clear based on their 'stats' when really their 'stats' just look OK compared to a community baseline (which was again proven to need to be un-correlated in our experiments).
If we as a society allow our law enforcing to become a lazy activity like our news reporting has become, the f ups in minority report will be the least of an every day persons problems.
And finally since nobody else touched it: PredPol is anti first amendment. /chew on that
The key phrase was that -if- you only use LEO on site reports for your data -then- it is self reinforcing.
Crime statistics are -not- gathered by LEO on site.
These academic morons wasted 7 years of my taxes on a study with a hypotheses so flawed even a /. AC could spot it -if- they bothered to read the summary.
Go back and read the last few lines of the summary.
If you are unable to instantly spot the fatal flaw in this useless study which wasted 7 years of tax payer dollars to come to a flawed and politically pre-determined solution then at a minimum you should stop voting and never breed.
I mean, is this a way to push money on things to hide criminal activity instead of really addressing the real problems that keep causing these activities?
In the words of EX narrator: "Design is about solving problems not hiding them".
I can't help but to see this as a warm patch instead of an actual medicine, having a notion of where a crime could be perpetrated in an area is useless if this keeps happening, at the end we will end up with a full surveillance system attached to our bodies... and when that happens we will be truly slaves of a few unknown people's fears.
that violent crime is less common then we like to think? Of course a small percentage is responsible for all violent crime. There isn't enough to go around.
Also, not sure about Sweden in the 80s but in America, even today, our prison system chews you up and spits out broken people. That's been equally well documented.
Finally pre-90s is a bad place to get crime statistics from. Lead in the air was pretty obviously creating unhinged people. Again, there's plenty of studies to back this up because it's the only thing that can explain the across the board drop in crime.
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
I'm a retired LEO. About 20 years ago my knack for coding, web development, and data mining got me baited-and-switched into Admin to make rudimentary 1980's systems accessible to desktop browsers, which led to developing over a dozen new DBs. We had 380+ sworn officers, 300K residents (plus 50K+ undocumented), and daytime population over 700K.
I don't know PredPol or its issues, but we used other crime analysis software that mined my new DBs and we had surprisingly good results. But to develop better predictions (it's all an "educated guessing game") you not only need clean normalized data, you also need to "understand" your data, know your tools AND their limitations, and correctly account for environmental variables (obstructions and barriers such as rivers, freeways, and traffic hours; patrol officer staffing densities and habits; etc.). Push-button analysis is not analysis.
In fact, as our Crime Analysis Unit was ramping up but not yet operational we had a serial rapist who, despite "old school" random surveillances, was eventually caught by citizens. But when we backfilled data (time machine back a few months), many "predictions" were spot on for timing and location (but not both together, which would have been way too easy) which could have helped to more effectively allocate surveillance resources and not blow up the overtime budget. Also, the prediction as to where the rapist likely lived, was 800' from his actual apartment. It was a great test case to learn our tools and incorporate those environmental variables.
But it's important that officers and detectives have faith in the numbers and analysis. All tools fail if people don't use them. I get why some of my fellow old school LEOs were initially reluctant, but when the predictions' batting average is getting higher than hunch-analysis, then maybe give it a chance. Case in point, a friend working an armed robbery spree didn't want anything from the unit. He finally threw in the towel, surveilled a prediction, and - BANG - the perp got shot the first night! Double win! People started requesting more analysis and predictions.
Related to crime analysis are workload studies to determine patrol staffing into their dedicated patrol areas. This usually relies on historical call data but variables are also important (beat size, traffic patterns, speed limits, ideal priority response times to critical incidents, distance to station and jail, how much proactive time do you want from officers, how often on average should a resident see a patrol unit go by, etc.). The actual end goal is "beat integrity" to balance the "average" of how often officers have to leave their beat to assist other beats, other than critical incidents (if a baby is choking, screw district integrity, closest officer wins).
So we crunched a full year of call response data, gathered our variables, and proposed a redistribution. One Commander in particular was pissed because he'd lose the most officers, but the the Chief bought in, so this better damn work! With only a single officer exception, we redeployed to the study. The year we redeployed was bumpy (the redeployment itself, and we sponsored a major regional LE event that affected everyone's schedules), but the following full year had amazing results. With only a 230 incident (0.1%) and 1/3 patrol officer (0.2%) difference between the 2 study years (as close as they can get!) the average integrity went from a range of 70% to 91% to a flat 87% across the board, with the only exception being the beat with that one extra guy which had fractionally less integrity and slightly more proactive activity (makes sense, because the extra body could help other beats more often and have more time for self-initiated activity).
that's corralling it. Stick with me on this, it's long.
I saw this in my old town (I live in an apartment full of Indian H1-bs now and while they do take my jerbs they're about the least criminal people on earth if you don't count them hacking my cheap router from time to time).
I used to live in a cheap, working class neighborhood. Up the street were crack houses galore. Never once did they bother me. Only trouble I ever had was from a loser friend who's girlfriend robbed me.
I used to wonder why and then I found out. "Broken Window" policing is actually "Busted Heads" policing. My bud lived in one of those crack house apartments following a bad divorce (only thing he could afford/get after getting his credit destroyed). Got robbed, they caught the guy when his apartment manager went into the apartment and recognized his stuff there. They released the guy a few days later and he was still living next door to my bud when he finally got his credit fixed and moved out.
That was OK because they'd kept the crime inside. Every now and then one of them would venture out of their little hell hole and rob a liquor store or something.
The cops would come down like a ton of bricks. Everyone got arrested. And since they all had at least some pot half were probably gonna do a year or two in the clink. Especially the Dads, who would take the rap for the pot so the mom could at least stay out of prison. And during the raid you better believe heads got busted like crazy.
This kind of shit is used to force the lower caste to stay in their lane. Keep their head down. It's the nastiest form of oppression possible. It lets you and me ignore the problem of widespread poverty because when the poor make trouble there's a cop there ready to beat them the fuck down and a private prison system happy to lock 'em up for 3-5 years.
This is also why the drug war hasn't ended. Locking up randos for minor drug crimes is how you keep those folks on edge.
Now, you might be thinking, so what? They're criminals anyway. That's all well and good, but think about it. When Capitalism goes south what's suppose to fix it? The Answer is that Mr Factory Owner won't let the country go to shit because he lives in it. But Mr Factory owner and even his middle class servants can use tricks like this to control the populace why bother? What's to stop him from letting everything go to shit except where he lives? This is where oligarchy comes from.
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/
Racism doesn't help profiling, but the inherent reason it's flawed is because it infers guilt rather than proves it. Essentially guilty before proven innocent. It's completely backwards from the values we hold and the way we want law enforcement to serve us.
This criticism is speculation. The way to test the software is to test the software. Simply run it but take no action. How well did it predict? That is the way any prediction algorithm, or medical test is rated. Other than that, it's just politics among blowhards.
Just like gravity is only a theory.
Various experiments including in the Netherlands have shown that the broken windows strategy works.
eom
n/t
Yeah, those pesky domain experts and their actual knowledge of systems with proper tools of validation. So stupid.
Here's a secret, a lot of crime is near section 8 housing. This is at least from my personal experience. One place I grew up as a kid was all section 8 housing, and had all kinds of crimes (drugs, rapings, muggings, burglaries), are just some of the experiences my siblings or I went through. That area now has no section 8 housing anymore and is all gentrified and safe. As an adult, I lived in another area where across the street was section 8 housing. My side of the road had a part were parents pushing their kids on swings and all butterflys and such, but on the other side that park usually had teens doing drugs. The 2 parks were only about a block away from each other, but totally different. About the article, having non-police also make reports would be an interesting idea. I'm thinking neighborhood watch groups and such. I think mixing section 8 housing with non-section 8 housing is the way to go (think diversification). That is, I think its better to have the poor, middle, and rich not completely isolated from each other. So instead of area X is for poor, area Y is for middle class, and area Z is for rich, to have them all live in the same zipcode.
The fact is that the majority of the perps constitute a small minority of the minorities.
When Fascism comes to America, it will call itself Anti-Fascism, and tell you to give up your guns.
Crime happens everywhere but it is not always reported. The worst, most critical crime in terms of its overall effect upon the public is never subject to a police dispatch. It happens at 1600 Pennsylvania Avenue and environs in Washington DC. It happens at Wall Street. And at Caracas, Mexico City, Dubai, Tel Aviv, Moscow, and even Geneva. It happens in marble buildings and fancy restaurants and hotels and involves people wearing Armani suits and fake sun tans. These crimes involve investment schemes, trade in drugs & guns, overthrowing governments, usurious loans to third world countries, manipulation of elections, monetary gifts to dictators and warlords, currency manipulations, deforestation for private profit and more. These criminals enrich each other at the expense of billions who suffer from the ravages of war, hunger, disease and deprivation.
The police are not called because these perpetrators are above the law.
...omphaloskepsis often...
Except the ISSUE IS NOT REPORTING or "noticing crimes" - IT'S PREDICTING CRIMES that's the issue.
Algorithm doesn't notice nor report crimes. It predicts where to send the police.
Resulting in a "garbage in = garbage out" predictive result based on reinforcement of outdated data.
E.g. If there was an arrest of a guy selling pot in front of a local Starbucks last month, and another guy arrested selling meth in a parking lot of a mall - algorithm now dictates through "near-repeat victimization" that both the Starbucks and the mall AND EVERYTHING AROUND THEM are likely locations of future crimes.
And should cops actually notice something in that area aroooouuund the location of a previous crime while being under pressure to fulfill their monthly quotas - it is seen as a validation of the predictive powers of the magical AI.
Rinse and repeat.
It's "Round up the usual suspects!" - only with locations and "supported" by math.
Pretty soon you have cops policing parking lots for broken tail lights and ID-checking everyone around a Starbucks, falling number of arrests for preventable crimes (such as selling drugs or opportunistic crimes) - with actual number of crimes on the increase city-wide.
Cause everyone is listening to the magical algorithm, designed to predict earthquake aftershocks.
Instead of having police patrolling even there where no crimes are being reported - e.g. cause the locals don't trust the police or are afraid of reprisals from the drug dealer next door.
Mit der Dummheit kämpfen Götter selbst vergebens
IMHO, the police can decide for themselves, if any software they use is really worth it, or not!
If any software they use, causes wasted effort, I for one, have no doubt, they would notice it!
So, IMHO, there is absolutely no need for any academics to try to analyze effectiveness/correctness of any software used by Police!!!
(Unless, any Police force specifically asks/needs/employs them to evaluate any of their software, of course!)
In the quote: "If you assume that the data collection for your system is generated by police whom you sent to certain neighborhoods, "
I would think the data is obtain by citizen everywhere around the city reporting crimes, be it to police directly or via insurance biz (who usually require a police report).
Furthermore - it should be fairly easy to verify if the system works after so many years. Are crimes not reported as predicted?
YES if it works you are influencing the model, but that is not necessarily a bad thing. You successfully disrupted the criminals; putting them out of their comfort zone, easier to catch elsewhere. Then it's harder to gather new data for a while.
Some murder go unreported, or at least are reported as disappearance and not necessarily in the area they happened ! If somebody get ride of the body, and has 40% chance to get out of it some murder will necessarily never know where they happened) Furthermore many burglary are not reported due to various factor, fear from the person itself (maybe what was stolen was illegal, maybe the person was threatened, maybe the person is illegal, maybe the person don't trust the police, maybe what was stolen is low value etc....). Not all crime are reported or detected, not even murder (I won't even go into the resolve rate which is around 60% in average in the US). So yes even for burglary or murder, detection is a problem.
C. Sagan : A demon haunted world:
http://www.amazon.com/gp/product/0345409469/
visit randi.org
For decades the police and "justice" system had a sweet thing going. Grab anyone they fancy for a crime and if there's no evidence, coerce a confession and/or produce a suitably trained and bribed witness. Then along came DNA testing and ruined their system, evidenced by the increasing number of incarcerated people who have been exonerated.
PredPol is a gift from heaven. They can now claim arrests based on "scientific proof", but in reality it will just be a return to the old days where they could frame anyone they wanted to. After all, it's a numbers game. Make an arrest and the crime is marked solved; off the books. Doesn't matter if the poor bastard is innocent or guilty.
Crime follows poverty, inequity, and addiction. Fix your society to live up to the claims of your propaganda and get above an African-level of wealth desparity and you might start to lower your crime ratea.
But youre mostly racist assholes so Blah people right!
Did you catch this quote from the article? police shootings between 2011 and 2015 were 3.49 times more likely on average to target black individuals compared to white
Umm... gee... check the murder rates. Blacks kill blacks at a rate six times more than whites kill whites. Which means both they kill and they are killed. So the police could be considered as under-performing here. That stat was provided out-of-context, much like the rest of it. For example, the study... what did it say? Basically that areas with better data will have better results. Well, what did you expect?
This isn't a case of bad data making bad results. If nobody was committing a crime, there would be no feedback loop reinforcing the patrols. It's as if the study authors (and lamer Slashdotters) believe that catching criminals is wrong if it's in the wrong area! But no, what did the study consider the problem to be?
the over-patrolling of communities of color
How is this a problem, when those areas still are the higher-crime areas, and when patrols shouldn't bother you if you're not committing a crime?
they don't arrest them until they bother folks well above their social class. They're not enforcing the law, they're enforcing a twisted social order. That's why poor folk don't take kindly to the cops. They're not there to serve and protect. They're there to oppress and contain.
And no I'm not a "Fuck the Police" type. They're a product of the system. If nobody calls out the system's bullshit nothing changes.
Hi! I make Firefox Plug-ins. Check 'em out @ https://addons.mozilla.org/en-US/firefox/addon/youtube-mp3-podcaster/