Serial Killers Should Fear This Algorithm
February 9, 2017 11:01 AM Subscribe
Building software to find killers Data analyst Thomas Hargrove believes each major city has a "few" serial killers.
Wow, it turns out Criminal Minds was right after all.
posted by Gelatin at 11:10 AM on February 9, 2017 [3 favorites]
posted by Gelatin at 11:10 AM on February 9, 2017 [3 favorites]
Soon to be an app for that...
posted by Alexandra Kitty at 11:13 AM on February 9, 2017 [2 favorites]
posted by Alexandra Kitty at 11:13 AM on February 9, 2017 [2 favorites]
A killer app even.
posted by mazola at 11:17 AM on February 9, 2017 [25 favorites]
posted by mazola at 11:17 AM on February 9, 2017 [25 favorites]
Should he ever be in the mood to aim this shit in the direction of some really high RPM fans, he could use historical data from, for example, the area served by the Ramparts division in LA to tune his algorithm to sift out unsolved murders committed by the police themselves.
posted by jamjam at 11:42 AM on February 9, 2017 [14 favorites]
posted by jamjam at 11:42 AM on February 9, 2017 [14 favorites]
Reminds me of the Center for Homicide Research in Minnesota, which got its start with investigating the circumstances of gay homicides, but has expanded into other areas.
posted by larrybob at 11:52 AM on February 9, 2017
posted by larrybob at 11:52 AM on February 9, 2017
I have recently wondered (as one does, obviously) if the apparent decrease in the number of serial killers operating in the US has more to do with how we gather and analyze crime data than any change in the reality of serial killing and killers. But Hannibal and Criminal Minds aside, serial killers are not much of an existential threat to Americans not engaged in sex work, drug consumption and/or trafficking, homelessness, etc. Consequently it seems that most police departments don't prioritize cases involving the 15th strangled prostitute in 3 years. See: the LISK.
As to why police aren't getting better at solving murders, I sort of instinctively assumed that it had something to do with changes in law and technology. In the 1940's you could literally call in a nurse to shoot up your suspect with adrenaline to keep him awake in the interrogation room. Nowadays you have pesky shit like DNA, video surveillance, and Miranda warnings to keep detectives from really getting in there and making the Reid technique work for them and their clearance rates. I'm not inclined to be up in arms over declining clearance rates until I know why they're declining; if we're just imprisoning fewer innocent people I'm actually fine with that.
posted by xyzzy at 12:19 PM on February 9, 2017 [10 favorites]
As to why police aren't getting better at solving murders, I sort of instinctively assumed that it had something to do with changes in law and technology. In the 1940's you could literally call in a nurse to shoot up your suspect with adrenaline to keep him awake in the interrogation room. Nowadays you have pesky shit like DNA, video surveillance, and Miranda warnings to keep detectives from really getting in there and making the Reid technique work for them and their clearance rates. I'm not inclined to be up in arms over declining clearance rates until I know why they're declining; if we're just imprisoning fewer innocent people I'm actually fine with that.
posted by xyzzy at 12:19 PM on February 9, 2017 [10 favorites]
It says something that in an age of big data where corporations can determine your consumer electronics preferences or political inclinations with startling accuracy, the science of data analysis in law enforcement is so undeveloped. Thomas Hargrove should be one of the pioneers of a new field and the police departments should be required to report everything to the FBI, so new versions of the algorithm can be made more accurate. Maybe they could even correlate other crimes like rape and animal abuse with the murder data, so they can find serial killers who are just getting started. Or maybe you could correlate the data with gang affiliations and use statistics to find the "hit men" in organized crime. The applications are endless.
posted by Kevin Street at 12:22 PM on February 9, 2017 [3 favorites]
posted by Kevin Street at 12:22 PM on February 9, 2017 [3 favorites]
Lucky for us, Dexter is still out there somewhere, chopping down trees in a sexy lumberjack beard.
posted by Saxon Kane at 12:31 PM on February 9, 2017 [5 favorites]
posted by Saxon Kane at 12:31 PM on February 9, 2017 [5 favorites]
I guarantee that Chicago has several serial killers. With a murder clearance rate in the high twenties to low thirties more than two thirds of all murders get a chance to kill again. Plus the severely constrained geographical concentration of the crime makes it all that much more likely that it is repeat offenders.
"Pulling up information from 218 metropolitan jurisdictions in the 2014 Uniform Crime Report, he found that in the places with poor clearance rates, the homicide rate was almost double that of places where the clearance rate was better—from 9.6 homicides to 17.9 per 100,000 people.
“It makes perfect sense,” Hargrove says. “If you leave the killers to walk the street, why wouldn’t that cause more killings? The answer is, it does.”"
His tools will be useless for this because the Chicago murders don't have differentiating features. It's mostly young men shooting young men and often gang related - criteria he explicitly excluded. It's not some sort of stylistic Dexter type thing. It's idiots with guns.
"I joke that what we’ve done is to create what amounts to a failed government detector.”
haha fuck.
posted by srboisvert at 12:39 PM on February 9, 2017 [2 favorites]
"Pulling up information from 218 metropolitan jurisdictions in the 2014 Uniform Crime Report, he found that in the places with poor clearance rates, the homicide rate was almost double that of places where the clearance rate was better—from 9.6 homicides to 17.9 per 100,000 people.
“It makes perfect sense,” Hargrove says. “If you leave the killers to walk the street, why wouldn’t that cause more killings? The answer is, it does.”"
His tools will be useless for this because the Chicago murders don't have differentiating features. It's mostly young men shooting young men and often gang related - criteria he explicitly excluded. It's not some sort of stylistic Dexter type thing. It's idiots with guns.
"I joke that what we’ve done is to create what amounts to a failed government detector.”
haha fuck.
posted by srboisvert at 12:39 PM on February 9, 2017 [2 favorites]
I agree that Hargrove's algorithm is just a first step, but with the right approach data analysis could help Chicago. If the police have any suspects at all, then some of those men are going to show up in the data as connected to multiple cases, which would make them targets for further investigation. Assuming there's any money to do the analysis or pay detectives for necessary overtime, that is.
posted by Kevin Street at 12:45 PM on February 9, 2017
posted by Kevin Street at 12:45 PM on February 9, 2017
I'm not inclined to be up in arms over declining clearance rates until I know why they're declining; if we're just imprisoning fewer innocent people I'm actually fine with that.
Yeah, the thing that most enraged me while reading this piece was conflating "clearance rates" with "murders solved".
posted by Etrigan at 12:45 PM on February 9, 2017 [7 favorites]
Yeah, the thing that most enraged me while reading this piece was conflating "clearance rates" with "murders solved".
posted by Etrigan at 12:45 PM on February 9, 2017 [7 favorites]
Another thing that strikes me as odd after reading this article is that there always seems to be lots of money for SWAT teams or bigger guns, but not enough to hire extra homicide detectives. Maybe it's easier to justify buying hardware than spending the money on skilled labor.
posted by Kevin Street at 12:50 PM on February 9, 2017 [4 favorites]
posted by Kevin Street at 12:50 PM on February 9, 2017 [4 favorites]
Another thing that strikes me as odd after reading this article is that there always seems to be lots of money for SWAT teams or bigger guns, but not enough to hire extra homicide detectives.
A lot of the equipment is paid for (partially or totally) by federal grants. Some of the military stuff even brings a requirement that if it's not used within a year or two, it has to be returned, which creates some pretty fucked-up incentives.
posted by Etrigan at 12:54 PM on February 9, 2017 [10 favorites]
A lot of the equipment is paid for (partially or totally) by federal grants. Some of the military stuff even brings a requirement that if it's not used within a year or two, it has to be returned, which creates some pretty fucked-up incentives.
posted by Etrigan at 12:54 PM on February 9, 2017 [10 favorites]
Clearance rate seems like a terrible KPI because it doesn't necessarily imply getting a killer off the street, and it's an easy number for crooked or just plain stupid cops to inflate. That is to say, I don't want cops making arrests unless it's a good quality arrest that leads to a good quality conviction.
Chicago has attempted "predictive policing," there's a lot of questions to be asked about it's application too, tending to affirm the sensibility in being cautious about the application of big data to criminology. Big data algorithms can discriminate and they do, Facebook provides the machinery to help -- though they're reigning it in a bit now. More often than not, big data is likely to show a pattern of bias in, bias out.
Previously 1, 2.
posted by Matt Oneiros at 1:03 PM on February 9, 2017 [7 favorites]
Chicago has attempted "predictive policing," there's a lot of questions to be asked about it's application too, tending to affirm the sensibility in being cautious about the application of big data to criminology. Big data algorithms can discriminate and they do, Facebook provides the machinery to help -- though they're reigning it in a bit now. More often than not, big data is likely to show a pattern of bias in, bias out.
Previously 1, 2.
posted by Matt Oneiros at 1:03 PM on February 9, 2017 [7 favorites]
Given data science's recent performance in the field of electoral politics, I hope you'll forgive me if I'm a bit leery of putting the awful weight of the criminal justice system behind it.
About the clearance rates that are in decline, I wonder what the criteria are for a case to be cleared? Is it measuring homicides versus indictments?
posted by whir at 4:21 PM on February 9, 2017 [1 favorite]
About the clearance rates that are in decline, I wonder what the criteria are for a case to be cleared? Is it measuring homicides versus indictments?
posted by whir at 4:21 PM on February 9, 2017 [1 favorite]
This would be really pretty good screenplay, with the totally obvious Act 2 thing that the algorithm is wrong and leading the cops down the wrong trail while the Cuckoo Clock Killer (tm) or whatever continues to plie his grisly trade.
I work with this stuff every day of my work life. No. It should not be trusted to administer justice.
I feel the same way with about it's tentative use in some branches of predictive medicine, outside of some very, very specific use cases that have been utterly well studied for decades.
This shit breaks every day of the year guys. And I'm talking about the incredibly expensive stuff from the big 5 vendors that costs ($X_M USD). You can plug it in backwards. The Scala team can do shit wrong. The Spark team can f_shit up. The data science group often does stuff wrong in R.
A lone data 'expert' tracking serial killers. No. I do not buy it, despite whatever he's selling to individual police departments at $350/hour.
posted by mrdaneri at 7:39 PM on February 9, 2017 [5 favorites]
I work with this stuff every day of my work life. No. It should not be trusted to administer justice.
I feel the same way with about it's tentative use in some branches of predictive medicine, outside of some very, very specific use cases that have been utterly well studied for decades.
This shit breaks every day of the year guys. And I'm talking about the incredibly expensive stuff from the big 5 vendors that costs ($X_M USD). You can plug it in backwards. The Scala team can do shit wrong. The Spark team can f_shit up. The data science group often does stuff wrong in R.
A lone data 'expert' tracking serial killers. No. I do not buy it, despite whatever he's selling to individual police departments at $350/hour.
posted by mrdaneri at 7:39 PM on February 9, 2017 [5 favorites]
According to A&E's series 'The Killing Season', there are well over 200 active serial killers in the US, and around 70% of their victims are sex workers. The series also says that many of them are over-the-road truckers, which makes seeing patterns even harder.
Serial killers have smartened up. They don't go after college students any more. They know that sex workers are still considered expendable.
posted by LindsayIrene at 7:47 PM on February 9, 2017
Serial killers have smartened up. They don't go after college students any more. They know that sex workers are still considered expendable.
posted by LindsayIrene at 7:47 PM on February 9, 2017
Yeah, this "clearance rates" metric seems like it would just create a lot of wrongful arrests or charges.
posted by AV at 8:14 PM on February 9, 2017
posted by AV at 8:14 PM on February 9, 2017
"It says something that in an age of big data where corporations can determine your consumer electronics preferences or political inclinations with startling accuracy, the science of data analysis in law enforcement is so undeveloped. Thomas Hargrove should be one of the pioneers of a new field and the police departments should be required to report everything to the FBI, so new versions of the algorithm can be made more accurate. Maybe they could even correlate other crimes like rape and animal abuse with the murder data, so they can find serial killers who are just getting started. Or maybe you could correlate the data with gang affiliations and use statistics to find the "hit men" in organized crime. The applications are endless."
In 1900, this breathless assessment would have described the promise of phrenology to detect the nascent criminal class.
The FBI estimate is that there are 25 to 50 serial killers in America at any one time, and that murders by serial killers are about 150 out of the 15,000 murders every year. That's about 1%, so arguments like the article makes about the overall clearance rate are unlikely to be a major source of unsolved crimes. The premise, that our inability to solve murders is because we're missing serial killers who act with impunity, has to be balanced against a null argument that the decrease in clearance rate is related to the overall decline of crime in general. That murder has been stubbornly persistent compared to the other general crime decrease is more likely tied to America's culture of gun violence, as overall assaults have decreased; guns are really effective ways to elevate non-lethal to lethal encounter, and many of our trackable reductions in the murder rate correspond to advances in battlefield medicine. That the relative rate of unsolved murders has gone up (or the clearance rate has gone down) would be consistent with a decrease in overall crime — the crimes most likely to get solved are the easiest to solve.
Because of that, an algorithm designed to detect unsolved cases that are the work of serial killers based on previously unsolved cases by documented serial killers is likely to have a significant false positive rate, and therefore (with limited police resources) may not help solve murders, and may actually hurt the police's ability to solve murders.
Likewise, the assertion that killers would kill again, which I think is fairly distinct from "serial killer" (we usually don't think of hitmen as serial killers), isn't necessarily implied by the decline in clearance rates — if someone isn't caught, why wouldn't that also encourage people who haven't killed to kill for the first time?
And that's before we get into the Goodhart's law problem: "When a measure becomes a target, it ceases to be a good measure." "Clearance rates" are necessarily a proxy for justice being served, but given the significant problem of wrongful conviction, the article's argument that an increased focus on clearances means that more crimes are solved becomes tautological.
The whole thing smacks of overfitting a model to find an outlier in a set, and encouraging others to do the same. Or: phrenology.
posted by klangklangston at 9:50 PM on February 9, 2017 [6 favorites]
In 1900, this breathless assessment would have described the promise of phrenology to detect the nascent criminal class.
The FBI estimate is that there are 25 to 50 serial killers in America at any one time, and that murders by serial killers are about 150 out of the 15,000 murders every year. That's about 1%, so arguments like the article makes about the overall clearance rate are unlikely to be a major source of unsolved crimes. The premise, that our inability to solve murders is because we're missing serial killers who act with impunity, has to be balanced against a null argument that the decrease in clearance rate is related to the overall decline of crime in general. That murder has been stubbornly persistent compared to the other general crime decrease is more likely tied to America's culture of gun violence, as overall assaults have decreased; guns are really effective ways to elevate non-lethal to lethal encounter, and many of our trackable reductions in the murder rate correspond to advances in battlefield medicine. That the relative rate of unsolved murders has gone up (or the clearance rate has gone down) would be consistent with a decrease in overall crime — the crimes most likely to get solved are the easiest to solve.
Because of that, an algorithm designed to detect unsolved cases that are the work of serial killers based on previously unsolved cases by documented serial killers is likely to have a significant false positive rate, and therefore (with limited police resources) may not help solve murders, and may actually hurt the police's ability to solve murders.
Likewise, the assertion that killers would kill again, which I think is fairly distinct from "serial killer" (we usually don't think of hitmen as serial killers), isn't necessarily implied by the decline in clearance rates — if someone isn't caught, why wouldn't that also encourage people who haven't killed to kill for the first time?
And that's before we get into the Goodhart's law problem: "When a measure becomes a target, it ceases to be a good measure." "Clearance rates" are necessarily a proxy for justice being served, but given the significant problem of wrongful conviction, the article's argument that an increased focus on clearances means that more crimes are solved becomes tautological.
The whole thing smacks of overfitting a model to find an outlier in a set, and encouraging others to do the same. Or: phrenology.
posted by klangklangston at 9:50 PM on February 9, 2017 [6 favorites]
That murder has been stubbornly persistent compared to the other general crime decrease is more likely tied to America's culture of gun violence
It's because there's bodies left over. You can play classification games with most other crimes, it's all bargain bin cooked books type stuff, but bodies are harder to hide.
posted by Strange_Robinson at 11:10 PM on February 9, 2017 [3 favorites]
It's because there's bodies left over. You can play classification games with most other crimes, it's all bargain bin cooked books type stuff, but bodies are harder to hide.
posted by Strange_Robinson at 11:10 PM on February 9, 2017 [3 favorites]
Well, but the overall decline in crime isn't just numbers juking. It's too widespread, prolonged and significant. It's happened in other countries too, and one of those distinctly American things is that we have more fatalities in large part because we have more gun-related violence, and guns are simply more lethal than most other personal weapons. If you're stabbed or beaten, you're more likely to survive than if you've been shot. But yeah, having bodies does make it one of the most reliable measurements of crime rates overall.
posted by klangklangston at 12:25 AM on February 10, 2017 [2 favorites]
posted by klangklangston at 12:25 AM on February 10, 2017 [2 favorites]
Somehow, the free market has not yet been able to solve the serial-killer problem.
posted by acb at 3:34 AM on February 10, 2017 [2 favorites]
posted by acb at 3:34 AM on February 10, 2017 [2 favorites]
there are well over 200 active serial killers in the US, and around 70% of their victims are sex workers. The series also says that many of them are over-the-road truckers, which makes seeing patterns even harder.
So the arrival of self-driving trucks should put a dent in the serial-killing rate, or at least coop all the serial killers in regional pools where they can be easily picked off with statistical data-mining?
posted by acb at 3:37 AM on February 10, 2017 [3 favorites]
So the arrival of self-driving trucks should put a dent in the serial-killing rate, or at least coop all the serial killers in regional pools where they can be easily picked off with statistical data-mining?
posted by acb at 3:37 AM on February 10, 2017 [3 favorites]
There's bound to be serious problems with most crime stats. The impossibility of perfectly aligned incentives pretty much guarantees any system can be gamed, and in short order. The theory that removal of lead somehow generated the difference seems so simple as to be absurd. All I've left is the weird notion that we're somehow better people than those from 50 years ago. Another possibility, we've reduced liberties to such an extent that folks prone to pushing against boundaries are being jailed? In terms of a 'repression' index, I wonder sometimes how this era compares to others.
We're awash in data that has been force fit to expectations. I'd not place much faith in crime data.
posted by Strange_Robinson at 11:04 AM on February 10, 2017 [1 favorite]
We're awash in data that has been force fit to expectations. I'd not place much faith in crime data.
posted by Strange_Robinson at 11:04 AM on February 10, 2017 [1 favorite]
Well, we are much nicer to our children, we no longer think a good beating is an essential part of parental discipline, and we don't routinely leave babies to cry. And children are taken into care if it is deemed that their parents are damaging them. At least in Western Europe. So many less people are going to grow up utterly emotionally fucked, maybe even enough to account for the decrease in crime rates.
posted by glasseyes at 6:27 PM on February 10, 2017
posted by glasseyes at 6:27 PM on February 10, 2017
« Older “The key is, you have to love them like your own." | Party on Wayne, party on Garth Newer »
This thread has been archived and is closed to new comments
posted by grumpybear69 at 11:06 AM on February 9, 2017 [2 favorites]