:o
Quote
Eddie Johnson, the Chicago Police Department superintendent. "We know we have a lot of violence in Chicago, but we also know there's a small segment that's driving this stuff," he said in a recent interview.
CHICAGO — In this city's urgent push to rein in gun and gang violence, the Police Department is keeping a list. Derived from a computer algorithm that assigns scores based on arrests, shootings, affiliations with gang members and other variables, the list aims to predict who is most likely to be shot soon or to shoot someone. Shaquon Thomas was on it.
His first arrest came at age 13, and others quickly followed, his face maturing in a progression of mug shots. By 18, Mr. Thomas, who was known as the rapper Young Pappy, was wounded in a shooting, the police say. Then last May, Mr. Thomas, 19, was fatally shot in what the police say was a running gang feud. His score was more than 500, putting him near the top of the Chicago Police Department's list.
"We know we have a lot of violence in Chicago, but we also know there's a small segment that's driving this stuff," Eddie Johnson, the police superintendent, said in a recent interview.
The authorities hope that knowing who is most likely to be involved in violence can bring them a step closer to curtailing it. They are warning those highest on the list that they are under intense scrutiny, while offering social services to those who want a path away from the bloodshed.
About three years into the program and on a fourth revision of the computer algorithm that generates the list, critics are raising pointed questions about potential breaches to civil liberties in the creation of such a ranking. And the list's efficacy remains in doubt as killings and shootings have continued to rise this year.
http://www.msn.com/en-us/news/crime/chicago-police-try-to-predict-who-may-shoot-or-be-shot/ar-BBtmmaY?ocid=ansmsnnews11
Eddie Johnson's statement was so important we had to put it in twice. Once before the byline.
Quote from: Razgovory on May 23, 2016, 09:09:39 PM
Eddie Johnson's statement was so important we had to put it in twice. Once before the byline.
I believe he is stereotyping.
One of these days you and Seedy need to do a cop-off.
Quote from: Razgovory on May 23, 2016, 09:22:27 PM
One of these days you and Seedy need to do a cop-off.
Has nothing to do with the above. To me, the system has a potential for incredible abuse, cloaked behind, "Well the algorithm says so."
and policy creep
That's all it is: profiling with a spreadsheet.
Quote from: 11B4V on May 23, 2016, 09:28:33 PM
Quote from: Razgovory on May 23, 2016, 09:22:27 PM
One of these days you and Seedy need to do a cop-off.
Has nothing to do with the above. To me, the system has a potential for incredible abuse, cloaked behind, "Well the algorithm says so."
and policy creep
It wasn't suppose to do with anything above. It was just an idea I found amusing. I don't know much about being a cop. I quickly discovered it was not the place for me.
That's like complaining about the government building anti-flood infrastructure only near rivers, as opposed to everywhere.
Quote from: Razgovory on May 23, 2016, 11:13:31 PM
Quote from: 11B4V on May 23, 2016, 09:28:33 PM
Quote from: Razgovory on May 23, 2016, 09:22:27 PM
One of these days you and Seedy need to do a cop-off.
Has nothing to do with the above. To me, the system has a potential for incredible abuse, cloaked behind, "Well the algorithm says so."
and policy creep
It wasn't suppose to do with anything above. It was just an idea I found amusing. I don't know much about being a cop. I quickly discovered it was not the place for me.
Ah, no offense intended.
Quote from: Martinus on May 23, 2016, 11:46:05 PM
That's like complaining about the government building anti-flood infrastructure only near rivers, as opposed to everywhere.
Don't follow
You put resources where they do the most good.
Cool, I bet it's a pretty interesting challenge to build these algorithms.
Quote from: Admiral Yi on May 24, 2016, 12:10:19 AM
You put resources where they do the most good.
Is this a principle you believe, in regards to law enforcement?
Quote from: Razgovory on May 24, 2016, 01:23:11 AM
Is this a principle you believe, in regards to law enforcement?
Can you just do the gotcha part without the question?
Quote from: Admiral Yi on May 24, 2016, 01:33:11 AM
Quote from: Razgovory on May 24, 2016, 01:23:11 AM
Is this a principle you believe, in regards to law enforcement?
Can you just do the gotcha part without the question?
:D
It sounds like law enforcement 101. Nobody is talking about putting people in jail preemptively, like in the Minority Report - but good intel should clearly be one of the biggest parts of both crime prevention and crime investigation.
This isn't about good intel; it's about Compstat and making the numbers fit. Anything that ports over to law enforcement the very same methodologies that are breaking everything else (Six Sigma, etc.) should be a clue.
Want good intel? Get the cops out of their cars, walk their beats, know their neighborhoods, learn the streets the old-fashioned way. Fuck Compstat. Goddamned Vietnam Whiz Kids bodycounting bullshit.
Computer models don't replace human judgment, they supplement it.
Quote from: CountDeMoney on May 24, 2016, 02:26:16 AM
This isn't about good intel; it's about Compstat and making the numbers fit. Anything that ports over to law enforcement the very same methodologies that are breaking everything else (Six Sigma, etc.) should be a clue.
Want good intel? Get the cops out of their cars, walk their beats, know their neighborhoods, learn the streets the old-fashioned way. Fuck Compstat. Goddamned Vietnam Whiz Kids bodycounting bullshit.
And get rid of them goddamned Glocks and go back to the humble, trusty .38 revolver, amirite??
Quote from: CountDeMoney on May 24, 2016, 02:26:16 AM
This isn't about good intel; it's about Compstat and making the numbers fit. Anything that ports over to law enforcement the very same methodologies that are breaking everything else (Six Sigma, etc.) should be a clue.
Want good intel? Get the cops out of their cars, walk their beats, know their neighborhoods, learn the streets the old-fashioned way. Fuck Compstat. Goddamned Vietnam Whiz Kids bodycounting bullshit.
Preach it, Brother Seedy. The naysayers just Don't Get It.
Check at figure 32 on this Presidential report.
https://www.whitehouse.gov/sites/default/files/page/files/20160423_cea_incarceration_criminal_justice.pdf
US has more than 150% more corrections officers than other countries. Same amount of prosecutors, a few less judges.
But 34.6% fewer police officers.
That's the reality of modern US law enforcement. Fight crime by grabbing any "risky" person and warehouse them in jail.
The GOP seems to think the problem is that we don't put enough people in prison. Cotton actually said that we have a "under-incarceration problem".
Quote from: Admiral Yi on May 24, 2016, 01:33:11 AM
Quote from: Razgovory on May 24, 2016, 01:23:11 AM
Is this a principle you believe, in regards to law enforcement?
Can you just do the gotcha part without the question?
Gotcha.
Quote from: Berkut on May 24, 2016, 12:21:42 PM
The GOP seems to think the problem is that we don't put enough people in prison. Cotton actually said that we have a "under-incarceration problem".
Do you dispute that people are prison insecure?
Quote from: DGuller on May 24, 2016, 08:16:42 AM
Computer models don't replace human judgment, they supplement it.
... in many cases computer models justify executives and senior management overriding the judgement of experienced people based on numbers they can make fit their own preconceptions.
Which I guess is just a longer rephrasing of what you just said :)
Quote from: Jacob on May 24, 2016, 04:48:12 PM
Quote from: DGuller on May 24, 2016, 08:16:42 AM
Computer models don't replace human judgment, they supplement it.
... in many cases computer models justify executives and senior management overriding the judgement of experienced people based on numbers they can make fit their own preconceptions.
Which I guess is just a longer rephrasing of what you just said :)
No, that's not really it at all.
Humans have certain advantages over algorithms. But they also have their cognitive biases and limitations on the amount of data they can process, so algorithms have certain advantages over humans. If you don't want to be a fossil who gets left behind in the age of algorithms, you need to recognize this fact and focus on learning how you can augment algorithms most effectively.
The algorithm can only be as good s the quality of the info going in and if the automation is being used as a justification for cutting the human budget you are in GIGO. Also how does when verify the model without running into problems of reverse causality? I.e. the algorithm targets certain at risk people, who are then subject to certain interventions which may in turn lead them to asocial responses.
Quote from: The Minsky Moment on May 24, 2016, 05:28:08 PM
The algorithm can only be as good s the quality of the info going in and if the automation is being used as a justification for cutting the human budget you are in GIGO.
The issue of data quality is both vastly underemphasized and vastly overemphasized at the same time. People who never built models generally don't appreciate just how crappy real life data is. That said, people who never built models also don't appreciate how models built on imperfect data can still be useful, just not
as useful. Which can still be a whole lot more useful than no model at all.
QuoteAlso how does when verify the model without running into problems of reverse causality? I.e. the algorithm targets certain at risk people, who are then subject to certain interventions which may in turn lead them to asocial responses.
You validate the model when you build it, not after deploying it. And you have causation issues any time you want to intervene in something or prevent something, so if you're concerned about that, crime fighting is not for you.
Quote from: DGuller on May 24, 2016, 05:05:00 PM
No, that's not really it at all.
Humans have certain advantages over algorithms. But they also have their cognitive biases and limitations on the amount of data they can process, so algorithms have certain advantages over humans. If you don't want to be a fossil who gets left behind in the age of algorithms, you need to recognize this fact and focus on learning how you can augment algorithms most effectively.
I think what I've been exposed to is people who talk the talk you just talked there, and who wielded spreadsheets and data to advance their arguments; but they have done so in areas where the proper application of algorithms is not well understood or where they themselves have no real understanding of the data they are wielding.
... cargo-cult and/or management fad applications of algorithms, as it were.
Quote from: Jacob on May 24, 2016, 06:35:51 PM
Quote from: DGuller on May 24, 2016, 05:05:00 PM
No, that's not really it at all.
Humans have certain advantages over algorithms. But they also have their cognitive biases and limitations on the amount of data they can process, so algorithms have certain advantages over humans. If you don't want to be a fossil who gets left behind in the age of algorithms, you need to recognize this fact and focus on learning how you can augment algorithms most effectively.
I think what I've been exposed to is people who talk the talk you just talked there, and who wielded spreadsheets and data to advance their arguments; but they have done so in areas where the proper application of algorithms is not well understood or where they themselves have no real understanding of the data they are wielding.
... cargo-cult and/or management fad applications of algorithms, as it were.
Yes, models can be dangerous in the hands of people with little knowledge. So can chainsaws, but that doesn't mean we should go back to using axes.
Quote from: Jacob on May 24, 2016, 06:35:51 PM
I think what I've been exposed to is people who talk the talk you just talked there, and who wielded spreadsheets and data to advance their arguments; but they have done so in areas where the proper application of algorithms is not well understood or where they themselves have no real understanding of the data they are wielding.
... cargo-cult and/or management fad applications of algorithms, as it were.
Yes. people who build models are frequently shockingly ignorant of the real-world constraints that limit their models, especially on the margins. I remember a briefing by representatives of [corporation name redacted] at one of the annual ASW strategic games my company hosted; they claimed that the SSN-21 class would have a 350-1 kill ratio against 4th generation Soviet subs (this was back when there was a Soviet Union) and that one
Seawolf had a 50% chance of sinking every Northern Fleet submarine in 90 days. They had neglected the fact that the Seawolf didn't carry the 200 or so torpedoes necessary for that task, as well as the fact that their kill ratios were simply unbelievable.
Quote from: derspiess on May 24, 2016, 08:28:10 AM
And get rid of them goddamned Glocks and go back to the humble, trusty .38 revolver, amirite??
Would've saved more than one locker room sink from the ignominy of an accidental discharge. :lol:
Quote from: Jacob on May 24, 2016, 06:35:51 PM
... cargo-cult and/or management fad applications of algorithms, as it were.
That's all it is: management fads. Oh boy oh boy oh boy, how can we import Agile project management into gang violence mitigation?
We had algorithms back in my day, you know. Analog 2D mainframes, mounted directly on the wall. The black pins were shootings, the red pins were homicides, the green pins were drug arrests, etc. We algorithmed the shift's patrols where the pins were all bunched together, all algorithmically and shit.
Quote from: CountDeMoney on May 24, 2016, 07:57:27 PM
Quote from: Jacob on May 24, 2016, 06:35:51 PM
... cargo-cult and/or management fad applications of algorithms, as it were.
That's all it is: management fads. Oh boy oh boy oh boy, how can we import Agile project management into gang violence mitigation?
We had algorithms back in my day, you know. Analog 2D mainframes, mounted directly on the wall. The black pins were shootings, the red pins were homicides, the green pins were drug arrests, etc. We algorithmed the shift's patrols where the pins were all bunched together, all algorithmically and shit.
Predictive modeling has made some strides in recent years. Quite a few, in fact.
Quote from: CountDeMoney on May 24, 2016, 07:57:27 PM
Quote from: Jacob on May 24, 2016, 06:35:51 PM
... cargo-cult and/or management fad applications of algorithms, as it were.
That's all it is: management fads. Oh boy oh boy oh boy, how can we import Agile project management into gang violence mitigation?
We had algorithms back in my day, you know. Analog 2D mainframes, mounted directly on the wall. The black pins were shootings, the red pins were homicides, the green pins were drug arrests, etc. We algorithmed the shift's patrols where the pins were all bunched together, all algorithmically and shit.
And there was probably some old timer bitching about how these new fangled pins and such was destroying real police work.
What I find amusing about this is that what you describe is in fact using data modelling and alogirthmic predictions to target high risk areas and individuals. The cop on the beat, learning his neighborhood, he is doing the *exact same thing* except he is doing so with a human being the data collection and analysis tool. And you know what? Humans are actually really fucking good at that. In fact, the human brains ability to put disparate data together isn't even well understood yet - it manages to put data together in ways that come to useful conclusions that are beyond our ability to rationally understand at times.
Of course, it is also rather good at putting data together in ways that lead to radically incorrect conclusions. See around a million easy to cite examples of police putting innocent people in jail based on jumping to conclusions about what they "know" to be true, that turns out to not be true at all, and fuck any evidence to the contrary.
Bitching about data modelling and non-human data analysis is idiotic. We cannot function without collecting data and interpreting it - hell, you could argue that that is the fundamental computational activity of all police work. No matter how you collect the data, and how you store the data, and how you analyze the data, if you do any of those steps poorly, you will lead to poor conclusions and poor outcomes.
Arguing that one particular method is inherently not to be trusted because it is new is just rank Ludditism. It is no different than people X years ago complaining that cops have no business being in cars, or having radios, or whatever.
A programme isn't perfect, but has the potential to be better than humans.
Quote from: DGuller on May 24, 2016, 05:45:45 PM
You validate the model when you build it, not after deploying it.
If the model is being used to generate policy responses in the real world, then validation requires correctly anticipating behavioral response across a broad domain. I.e. pre-validation will not be effective.
Quote from: The Minsky Moment on May 25, 2016, 10:52:51 AM
Quote from: DGuller on May 24, 2016, 05:45:45 PM
You validate the model when you build it, not after deploying it.
If the model is being used to generate policy responses in the real world, then validation requires correctly anticipating behavioral response across a broad domain. I.e. pre-validation will not be effective.
In that case, you go sit in a corner and sob about the futility of it all.
Quote from: Berkut on May 25, 2016, 10:14:19 AM
And there was probably some old timer bitching about how these new fangled pins and such was destroying real police work.
That's goddamned right, what with ruining perfectly good maps with holes like that.