Slashdot Mirror
Wisconsin's Prison-Sentencing Algorithm Challenged in Court (engadget.com)
"Do you want a computer to help decide a convict's fate?" asks Engadget, telling the story of a Wisconsin convict who "claims that the justice system relied too heavily on its COMPAS algorithm to determine the likelihood of repeat offenses and sentenced him to six years in prison." Sentencing algorithms have apparently been in use for 10 years.
His attorneys claim that the code is "full of holes," including secret criteria and generic decisions that aren't as individually tailored as they have to be. For instance, they'll skew predictions based on your gender or age -- how does that reflect the actual offender...?
[T]he court challenge could force Wisconsin and other states to think about the weight they give to algorithms. While they do hold the promise of both preventing repeat offenses and avoiding excessive sentences for low-threat criminals, the American Civil Liberties Union is worried that they can amplify biases or make mistakes based on imperfect law enforcement data.
The biggest issue seems to be a lack of transparency, which makes it impossible to determine whether convicts actually are receiving fair sentences.
[T]he court challenge could force Wisconsin and other states to think about the weight they give to algorithms. While they do hold the promise of both preventing repeat offenses and avoiding excessive sentences for low-threat criminals, the American Civil Liberties Union is worried that they can amplify biases or make mistakes based on imperfect law enforcement data.
The biggest issue seems to be a lack of transparency, which makes it impossible to determine whether convicts actually are receiving fair sentences.
If you think a computer is biased and unfair and makes decisions using secret criteria, wait till you meet a human!
Don't waste your vote! Vote for whoever you want, unless you live in a swing state it won't matter anyways
And this is different how to inscrutable judges who decide sentencing? Case in point Judge Aaron Persky and the Stanford rape case. Was there any openness or fairness in that sentence? I don't see Brock Turner complaining much.
I am Slashdot. Are you Slashdot as well?
In the US there are already way too many people in jail, letting buggy software sentence people is a joke.
use algorithms, if we believe we can generalize and use statistics and averages, thinking that people are more or less robots, obeying instincts and influence of environment and society, without free will.
on the other hand, if we believe that each individual is different and has free will, there cannot be anything but superficial use for generalizations and averages, and certainly no sentencing algorithms.
of course judges etc, who will sentence without algorithms can make mistakes, but that is the price of free will. right to make mistakes, even right to be unjust and be evil( as well as good and just). but it may be better than be considered mere robots.
Wait, it's iCurrentYear, why would we be giving any weight to gender in sentencing? Men and women have equal rights, yes? So why should say, someone who raped a teenager over 50 times get off scott free just because of their gender?
... and I thought Facebook was fucked up.
It little behooves the best of us to comment on the rest of us.
Repeat offender = Longer prison term. The ACLU can go F themselves.
Hmmm.. I reckon the algos started around 1980.
https://commons.wikimedia.org/...
46137
It is a negative, self-reinforcing feedback loop.
While the idea of using an algorithm to sentence a human being is bone-chilling, you might be able to justify this as a "formula" for sentencing -- which, of course, merits its own debate.
What is unconscionable about this is the fact that it's a SECRET algorithm. As in closed source. Essentially a secret law.
This has no place in democracy.
(Also, any algorithm which ingests statistical and demographic data is bound to come up with unpalatable and/or spurious demographic correlations (since there is a causal link between poverty and crime and a historic link between race and poverty) which I wold rather have society refrain from codifying -- in law or in actual computer code).
It doesn't and it's not intended to.
It's statistics and probabilities, just like how insurance premiums work.
Confucius say, "Find worm in apple - bad. Find half a worm - worse."
Why don't they use the fundamentals used in developing the algorithm to identify and eliminate the root causes of crime (through rehabilitation/counseling) and thus reduce crime in highly prone areas? If using the software they show that they can reduce crime in, say Chicago, to 50% in 5 years, the researchers would have accomplished the 'real' thing.
The courts and police are moving in a heavily wrong direction with the software.
A human can be assessed for bias, while a program must produce results (in this case jail sentences) before it can be judged. Problem is that the program hides the source code, and has a black box compiled state. Both must not only be known but be publicly disclosed before any faith can be put in the sentencing.
The link supplied goes to a page with barely more text than a slashdot summary. Skip the middleman and go to the actual source.
Quattuor res in hoc mundo sanctae sunt: libri, liberi, libertas et liberalitas.
Longer sentence increase the chance you re-offend, BUT often male are for a similarly violent offense given historically a longer or harsher sentence than female. So if the software use historical data to determine male should get a longer sentence , then it is a self fulfilling prediction.
An algorithm should NEVER give a sentence length based on recidivism rate, it should solely based on the crime itself, and past recidivism. Giving longer sentence based on other criteria like probability f recidivism is pre-crime sentencing and utterly disgusting.
C. Sagan : A demon haunted world:
http://www.amazon.com/gp/product/0345409469/
visit randi.org
His attorneys claim that the code is "full of holes," including secret criteria and generic decisions that aren't as individually tailored as they have to be.
This statement was kinda funny . Unless one of his attorney is named Alan Cox and has a merlins beard ( which is highly unlikely).
Why are there so many stores about prisons on slashdot? Is it simply the area of the country I live in? Is the readership here more inclined to be law breakers? I'm by no means rich or privileged or anything and am from a poorer immigrant background yet nobody I have ever known has been arrested or put in jail with the one exception of a neighbor who was making meth in his bathroom. Are there no more tech stories or educational articles? The few that I've submitted are ignored. Is tech dead? Is that a valid ask slashdot submission?
Sentencing based on crimes you are expected to commit in future? WTF? Are we talking about law or socioeconomics?
What is this? Minus three strikes, and you're in?
Transparency can be good or bad.
The algorithm is likely to do a better job of actual justice than the human judge, but that doesn't mean much to a lawyer. The job of the defense lawyer is to get their client off even if the client is guilty. There is no actual justice because they want to believe that if it is possible to get the client off, then the real guilt must not exist.
These are folks for whom the English language is very much perceived like playdough. The word "is" can be challenged or redefined with cavalier indifference to the reality that there is an actual reality that is. Science is regularly treated as if it was 13th century voodoo because it is a successful attack even when the science is valid.
They will use the grammar used to describe the mathematical fundamentals to try and invalidate the algorithm, not because there is any flaw in the algorithm or the fundamentals, but because they can attack something written in words. If they can be successful in any such, then "of course", they will assume to have been in the right and to have saved the world from a fallacy of mathematical truth, and will feel no remorse for having thrown away a tool of justice substantially as strong as the count of people on a jury required to get justice.
Whats in the algorithim: if $offender = poor { $sentence = $sentence + 5 } if $offender = homeless { $sentence = $sentence + 10 } if $offender = latino { $sentence = $sentence * 2 } elseif $offender = black { $sentence = $sentence * 3 }
Since when is it a problem to solve a social issue with a technical solution? Oh, right. Since forever.
Don't fight for your country, if your country does not fight for you.
https://www.youtube.com/watch?v=DXDd1zO9F1I&list=PLhavZM1-Ey4H9NY2rxpsbaCQ8dDLNNs6k
at 27:29.. the "Judgement Machine" weighs the evidence, with some charming FX typical to the budget and the era ;-)
Obviously, secret rules and presets made sure that Blake, a freedom fighter and threat to the powers that be, is locked up forever. Oh and those powers call themselves "The Federation" :_)
-f
Look, I get that driverless cars are all the rage these days, and nonprogrammers like you think that computers are some kind of intelligent living beings, but it's really not so at all.
A computer program is an inflexible, dumb piece of logic that is incomplete and crashes all the time when unexpected inputs are introduced. Read Kafka if you want to know what a world directed by computers is like.
Lack of transparency? Transparency another word that means so much it has lost any meaning other than forcing one to dance to the tune of the piper.
Sounds like the same algorithms that are used to calculate your insurance premium...
http://www.jsonline.com/news/c...
They rely on cookies to limit the number of articles that you read. Unlike similar sites, they don't block you if you block cookies.
The problem is the algorithms were optimized for a specific set of people, and guess what, they weren't the most common offenders.
That is, the algorithms are supposed to predict what you will do after you get out - go legit or commit more crimes.
But they are very accurate for older, white offenders, but very inaccurate for younger, black offenders. Effectively the algorithms were written in an attempt to be race blind by ignoring race. But certain activities, like number of previous interactions with the police, contain built in bias against younger, black people.
While a 40 year old white man that's been questioned by the police 5 times is likely a habitual criminal, a 20 year old black man that been questioned 10 times was just walking while black.
excitingthingstodo.blogspot.com
For instance, they'll skew predictions based on your gender or age -- how does that reflect the actual offender...?
Insurance companies use the same statistics to charge people for insurance. New drivers, regardless of their race, are more likely to be involved in accidents than more experienced drivers.
However, once a person reaches a certain age their experience is outpaced by their declining health (poor eyesight, slower reflexes, impaired mental conditions, etc) at which point insurance rates rise.
I have never had an accident in my decades of driving yet the cost of my insurance is based on the totality of everyone in my age group. How is that fair?
All of the above is the same thing when it comes to criminals. Certain age groups have a propensity for committing crimes and gender plays a role as well. Having people charged based on these statistics is exactly the same as being charged for your insurance. If you're going to claim someone should be treated based on their crime rather than what others in their group have done then the same should apply to insurance.
We will bankrupt ourselves in the vain search for absolute security. -- Dwight D. Eisenhower
"claims that the justice system relied too heavily on its COMPAS algorithm to determine the likelihood of repeat offenses and sentenced him to six years in prison."
So a defendent's past isn't supposed to be used in trial, but some unknown factor about what he MIGHT do in the future is allowed in sentencing? I'm all for executing criminals, but this sort of nonsense is just plain dangerous.
Yes, I'm saying credit score is a point for consideration in sentencing - not decision of guilt or innocence, but once guilt has been established, credit score tells about a person's history of making good on commitments, and should be a strong predictor of their likelihood of meeting terms of a suspended sentence, probation or parole.
And pray tell what good is that for the double digit percentage of the population who have bad credit scores scores for reasons beyond their control? People who are young, very old, poor, disabled, unhealthy or have just had a bad run of luck can have a terrible credit score and it has no correlation with anything that should be a consideration in sentencing. Not everybody is a 45 year old white male with a college degree and a steady job and 2.5 kids. Credit scores are complicated and there are many reasons why someone might have a bad credit score. Furthermore there are plenty of criminals who have pristine credit records. I cannot see any reasonable argument for criminal sentencing where credit scores should play any role at all. All that does is favor rich privileged people who probably don't really need the help.
Eventually they could always use the Sybil System (http://psychopass.wikia.com/wiki/Sibyl_System)
Yes, I'm saying credit score is a point for consideration in sentencing - not decision of guilt or innocence, but once guilt has been established, credit score tells about a person's history of making good on commitments, and should be a strong predictor of their likelihood of meeting terms of a suspended sentence, probation or parole.
Wish I had mod points. Greatr point, that, and I wish I'd thought of it....
Why? Do you have scientific evidence that there is any meaningful strong correlation between credit scores and recidivism? I'm guessing you are just assuming there is a link without actually having data to back it up. How do you plan to account for people who have terrible credit scores for reasons beyond their control (age, poor health, bad luck, etc)?
"Do you want a computer to help decide a convict's fate?" asks Engadget...
Why not, *if* the algorithm is reasonably crafted? I think the controversy is that the algorithm in question took the probability of repeat offending into consideration. However, if certain metrics can be calculated, particularly with respect to culpability, victim impact, remorse, etc, then why shouldn't an algorithm be used to determine the optimum sentence?
While I'll be the first to state that I don't know what the standard would look like for such variables, they are all things that can be measured, modeled, analyzed and applied. By integrating an algorithm for optimal sentencing I think we could get rid of a ton of waste in the penal system incarcerating people who may not necessarily need to be incarcerated any longer (and do away with ridiculous minimum-sentencing garbage).
As others have said, the humanitarian side of the system errs on the side of forgiveness. If someone's credit is tanked, then don't consider their credit rating as a positive indicator of their compatibility with societies rules and expectations, look for other positive signs. This isn't saying "bad credit score: throw them under the jail and let them rot" this is saying "good credit score: indication that this person makes good on their promises, on a first offense give them benefit of the doubt and simple probation"
Unless you can show me some causal relationship between credit score and recidivism it should not be a consideration unless the case is about something related to personal finance. Even using it only as a positive has a skewing effect. Minorities, poor, elderly, young all are more likely to have poor credit ratings. So you are not helping the most disadvantaged groups but you are helping white, middle/upper class privileged people based on "evidence" that is unrelated to the case. It's no different than saying "he looks like a nice young man so we shouldn't punish him too hard". It makes zero sense.
How would the use (or alleged use) of this algorithm affect cases where judges want to "Make an example" of someone and basically throw the book at them for a first offense or for not taking a plea deal? I am talking cases like the Aaron Schwartz case, where he was trying to free up educational and research information for public use (as in the greater good) and ended up with basically 30 years in prison for what was essentially a financially based copyright case (Though it was not even clear what specific copyright was being violated.)
Would the algorithm suggest saner consequences in Aaron Schwartz's case? I don't know if there is any good way to find out without committing a "crime" like that.
Y'all ever been to Texas?
Not only have I been there I have family that lives there. Outside of Austin they should ask for a passport for most of the state to go there. I make no apologies when I say many Texans have some seriously messed up ideas about what constitutes "justice".
Algorithmic sentencing is one way to move toward a more consistent system, less subject to the vagaries of individual jurisdictions.
Consistency isn't necessarily as valuable as you might be implying. The entire point of having a judge is, you know, to judge things and come to a reasoned opinion about how the law should apply to a particular case. I could see an algorithm being useful as an aid to advise a judge on possible options but there are WAY too many corner cases for it to be a good idea to put it front and center. Are judges imperfect? Yep! That's why we have appeals systems. But you literally cannot come up with an algorithm that will properly address all the corner cases. You are merely turning the programmer and lawmakers into the de-facto judge which is a terrible idea. See three strikes laws if you need an example of how stupid algorithmic sentencing can be.
Judgements still have to be made based on something, and credit scores (and, more specifically, the underlying data from which they are computed) are one of the strongest windows into personality and prediction of future behavior we have in today's society.
I reject your framing of the issue. You are presuming that credit scores have any meaningful relationship to criminality without presenting any actual evidence that such an assertion is backed by facts. You are extrapolating purchasing and financial management behavior to have some relationship to criminality without any basis. Even a correlation isn't adequate because there are all sorts of ridiculous correlations between completely unrelated things. You have to PROVE a causal relationship between an individual person's credit score and their likelihood to commit future crimes for your argument to have any basis at all. Good luck with that.
I sentence you to 2d20 years in prison. The dice are tested to be fair.
If I have been able to see further than others, it is because I bought a pair of binoculars.
Sentencing scoresheets are already extremely widespread at the state and federal level. These scoresheets ingest data about the current crime, prior crimes, victim harm and assign numerical values to each of these things. This is then crunched to produce a range of prison sentences in months.
For example, in Florida, someone with a clean record who commits grand theft wouldn't score any prison at all by default. A person with 12 prior residential burglaries might score 10 years in prison for the same crime (ie, more than the statutory maximum for grand theft). Then you add in enhancements for recently being released for prison, enhancements for using a gun, enhancements for having multiple prior felonies, etc.
Judges and prosecutors have tons of discretion to deviate from these guidelines when offering deals or sentencing after trials.
But the point is that "algorithms" are commonly used in sentencing and have been for years. They're not secret and they don't use demographic data.
Do you want uniformity or individually tailored sentences for criminals? You can't have both.
Comment removed based on user account deletion
Comment removed based on user account deletion