Slashdot Mirror
Innocent Until Predicted Guilty
theodp writes "Gizmodo has an angry piece on IBM helping Florida to predict how delinquent your child's going to be. The Florida Department of Juvenile Justice has decided to start using IBM predictive analytics software to help them determine which of the 85,000 kids who enter their system each year poses the biggest future threat. From IBM's sales pitch: 'Predictive analytics gives government organizations worldwide a highly-sophisticated and intelligent source to create safer communities by identifying, predicting, responding to and preventing criminal activities. It gives the criminal justice system the ability to draw upon the wealth of data available to detect patterns, make reliable projections and then take the appropriate action in real time to combat crime and protect citizens.'"
your child doesn't fall into the minority report.
It seems to me that if the government thinks it can predict these things and takes certain actions in prevention, it might actually cause the problem that is predicted, and thus validate the method.
The best part is, if the software doesn't currently consider you a "threat" we can always tweak it to push you over the threshold! Remember that come next election, or next time you purchase something we don't think you should, or even the next time you pass us and don't give us a compliment!
Do predictive analytics work for other demographics as well, e.g. middle aged white man from prominent Ivy League university running an energy company more likely to steal billions of dollars over young Latino kid living in downtown Miami?
I believe the best use of this technology is as a means for monitoring our government officials and representatives (starting with the folks thinking about using it here.) It is arguable that the harm done by the average juvenile delinquent pales in comparison to the social and economic harm done by politicians and lawless officials. We should be using predictive technology keep them in check, and ensure that liberty is being preserved for future generations...
This isn't even thought crime. That at least requires you to think about doing something "subversive". This merely requires an arbitrary set of parameters to flag you as potentially "subversive". No thought required.
Sure you can, the entire insurance industry is based on doing just that.
There are 4 boxes to use in the defense of liberty: soap, ballot, jury, ammo. Use in that order. Starting now.
Okay, all I see here is a slippery slope argument. Juvenile delinquents who have been convicted of a crime are generally sentenced to probation, attendance in educational programs, counseling, etc. The only difference here is now they're using computer models to decide which programs are most appropriate for a given youth based upon the data they put in... instead of the court making the decision based upon a less complete set of data and a less methodical prediction of what would work best for that individual.
Now I'm not saying IBM's system works. It may or may not and that needs to be carefully studied. I have no problem, however, with computer models being used to determine which juvenile delinquents are most likely to benefit from specific programs and which are most in need of them when resources are limited. Appeals to various constitutional amendments are just empty rhetoric, given these kids have been convicted of a crime and this is part of their rehabilitation. In fact this whole article looks like an excuse for sensationalism and a reason to display cool graphics from "Minority Report". Lame Mr. Diaz.
Sure you can. You just can't be particularly sure that it'll be accurate.
Luckily, the people most likely to be (questionably accurately) judged to be pre-crime risks are likely to be members of more or less unsympathetic and disliked groups, so the people who actually count won't much care whether you are accurate or not. If anything, the "good upstanding citizens" will howl with rage and demand that they tolerance for false positives be increased every time the blood-spattered story of a false negative hits the cable news...
Sandra Bullock and Tom Cruise are doing a movie together? I'm not sure my gag reflex is strong enough for that.
How we know is more important than what we know.
Suppose those factors like increased absences and a couple of minor contacts with police indicate that Johnny is extremely likely to drop out of school. Maybe that's a good hint that someone needs to talk to Johnny and see if something correctable is going on in his life.
But that's not what's happening in our schools already; Just look at the Zero Tolerance statutes!
Do you really think that the same people who would expel a 9th grade girl for bringing a butter knife to school can be trusted to be rational with this kind of information?
Let's stop pretending that this is anything but a technological tool for doing what is already happening. Individuals are already differentially sentenced for all kinds of reasons, many of them terrible. Far better to use well-understood machine-learning/data-mining techniques instead of the discretion of individual judges and all its attendant biases.
N.B. This obviously has the potential for misuse (e.g. the first time some political hack suggests it is great for preemption.), but it is not a prima facie violation of individual's rights.
Eternal Vigilance, etc.
Do we really need a computer program to tell us that kids who are growing up in inner city neighborhoods surrounded by drugs and crime need more support?
I'm pretty sure we know that now, and we don't give them the support they need.
The fear with regard to this system is that it will lead to punishing first time offenders more harshly if they are high risk. The first time someone who set off a "high risk" trigger gets out of his first offense counciling and goes on to commit a violent crime the media and the public will be outraged that we didn't do more to prevent it (ie lock them up long term).
This software could be useful but it is not an answer to the social problem we have right now; That we don't give support to high risk minority kids and that we prefer to incarcerate them when possible.
"In America, first you get the sugar, then you get the power, then you get the women..." -H. Simpson
Did you mean, weak enough?
You know what? I can predict who will become involved in criminal activities. The answer to one simple question can peg it every time: ARE YOU HUMAN?
People are criminals. Republican, Democrat, bullshit. You ever speed? Criminal. You ever not come to a 100% complete stop before advancing past a stop sign? Criminal. Smoke weed? Criminal. Drink before the age of 21? Criminal.
Laws exist to make us criminals. We all are. We just get away with our crimes until we piss off the wrong people.
Yes we do because we need to find out the real statistical weight (significance) of which regressors gives us the best return on our money.
An excellent real world example is the disparity in blacks for sentencing. We find that it is more statstically significant that the reasons black comprise more of the prision population isn't because they are black per say, it just happens that blacks comprise more of the lower income. Income is more significant then just being black so if you want to reduce the disparity in sentencing, dollar for dollar you are better off spending money on improving job opportunities then say, sensitivity training for prosecutors.
The point of using computer systems is to dig through bias and perceptions and get to a root cause.
In the case of youth intervention it comes down to finding out which circumstances are the most relevant and addressable. Yeah we know of about 4000 regressors that factor into a child's success... it's finding out which of the 4000 we can address and which of those not only help the child, but save us money rather then just subsidize bad behaviors.
The fear is unfounded since it wouldn't be admissible in sentencing guidelines, no more then a genetic predisposition would be. The courts are pretty clear on what can be taken into account in sentencing and none of this would be allow to be factored in under existing laws and guidelines. Even if they tried it wouldn't make it past an appeal, (IANALBMWIAP)
-=[ Who Is John Galt? ]=-
Take a step back here, guys. This system is NOT being implemented for children entering school, or anything like that. It is being used for children who are actually entering the juvenile justice system. Kids who are already very much in trouble. These kids are already being pigeonholed by caseworkers, DAs, judges, probation officers, and a host of outsiders of the system such as school officials.
They are actually trying to use some actual data to try to direct these kids within the justice system. That's not such a terrible thing - in fact, it's what people in the system who are trying to do the right thing are trying to do - get kids the help they need to save their future. Like anybody, I would hope that the software is not blindly relied upon, but the people in the system are still going to be there. It's pretty hard right now to fight for a kid to get the attention they need and be directed the way they should in that system. Because these are kids who are already being judged by the law, there's not really any worse situation they can get into. They can already have a judge or caseworker who arbitrarily hates them, or who wants to help them. The factors for re-offending are already being examined, it's just that now some software can spit out a report based on these factors.
There's not much to see here.
Which is exactly what judges and parole officers do today using guesswork and Kentucky windage. I think it's hard to maintain that making that process more objective and automated is a bad thing.
Of course, there's potential for abuse through its misapplication to other areas.
I think the reason why this strikes a discordant note is that no one wants to be "labeled" by a machine no matter how sophisticated. Most, I think, want a human in the loop. Even then the mechanistic idea that humans can be categorized to predict future behavior seems so wrong even if it is necessary if we are to control social problems.
"Mrs. King, we have determined that your son Martin is statistically likely to grow up to be a troublemaker. We advise you begin Prozac treatments to curb these tendencies and allow him to live a normal life. Now please move to be back of the bus."