Slashdot Mirror
AI is Sending People To Jail -- and Getting it Wrong (technologyreview.com)
At the Data for Black Lives conference last weekend, technologists, legal experts, and community activists snapped the kind of impact AI has on our lives into perspective with a discussion of America's criminal justice system. There, an algorithm can determine the trajectory of your life.
From a report: The US imprisons more people than any other country in the world. At the end of 2016, nearly 2.2 million adults were being held in prisons or jails, and an additional 4.5 million were in other correctional facilities. Put another way, 1 in 38 adult Americans was under some form of correctional supervision. The nightmarishness of this situation is one of the few issues that unite politicians on both sides of the aisle.
Under immense pressure to reduce prison numbers without risking a rise in crime, courtrooms across the US have turned to automated tools in attempts to shuffle defendants through the legal system as efficiently and safely as possible. This is where the AI part of our story begins. Police departments use predictive algorithms to strategize about where to send their ranks. Law enforcement agencies use face recognition systems to help identify suspects. These practices have garnered well-deserved scrutiny for whether they in fact improve safety or simply perpetuate existing inequities.
Researchers and civil rights advocates, for example, have repeatedly demonstrated that face recognition systems can fail spectacularly, particularly for dark-skinned individuals -- even mistaking members of Congress for convicted criminals. But the most controversial tool by far comes after police have made an arrest. Say hello to criminal risk assessment algorithms.
Under immense pressure to reduce prison numbers without risking a rise in crime, courtrooms across the US have turned to automated tools in attempts to shuffle defendants through the legal system as efficiently and safely as possible. This is where the AI part of our story begins. Police departments use predictive algorithms to strategize about where to send their ranks. Law enforcement agencies use face recognition systems to help identify suspects. These practices have garnered well-deserved scrutiny for whether they in fact improve safety or simply perpetuate existing inequities.
Researchers and civil rights advocates, for example, have repeatedly demonstrated that face recognition systems can fail spectacularly, particularly for dark-skinned individuals -- even mistaking members of Congress for convicted criminals. But the most controversial tool by far comes after police have made an arrest. Say hello to criminal risk assessment algorithms.
People want a perfect world. People don't know history and how far humanity has come. People complain about modern life like we live in 1850. People should continue to work to fix things - but the constantly bitching , finger pointing and dividing of people into groups *does not help*.
Algorithms and bad statistics are not artificial intelligence. People using algorithms and bad statistics in idiotic ways is also not ai. Words mean things. Use them with care and precision.
The most blatant statistic that shows cultural racism is the crime clearance rate by race of the victim.
The computer should send many more cops into 'communities of color', not doing so is racist!
They aren't getting their 'fair share' of law enforcement, as seen by the fact that blacks are shot at a lower rate than their share of crime committed. Until 40%+ of those shot by cops are black, they aren't being treated fairly.
John McAfee 'It was like that time I hired that Bangkok prostitute; to do my taxes, while I fucked my accountant'
Because most risk assessment algorithms are proprietary, it’s also impossible to interrogate their decisions or hold them accountable.
You can't have secret laws, or secret government. Government and criminal justice must have the ability to scrutinize the decisions, and you can't scrutinize a secret algorithm.
This is the same thing that's happened with blood alcohol testing machines. Courts have ruled they have to allow scrutiny of the source code. The same should happen for these algorithms.
The next step is to simply decide which data is legal to make decisions on. You can't use race, but I bet you a lot of money they use where you live, and income levels. Discriminating on either of those isn't just a matter of racism, it's just not right to ding people because they're poor, or live in a poor neighborhood.
The data that gets used should be things more in peoples direct control. Like the crime committed, the number of other crimes you've been convicted of, etc.
The United States doesn't a a Justice system, but a punishment system.
It is running off the Old Idea. If we treat the population like pre-teen kids, where punishment is an effective way to curve behavior, and prevent this from happening.
Now lets not straw man this, and talk about murderers, and the harden criminals, where harsher sentence are needed.
Most Americans Jailed are for lower level crimes, crimes of passion, or crimes because they couldn't find an effective legal way out.
The cost of keep these people in jail, is often far more then their hindrance too society that they caused.
We can be tough on crime, without jailing everyone. Jailing should be used only if the criminal is considered too much of a risk to the general public. They are other ways to punish and rehabilitate criminals. Such as Home Confinement or Monitored Home Confinement, where the criminal can still go to work, and live their life, but just cannot travel anywhere he wants and when. Giving them a life, while making sure they don't go out of bounds. There is also just general relocation, sometimes the criminal causes crime, because they are living in a place that fosters such activity. Then there should be more effort in educations, and showing people a better way out.
People shouldn't be able to get away with criminal activity. But just locking them up isn't justice. It is just being cruel, and wanting revenge for their damage.
If something is so important that you feel the need to post it on the internet... It probably isn't that important.
All your insurance is based on statistical risk assessment too... is it all invalid? Should you be in the same risk pool as alcoholics and meth users?
I've abandoned my search for truth; now I'm just looking for some useful delusions.
Quoting the Slashdot story summary: "The US imprisons more people than any other country in the world."
It is more correct to say, "The US imprisons a higher percentage of its people than any other country in the world."
Prison is a big, profitable business in the United States. The companies that manage prisons are paid up to $70,000 per prisoner, per year.
Articles:
The Economics of the American Prison System (May 21, 2018)
The Prison Industry in the United States: Big Business or a New Form of Slavery? (Nov. 7, 2018)
When "blacks" stop doing culturally stupid shit like committing acts of violence then they wont be in jail. Stop judging them on their skin color and judge them based on their actions. And most importantly, don't give anyone a pass/handycap based on race; for that in of itself is racism!
Life is not for the lazy.
One important missing part of the story is how does the decision-making of algorithms fare against decision-making of humans?
Just like self-driving cars, the important thing for law enforcement AI isn't the absolute rate of errors in judgement, it's the relative rate of errors compared to humans making those decisions. Human decision making is far from perfect, so we shouldn't throw out algorithmic tools completely because they don't end up magically being correct 100% of the time. They just have to be at least 1% more consistent than we are to be of overall benefit. Remember, the goal here is to *reduce* the prison population through the use of AI. Less people will end up in prison due to the algorithms than would otherwise be there. Sure, it will make some mistakes, but overall, less people will be in prison compared to the human-judgement based system, because that is the metric the AIs are being trained to improve. If the prison population drops by 30% due to AI optimization, then that means a LOT less black people in prison, so even if the percentage error rate was a bit higher, less black people would be negatively impacted.
Of course they are leaving details out. Because the details about what happened wouldn't contribute any more toward their argument. By their own admission, GP did something worse than get "caught with a joint" and yet their punishment was much less severe than that hypothetical kid would have received for a much lesser infraction. I assume by the way they worded their post that GP was able to enter a plea deal, which is an opportunity that might be denied to someone who is determined by an algorithm to be more of a "risk" - and that's exactly the issue at hand.
Note that TFA uses language such as "defendant" and not "convicted criminal"; the problem is that one does not have to be convicted of a crime to have their life ruined. Actually, one doesn't even have to be charged with a crime. Imagine a scenario where someone is detained for a day or two then released without being charged with any crime, because an algorithm decides that they might be a risk, and in the meantime because they don't show up to work they lose their job. Whereas if another person, say someone charged with a crime, is let go within a very short timeframe on promise of making a phone call the next morning, or is allowed to post bail while awaiting trial, they might not suffer any major life interruption.