Page 1

CRIME & PUNISHMENT
— special report on “machine bias”
— homeless writer FDW re: prison

1

Page 2

An opportunity for opportunity:

SUBSCRIBE.

Your monthly subscription gives opportunity,
resources, and a monthly gift of free magazines
to a specific hard working street entrepreneur.

Find out more:
speakupmag.org/subscribe

Page 3

Speak Up is a voice
for the voiceless.

The magazine is sold by vulnerably-housed street
vendors who become micro-business owners,
develop a passion for inspired entrepreneurship,
and get back on their feet financially.

—

Speak Up Magazine is a 501(c)3 non-profit and is empowered
by your donations. Give at speakupmag.org.

Speak Up
501 Hawthorne Lane
Charlotte, NC 28204

speakupmag.org

3

Page 4

MACHINE BIAS
By Julia Angwin, Jeff Larson, Surya Mattu and Lauren Kirchner
This story was originally published in ProPublica

Page 5

On a spring afternoon in 2014, Brisha Borden for subsequently breaking into a warehouse
was running late to pick up her god-sister from and stealing thousands of dollars’ worth of
school when she spotted an unlocked kid’s electronics.
blue Huffy bicycle and a silver Razor scooter. Scores like this — known as risk assessments
Borden and a friend grabbed the bike and — are increasingly common in courtrooms
scooter and tried to ride them down the street across the nation. They are used to inform
in the Fort Lauderdale suburb of Coral Springs. decisions about who can be set free at every
Just as the 18-year-old girls were realizing stage of the criminal justice system, from
they were too big for the tiny conveyances assigning bond amounts — as is the case in
— which belonged to a 6-year-old boy — a Fort Lauderdale — to even more fundamental
woman came running after them saying, decisions about defendants’ freedom. In
“That’s my kid’s stuff.” Borden and her friend Arizona, Colorado, Delaware, Kentucky,
immediately dropped the bike and scooter Louisiana, Oklahoma, Virginia, Washington and
and walked away. Wisconsin, the results of such assessments are
But it was too late — a neighbor who given to judges during criminal sentencing.
witnessed the heist had already called the Rating a defendant’s risk of future crime is
police. Borden and her friend were arrested often done in conjunction with an evaluation of
and charged with burglary and petty theft for a defendant’s rehabilitation needs. The Justice
the items, which were valued at a total of $80. Department’s National Institute of Corrections
Compare their crime with a similar one: The now encourages the use of such combined
previous summer, 41-year-old Vernon Prater assessments at every stage of the criminal
was picked up for shoplifting $86.35 worth of justice process. And a landmark sentencing
tools from a nearby Home Depot store. reform bill currently pending in Congress
Prater was the more seasoned criminal. He would mandate the use of such assessments in
had already been convicted of armed robbery federal prisons.
and attempted armed robbery, for which In 2014, then U.S. Attorney General Eric
he served five years in prison, in addition to Holder warned that the risk scores might be
another armed robbery charge. Borden had injecting bias into the courts. He called for the
a record, too, but it was for misdemeanors U.S. Sentencing Commission to study their
committed when she was a juvenile. use. “Although these measures were crafted
Yet something odd happened when Borden with the best of intentions, I am concerned
and Prater were booked into jail: A computer that they inadvertently undermine our efforts
program spat out a score predicting the to ensure individualized and equal justice,”
likelihood of each committing a future crime. he said, adding, “they may exacerbate
Borden—who is black—was rated a high risk. unwarranted and unjust disparities that are
Prater—who is white—was rated a low risk. already far too common in our criminal justice
Two years later, we know the computer system and in our society.”
algorithm got it exactly backward. Borden The sentencing commission did not,
has not been charged with any new crimes. however, launch a study of risk scores. So
Prater is serving an eight-year prison term ProPublica did, as part of a larger examination
5

Page 6

of the powerful, largely hidden effect of algorithm was somewhat more accurate than a
algorithms in American life. coin flip. Of those deemed likely to re-offend,
We obtained the risk scores assigned to 61 percent were arrested for any subsequent
more than 7,000 people arrested in Broward crimes within two years.
County, Florida, in 2013 and 2014 and We also turned up significant racial
checked to see how many were charged disparities, just as Holder feared. In forecasting
with new crimes over the next two years, the who would re-offend, the algorithm made
same benchmark used by the creators of the mistakes with black and white defendants at
algorithm. roughly the same rate but in very different
The score proved remarkably unreliable in ways.
forecasting violent crime: Only 20 percent of The formula was particularly likely to falsely
the people predicted to commit violent crimes flag black defendants as future criminals,
actually went on to do so. wrongly labeling them this way at almost twice
When a full range of crimes were taken the rate as white defendants.
into account — including misdemeanors White defendants were mislabeled as low
such as driving with an expired license — the risk more often than black defendants.
6

Page 7

Could this disparity be explained by methodology and defended the accuracy of
defendants’ prior crimes or the type of crimes its test: “Northpointe does not agree that the
they were arrested for? No. We ran a statistical results of your analysis, or the claims being
test that isolated the effect of race from made based upon that analysis, are correct or
criminal history and recidivism, as well as from that they accurately reflect the outcomes from
defendants’ age and gender. Black defendants the application of the model.”
were still 77 percent more likely to be pegged Northpointe’s software is among the most
as at higher risk of committing a future violent widely used assessment tools in the country.
crime and 45 percent more likely to be The company does not publicly disclose the
predicted to commit a future crime of any kind. calculations used to arrive at defendants’
(Read our analysis.) risk scores, so it is not possible for either
The algorithm used to create the Florida defendants or the public to see what might be
risk scores is a product of a for-profit company, driving the disparity. (On Sunday, Northpointe
Northpointe. The company disputes our gave ProPublica the basics of its future-crime
analysis. formula — which includes factors such as
In a letter, it criticized ProPublica’s education levels, and whether a defendant has
7

Page 8

a job. It did not share the specific calculations, lawnmower and some tools. The prosecutor
which it said are proprietary.) recommended a year in county jail and follow-
Northpointe’s core product is a set of up supervision that could help Zilly with
scores derived from 137 questionsthat are “staying on the right path.” His lawyer agreed
either answered by defendants or pulled to a plea deal.
from criminal records. Race is not one of the But Judge James Babler had seen Zilly’s
questions. The survey asks defendants such scores. Northpointe’s software had rated Zilly
things as: “Was one of your parents ever sent as a high risk for future violent crime and a
to jail or prison?” “How many of your friends/ medium risk for general recidivism. “When I
acquaintances are taking drugs illegally?” look at the risk assessment,” Babler said in
and “How often did you get in fights while at court, “it is about as bad as it could be.”
school?” The questionnaire also asks people Then Babler overturned the plea deal that
to agree or disagree with statements such as had been agreed on by the prosecution and
“A hungry person has a right to steal” and “If defense and imposed two years in state prison
people make me angry or lose my temper, I and three years of supervision.
can be dangerous.”
The appeal of risk scores is obvious: The Criminologists have long tried to predict
United States locks up far more people than which criminals are more dangerous before
any other country, a disproportionate number deciding whether they should be released.
of them black. For more than two centuries, Race, nationality and skin color were often
the key decisions in the legal process, from used in making such predictions until about
pretrial release to sentencing to parole, have the 1970s, when it became politically
been in the hands of human beings guided by unacceptable, according to a survey of risk
their instincts and personal biases. assessment tools by Columbia University law
If computers could accurately predict which professor Bernard Harcourt.
defendants were likely to commit new crimes, In the 1980s, as a crime wave engulfed
the criminal justice system could be fairer and the nation, lawmakers made it much harder
more selective about who is incarcerated and for judges and parole boards to exercise
for how long. The trick, of course, is to make discretion in making such decisions. States
sure the computer gets it right. If it’s wrong in and the federal government began instituting
one direction, a dangerous criminal could go mandatory sentences and, in some cases,
free. If it’s wrong in another direction, it could abolished parole, making it less important to
result in someone unfairly receiving a harsher evaluate individual offenders.
sentence or waiting longer for parole than is But as states struggle to pay for swelling
appropriate. prison and jail populations, forecasting criminal
The first time Paul Zilly heard of his score risk has made a comeback.
— and realized how much was riding on it — Dozens of risk assessments are being used
was during his sentencing hearing on Feb. 15, across the nation — some created by for-
2013, in court in Barron County, Wisconsin. profit companies such as Northpointe and
Zilly had been convicted of stealing a push others by nonprofit organizations. (One tool
8

Page 9

being used in states including Kentucky and designed to provide judges with insight into
Arizona, called the Public Safety Assessment, the types of treatment that an individual might
was developed by the Laura and John need — from drug treatment to mental health
Arnold Foundation, which also is a funder of counseling.
ProPublica.) “What it tells the judge is that if I put you
There have been few independent studies on probation, I’m going to need to give you
of these criminal risk assessments. In 2013, a lot of services or you’re probably going
researchers Sarah Desmarais and Jay Singh to fail,” said Edward Latessa, a University of
examined 19 different risk methodologies used Cincinnati professor who is the author of a
in the United States and found that “in most risk assessment tool that is used in Ohio and
cases, validity had only been examined in one several other states.
or two studies” and that “frequently, those But being judged ineligible for alternative
investigations were completed by the same treatment — particularly during a sentencing
people who developed the instrument.” hearing — can translate into incarceration.
Their analysis of the research through 2012 Defendants rarely have an opportunity to
found that the tools “were moderate at best challenge their assessments. The results are
in terms of predictive validity,” Desmarais usually shared with the defendant’s attorney,
said in an interview. And she could not find but the calculations that transformed the
any substantial set of studies conducted in underlying data into a score are rarely
the United States that examined whether risk revealed.
scores were racially biased. “The data do not “Risk assessments should be impermissible
exist,” she said. unless both parties get to see all the data that
Since then, there have been some attempts go into them,” said Christopher Slobogin,
to explore racial disparities in risk scores. One director of the criminal justice program at
2016 study examined the validity of a risk Vanderbilt Law School. “It should be an open,
assessment tool, not Northpointe’s, used to full-court adversarial proceeding.”
make probation decisions for about 35,000 Proponents of risk scores argue they can
federal convicts. The researchers, Jennifer be used to reduce the rate of incarceration. In
Skeem at University of California, Berkeley, 2002, Virginia became one of the first states
and Christopher T. Lowenkamp from the to begin using a risk assessment tool in the
Administrative Office of the U.S. Courts, sentencing of nonviolent felony offenders
found that blacks did get a higher average statewide. In 2014, Virginia judges using the
score but concluded the differences were not tool sent nearly half of those defendants to
attributable to bias. alternatives to prison, according to a state
The increasing use of risk scores is sentencing commission report. Since 2005, the
controversial and has garnered media state’s prison population growth has slowed
coverage, including articles by the Associated to 5 percent from a rate of 31 percent the
Press, and the Marshall Project and previous decade.
FiveThirtyEight last year. In some jurisdictions, such as Napa County,
Most modern risk tools were originally California, the probation department uses
9

Page 10

risk assessments to suggest to the judge an prison in Massachusetts. But I guess they don’t
appropriate probation or treatment plan for count that here in Broward County.” In fact,
individuals being sentenced. Napa County criminal records from across the nation are
Superior Court Judge Mark Boessenecker said supposed to be included in risk assessments.
he finds the recommendations helpful. “We Less than a year later, he was charged with
have a dearth of good treatment programs, so two felony counts for shoplifting about $1,000
filling a slot in a program with someone who worth of tools from Home Depot. He said his
doesn’t need it is foolish,” he said. crimes were fueled by drug addiction and that
However, Boessenecker, who trains other he is now sober.
judges around the state in evidence-based
sentencing, cautions his colleagues that the Northpointe was founded in 1989 by Tim
score doesn’t necessarily reveal whether a Brennan, then a professor of statistics at the
person is dangerous or if they should go to University of Colorado, and Dave Wells, who
prison. was running a corrections program in Traverse
“A guy who has molested a small child City, Michigan.
every day for a year could still come out as Wells had built a prisoner classification
a low risk because he probably has a job,” system for his jail. “It was a beautiful piece of
Boessenecker said. “Meanwhile, a drunk guy work,” Brennan said in an interview conducted
will look high risk because he’s homeless. before ProPublica had completed its analysis.
These risk factors don’t tell you whether the Brennan and Wells shared a love for what
guy ought to go to prison or not; the risk Brennan called “quantitative taxonomy” —
factors tell you more about what the probation the measurement of personality traits such as
conditions ought to be.” intelligence, extroversion and introversion. The
two decided to build a risk assessment score
“I’m surprised [my risk score] is so for the corrections industry.
low. I spent five years in state prison in Brennan wanted to improve on a leading
Massachusetts.” risk assessment score, the LSI, or Level of
Sometimes, the scores make little sense Service Inventory, which had been developed
even to defendants. in Canada. “I found a fair amount of weakness
James Rivelli, a 54-year old Hollywood, in the LSI,” Brennan said. He wanted a tool
Florida, man, was arrested two years ago for that addressed the major theories about the
shoplifting seven boxes of Crest Whitestrips causes of crime.
from a CVS drugstore. Despite a criminal Brennan and Wells named their product the
record that included aggravated assault, Correctional Offender Management Profiling
multiple thefts and felony drug trafficking, the for Alternative Sanctions, or COMPAS. It
Northpointe algorithm classified him as being assesses not just risk but also nearly two dozen
at a low risk of reoffending. so-called “criminogenic needs” that relate
“I am surprised it is so low,” Rivelli said to the major theories of criminality, including
when told by a reporter he had been rated a 3 “criminal personality,” “social isolation,”
out of a possible 10. “I spent five years in state “substance abuse” and “residence/stability.”
10

Page 11

Defendants are ranked low, medium or high In 2011, Brennan and Wells sold
risk in each category. Northpointe to Toronto-based conglomerate
As often happens with risk assessment tools, Constellation Software for an undisclosed sum.
many jurisdictions have adopted Northpointe’s Wisconsin has been among the most
software before rigorously testing whether it eager and expansive users of Northpointe’s
works. New York State, for instance, started risk assessment tool in sentencing decisions.
using the tool to assess people on probation In 2012, the Wisconsin Department of
in a pilot project in 2001 and rolled it out to Corrections launched the use of the software
the rest of the state’s probation departments throughout the state. It is used at each step in
— except New York City — by 2010. The state the prison system, from sentencing to parole.
didn’t publish a comprehensive statistical In a 2012 presentation, corrections official
evaluation of the tool until 2012. The study Jared Hoy described the system as a “giant
of more than 16,000 probationers found the correctional pinball machine” in which
tool was 71 percent accurate, but it did not correctional officers could use the scores at
evaluate racial differences. every “decision point.”
A spokeswoman for the New York state Wisconsin has not yet completed a
division of criminal justice services said the statistical validation study of the tool and
study did not examine race because it only has not said when one might be released.
sought to test whether the tool had been State corrections officials declined repeated
properly calibrated to fit New York’s probation requests to comment for this article.
population. She also said judges in nearly Some Wisconsin counties use other risk
all New York counties are given defendants’ assessment tools at arrest to determine if a
Northpointe assessments during sentencing. defendant is too risky for pretrial release. Once
In 2009, Brennan and two colleagues a defendant is convicted of a felony anywhere
published a validation study that found that in the state, the Department of Corrections
Northpointe’s risk of recidivism score had an attaches Northpointe’s assessment to the
accuracy rate of 68 percent in a sample of confidential presentence report given to
2,328 people. Their study also found that the judges, according to Hoy’s presentation.
score was slightly less predictive for black In theory, judges are not supposed to give
men than white men — 67 percent versus 69 longer sentences to defendants with higher
percent. It did not examine racial disparities risk scores. Rather, they are supposed to
beyond that, including whether some groups use the tests primarily to determine which
were more likely to be wrongly labeled higher defendants are eligible for probation or
risk. treatment programs.
Brennan said it is difficult to construct a
score that doesn’t include items that can Overall, Northpointe’s assessment tool
be correlated with race — such as poverty, correctly predicts recidivism 61 percent of
joblessness and social marginalization. “If the time. But blacks are almost twice as likely
those are omitted from your risk assessment, as whites to be labeled a higher risk but not
accuracy goes down,” he said. actually re-offend. It makes the opposite
11

Page 12

mistake among whites: They are much more a stolen vehicle and fleeing from police, is
likely than blacks to be labeled lower risk but challenging the use of the score at sentencing
go on to commit other crimes. as a violation of his due process rights. The
But judges have cited scores in their state has defended Horne’s use of the score
sentencing decisions. In August 2013, Judge with the argument that judges can consider
Scott Horne in La Crosse County, Wisconsin, the score in addition to other factors. It has
declared that defendant Eric Loomis had also stopped including scores in presentencing
been “identified, through the COMPAS reports until the state Supreme Court decides
assessment, as an individual who is at high risk the case.
to the community.” The judge then imposed “The risk score alone should not determine
a sentence of eight years and six months in the sentence of an offender,” Wisconsin
prison. Assistant Attorney General Christine
Loomis, who was charged with driving Remington said last month during state
12

Page 13

Supreme Court arguments in the Loomis case. all the changes he was making in his life —
“We don’t want courts to say, this person his conversion to Christianity, his struggle to
in front of me is a 10 on COMPAS as far as quit using drugs and his efforts to be more
risk, and therefore I’m going to give him the available for his son. “Not that I’m innocent,
maximum sentence.” but I just believe people do change.”
That is almost exactly what happened to
Zilly, the 48-year-old construction worker sent Florida’s Broward County, where Brisha
to prison for stealing a push lawnmower and Borden stole the Huffy bike and was scored
some tools he intended to sell for parts. Zilly as high risk, does not use risk assessments
has long struggled with a meth habit. In 2012, in sentencing. “We don’t think the [risk
he had been working toward recovery with the assessment] factors have any bearing on
help of a Christian pastor when he relapsed a sentence,” said David Scharf, executive
and committed the thefts. director of community programs for the
After Zilly was scored as a high risk for Broward County Sheriff’s Office in Fort
violent recidivism and sent to prison, a public Lauderdale.
defender appealed the sentence and called Broward County has, however, adopted
the score’s creator, Brennan, as a witness. the score in pretrial hearings, in the hope
Brennan testified that he didn’t design his of addressing jail overcrowding. A court-
software to be used in sentencing. “I wanted appointed monitor has overseen Broward
to stay away from the courts,” Brennan said, County’s jails since 1994 as a result of the
explaining that his focus was on reducing settlement of a lawsuit brought by inmates in
crime rather than punishment. “But as time the 1970s. Even now, years later, the Broward
went on I started realizing that so many County jail system is often more than 85
decisions are made, you know, in the courts. percent full, Scharf said.
So I gradually softened on whether this could In 2008, the sheriff’s office decided that
be used in the courts or not.” instead of building another jail, it would begin
using Northpointe’s risk scores to help identify
“Not that I’m innocent, but I just believe which defendants were low risk enough to
people do change.” be released on bail pending trial. Since then,
Still, Brennan testified, “I don’t like the idea nearly everyone arrested in Broward has been
myself of COMPAS being the sole evidence scored soon after being booked. (People
that a decision would be based upon.” charged with murder and other capital crimes
After Brennan’s testimony, Judge Babler are not scored because they are not eligible
reduced Zilly’s sentence, from two years in for pretrial release.)
prison to 18 months. “Had I not had the The scores are provided to the judges who
COMPAS, I believe it would likely be that I decide which defendants can be released from
would have given one year, six months,” the jail. “My feeling is that if they don’t need them
judge said at an appeals hearing on Nov. 14, to be in jail, let’s get them out of there,” Scharf
2013. said.
Zilly said the score didn’t take into account Scharf said the county chose Northpointe’s
13

Page 14

software over other tools because it was easy friend Sade Jones, the teenage girls who
to use and produced “simple yet effective stole a kid’s bike and scooter, Hurley raised
charts and graphs for judicial review.” He said the bond amount for each girl from the
the system costs about $22,000 a year. recommended $0 to $1,000 each.
In 2010, researchers at Florida State Hurley said he has no recollection of the
University examined the use of Northpointe’s case and cannot recall if the scores influenced
system in Broward County over a 12-month his decision.
period and concluded that its predictive
accuracy was “equivalent” in assessing Sade Jones, who had never been arrested
defendants of different races. Like others, they before, was rated a medium risk.
did not examine whether different races were The girls spent two nights in jail before
classified differently as low or high risk. being released on bond.
Scharf said the county would review “We literally sat there and cried” the whole
ProPublica’s findings. “We’ll really look at them time they were in jail, Jones recalled. The
up close,” he said. girls were kept in the same cell. Otherwise,
Broward County Judge John Hurley, who Jones said, “I would have gone crazy.” Borden
oversees most of the pretrial release hearings, declined repeated requests to comment for
said the scores were helpful when he was a this article.
new judge, but now that he has experience he Jones, who had never been arrested before,
prefers to rely on his own judgment. “I haven’t was rated a medium risk. She completed
relied on COMPAS in a couple years,” he said. probation and got the felony burglary charge
Hurley said he relies on factors including reduced to misdemeanor trespassing, but she
a person’s prior criminal record, the type of has still struggled to find work.
crime committed, ties to the community, “I went to McDonald’s and a dollar
and their history of failing to appear at court store, and they all said no because of my
proceedings. background,” she said. “It’s all kind of difficult
ProPublica’s analysis reveals that higher and unnecessary.”
Northpointe scores are slightly correlated
with longer pretrial incarceration in Broward
County. But there are many reasons that could
be true other than judges being swayed by
the scores — people with higher risk scores
may also be poorer and have difficulty paying
bond, for example.
Most crimes are presented to the judge
with a recommended bond amount, but he
or she can adjust the amount. Hurley said he
often releases first-time or low-level offenders
without any bond at all.
However, in the case of Borden and her
14

Page 15

THE TRUTH ABOUT
PRISON (BY FDW)
Part 1 Part 2
Let me truthful with this story about prison. Let me tell you the truth about prison.Prison
When I was in prison somewhere in the United is not a place where anyone wants to go. The
States, I was raped by officers of the prison officers treat you like caged animals. You get
and inmates too. I told someone who worked your good ones and bad ones. The prison
at the prison, and they did nothing about it. I food is no good. The nurses are no good.
was so ashamed and hurt – my butt was sore They do not care about the inmates. They will
for days. They made me suck their meat a.k.a. let you die. I knew someone who was very sick,
dick, and then beat me. and he died on his bed. It took two hours for
I tried writing to people, like the news and the nurses to get to his cell after he died.
my parents, but they could not do anything at Now let me tell you about the bad officers.
the time. I then lost my mother. She died while They bring drugs to the inmates, and female
I was in prison, and they did not let me go officers have sex with male inmates. Then
to her funeral. I was so upset I wanted to kill they say that they were raped so they will not
myself. I talked with my doctor at the prison, lose their jobs. The truth is that the inmates
and he did not help me. He only drugged me did not rape the nurses or the female officers.
up more. The inmates then feel that they have to pay
I could not walk. I was walking like a real them money—$100, $200, sometimes $1,000.
zombie. They had me on a lot of mental illness I know this. This is the truth, and God knows
medications. These medications took my sex that this is the truth. Now the world knows the
drive away and my sun could not rise. All I truth.
could do was sleep. I remember that I was I’ve been raped in prison, and the officers
taking a shower, and an officer came into the did nothing about it. I tried to kill myself
shower and smacked me on my ass. ten times. I went to the hospital three times
He told me, “I’m going to hit that tonight.” and stayed seven days all three times. The
“No,” I said. sergeant, captain, lieutenant and doctor do
He raped me. I cried out for help and no not care about the inmates. They do not want
one helped me. That night, I tried to kill myself to give medicine to inmates. They will take
and was sent to the hospital. I stayed there money from an inmate ($20) for a sick call.
for seven days. I tried to tell the people at the The FBI needs to know the truth about state
hospital that I was raped, but the officer was in prisons.
the room with me, so I could not safely say a Trust me, I know.
word.
But now I’m not in prison, so I am telling it
all. This is the truth.
15