CNN Business  — 

Cash bail keeps many people behind bars before trial simply because they can’t afford to pay to get out. Can algorithms help do a better job?

That’s the question California voters are being asked November 3 as they decide whether to vote for or against Proposition 25.

If voters pass the measure, which would in turn uphold 2018’s Senate Bill 10, California will become the first state to wholly abolish pretrial cash bail. Instead, judges must consider the results of a risk-assessment tool – software that uses algorithms to predict whether a person is likely to show up for court or get arrested again in the meantime – in determining whether to hold or release a person prior to their court dates.

On its face, Prop 25 sounds like it would appeal to proponents of criminal justice reform. The California Democratic Party, Democratic Gov. Gavin Newsom and congressional leaders such as Rep. Karen Bass have expressed broad support for the measure.

And recent research from UC Berkeley’s California Policy Lab and the nonpartisan Public Policy Institute of California forecasts that if Prop 25 passes more people in the state will be released prior to arraignment.

As of 2019, 49 of the state’s 58 counties may have used pretrial risk assessment tools, according to PPIC. But despite that usage, it’s hard to discern how well they can forecast defendants’ future behavior and critics worry these tools will simply perpetuate racial inequality in policing due to the data they rely on – data that may include a defendant’s age and arrest history. The use of these algorithm-driven computer programs is so controversial and thorny that some former supporters have switched sides, coming out against risk assessments.

“It’s complicated, especially if it comes as a way to end cash bail or comes as part of bail reform – it’s hard to stand in the way of it and say this is not a good idea,” said Electronic Privacy Information Center fellow Ben Winters, highlighting the difficulty of the decision voters face.

In need of reform

Proponents of Prop 25 say the measure would provide much-needed reform to the state’s criminal justice system.

Assemblymember Rob Bonta, who co-authored Senate Bill 10, said the proposition “will finally end a for-profit, private, predatory system that punishes poor people simply for being poor” and will replace it with an “objective” assessment tool.

Opponents – on both sides of the political aisle – agree that cash bail is a flawed system that disproportionately punishes low-income offenders, but they believe Proposition 25 is not a solution.

The California Republican Party, in its official endorsement guide, said the proposition forces Californians accused of crimes to rely on the algorithms, which would free more people accused of crimes. The bail industry has raised millions to fight the proposition’s passage.

Other opponents, including the ACLU, Human Rights Watch, and nonprofit coalition JusticeLA, worry that the risk assessment tools rely on racially biased information like demographic, criminal history and socioeconomic status as indicators of recidivism. They argue that California would be building one deeply flawed system on top of another

Unclear effectiveness

If Prop 25 succeeds, it will require that defendants arrested for nonviolent misdemeanors, which covers the majority of jail bookings, be released within 12 hours of being booked. Other defendants will be subject to a risk assessment that considers factors such as their age, criminal history, and whether they’ve missed court dates in the past, to sort people into low-, medium-, and high-risk categories. Those deemed high risk – including people arrested for violent felonies – will not be released before their arraignments.

While the passage of Prop 25 would require each of California’s 58 counties to decide which risk-assessment tool to use, several that are already popular in the US give a good sense of which ones they could roll out and how they would work.

The most widely used risk assessment tool is Arnold Ventures’ Public Safety Assessment, or PSA. Data collected by MediaJustice and Movement Alliance Project, which support pretrial release, estimates the PSA is used in at least five states and 59 counties in the US. San Francisco, which did away with cash bail in early 2020, is among them.

First rolled out in 2014, it considers nine factors of various weights in its algorithm, including a defendant’s age when they were arrested, whether they have any pending charges and whether they’ve been convicted of a violent crime in the past. The results are predictions about the likelihood that the person will fail to appear in court or get arrested in the interim, which judges can use to help make a decision about whether to release or keep a person in jail.

It’s not clear how well the PSA or other tools work, though. Megan Stevenson, an associate professor at the University of Virginia School of Law who studies criminal justice reform and algorithmic risk assessment, said that thus far there isn’t a ton of evidence either way. In Kentucky, for instance, which started mandating the use of such tools in 2011, the risk assessment scores were ignored more often than not by judges, Stevenson said. Kentucky has since adopted a more comprehensive pretrial detention policy, she said.

“I think the most helpful thing they’ve done is they’ve facilitated reform,” Stevenson said of the tools.

Asked how well the PSA specifically can forecast what will happen to defendants before their court dates, CNN Business was told that it has been shown to be predictive.

“It’s something that needs to be studied on an ongoing basis,” said Alison Shames, co-director of Advancing Pretrial Policy and Research, which is an initiative supported by Arnold Ventures that works with jurisdictions to roll out the PSA and researches pretrial assessments. Shames added that jurisdictions using the tool are told to collect data and determine how well it works for them.

How it could go wrong

The new system would likely lead to more people charged with low-level offenses being released before their trials, according to Jack Glaser, a professor at the University of California, Berkeley’s School of Public Policy.

“Without the bail option, more low-level offense, low-risk individuals will be released, and this will especially benefit those who otherwise didn’t have means to post bail,” he said. “On the flip side, more violent offenders with relatively high flight risk will be held in jail because the bail option is gone.”

A recent study from UC Berkeley’s California Policy Lab tried to quantify how much these tools could help if Prop 25 passes. Researchers pored over bookings in San Francisco and Sonoma Counties in 2017 and 2018, and found that if Prop 25 had been law the percentage of cases eligible for pretrial release would climb from 44% to 59% in San Francisco and from 63% to 66% in Sonoma County.

However, while the researchers found that release rates would rise for all racial groups, they determined there would still be racial disparities in pretrial releases – particularly for Black people, who make up roughly 13% of the US population but 27% of arrests.

“Just using a risk-assessment tool doesn’t erase that reality from the data,” said Johanna Lacoe, co-author of the study and California Policy Lab research director. “It uses data where that’s a problem.”

That was the conclusion reached earlier this year by the Pretrial Justice Institute. For decades the non-profit supported the use of pretrial risk assessments as an alternative to cash bail. In February, it announced it had changed its mind, saying in a statement at the time, “we made a mistake”.

Cherise Fanno Burdeen, executive partner at the Pretrial Justice Institute, explained the about-face came down to a belief that policing in the US has not been equitable for hundreds of years, which would make any algorithm using criminal history data itself racially inequitable.

“Data itself is not biased; it’s simply a reflection of biased policies,” she said.