Recently, the Los Angeles Police Department announced it would stop using algorithmic-based programs to identify who's most likely to commit violent crimes, according to The Los Angeles Times. An audit of the program by the department's inspector general found, among other things, that the police department used inconsistent criteria to label people “chronic offenders.”

The LAPD is one of many U.S. police departments and courts leveraging artificial intelligence-backed software to assist in policing, bail and sentencing decisions. While such tools are grabbing headlines for disproportionately targeting blacks and Latinos, observers say the root of these tools' problems is biased data.

“All the algorithms we've seen being used in the criminal justice realm use prior criminal history as a primary factor,” said Nyssa Taylor, criminal justice policy counsel for the American Civil Liberties Union of Pennsylvania.

Over-policing and plea bargains entered into not necessarily out of guilt, but by a defendant being low-income, can create misleading incarceration and recidivism data. “When you train a tool on a data set of that, it's going to reproduce the same biases in society,” said Hannah Sassaman, a Media Mobilizing Project policy director and 2017 Soros Justice fellow focused on limiting sentencing based on predictive algorithms.

“What people don't understand is what algorithms are, they are basically opinions disguised as facts,” said Philadelphia-based criminal defense attorney Troy Wilson. “People just think you are using numbers, you're using experiences, institutional racism to justify a particular result.”

He added, “A predictive algorithm that is trying to predict that you are such a high risk wants to give you a higher jail sentence before you committed another crime.”

Wilson said such a decision should not be taken lightly. “You should freak out—that's another reason I'm vehemently against it. How in the hell are you going to predict I'm going to commit another crime” when using AI and flawed data, he asked.

While criminal justice reform advocates stressed the weight government policy and policing have on data used in sentencing and bail tools, they also noted proprietary algorithms also presents obstacles.

In many cases, there is a lack of transparency over how such software came to its conclusion. Wilson said such proprietary software doesn't allow lawyers to cross-examine the software's developers.

Advocates for reform agreed the courts need to take account of racial disparities before deciding a defendant's sentence or likelihood to commit a crime.

“From my perspective the angst around assessment tools is a mild distraction from what should be the angst about the data we are using in the first place,” said Cherise Fanno Burdeen, CEO of Pretrial Justice Institute, an organization that assists states and local government with pretrial reforms. “It's not the assessment tool itself that is biased, and data by itself can not be biased, but it's generated by policies that are biased.”

She added, “There's no no silver bullet fix to racial equity, we have to go back to the policy.”