California prisons block AI researchers from examining parole denials (Demo)

A team of researchers who want to develop a machine learning platform to help analyze and detect any patterns of bias in California parole-suitability decisions has been blocked for years by the state’s Department of Corrections and Rehabilitation (CDCR). In a lawsuit filed today by the Electronic Frontier Foundation (EFF), the researchers argue that the state’s public records law requires the release of race and ethnicity data they need to develop their work.

“We want to create a machine learning tool that can extract factors from parole hearing transcripts, describe the current decision-making process, and identify which decisions appear inconsistent with that process and might be worthy of reconsideration. We need race data to do that,” said Catalin Voss, a PhD student at Stanford University.

“Modeling any criminal justice process without accounting for race is unthinkable. Race is embedded in the American justice system. Empirically, the only way to disentangle race from other factors is with the race data,” said Jenny Hong, a PhD candidate at Stanford.

The team – which also includes Nick McKeown, a Professor of Computer Science at Stanford, and Kristen Bell, an Assistant Professor of Law at the University of Oregon – sent their first California Public Records Act (CPRA) request for race and ethnicity data in September 2018. However, officials from CDCR denied that request and several more since, claiming exemptions to the CPRA that a judge in an earlier case has already ruled do not apply to this data.

“When I first connected with Catalin about developing this research, I assured him we’d be able to get data about race because it’s public record. I thought it might take two months. It has been two years, and we have no real answers,” said Bell.

“The California Department of Corrections and Rehabilitation is making the same arguments with us that have already lost in court,” said EFF Staff Attorney Saira Hussain. “Our clients want to use machine learning to identify patterns of discrimination – something you’d think prison officials might want to learn more about.”

Bell has conducted previous research on California parole-suitability decisions and found that race and other illegitimate factors had a significant influence on parole decisions for individuals sentenced to life with the possibility of parole as juveniles. During negotiations between the researchers and CDCR about a potential data release, one official said that the records would be provided only if Bell ceased her involvement in the project.

“What’s most upsetting is knowing that our experience is likely not unique,” said Bell. “There has been much debate about evidence-based criminal justice reform in California, but how can we know if we’re moving any closer to justice when the prison system is preventing independent researchers from accessing race data?”

“Government officials should not be deciding which researchers get access to information based on whether they think they will like the answers,” said EFF Staff Attorney Cara Gagliano. “Our clients simply want CDCR to follow the law and provide the records they need to do their work.”

For the full petition in Voss v. CDCR:
https://www.eff.org/document/petition-voss-v-cdcr