Machine-based risk predictions can cut crime or jail populations dramatically

Evidence

This is not the only example of legal technology working for good. As discussed above, in San Francisco, AI is being used to blind police reports so that prosecutors cannot tell the race of defendants (Gecker, 2019). Jurisdictions across the country are using the Public Safety Assessment tool (PSA), which was designed to reduce bias in judicial decision-making. In assessing dangerousness, the tool takes into account age and history of criminal convictions, but it elides race, gender, employment background, where a defendant lives, and history of criminal arrests (Arnold Foundation, 2018). Finally, there is groundbreaking theoretical work showing that machine predictions can lead to drastically improved legal outcomes (Kleinberg et al., 2018). The authors found that their policy simulation would reduce crime by up to 25 percent without causing an increase in jail populations. Alternatively, it would be possible to keep crime rates steady while reducing jail populations by 42 percent. — Joseph Avery, Jim Sidanius, Joel Cooper, Gregory Davis, Amanda N., Bias in the Law, 1172 (#expands, via linked claim)

Before her arrest, the state tracked every terrestrial and virtual footstep Jelilova Gulbahar left behind. Every online purchase and social media exchange, every phone conversation and checkpoint stop supplied the state with a fluid stream of data, data fed into a policing algorithm that led to Jelilova’s identification as a “terrorist.” — Khaled A. Beydoun, The New Crusades_ Islamophobia and the…, 531 (#complicates, via linked claim)

The long-overdue evaluation, in 2006, found that four of the variables in their model—whether the defendant had lived in Minnesota for more than three months, whether they lived alone, their age at the time of booking, and whether a weapon was involved in their charge—had virtually no bearing whatsoever on their actual risk of committing a new offense while waiting for their trial date or of missing a required court appearance. And yet the model had been recommending pretrial detention on those grounds. Worse, three of those four factors strongly correlated with race. — Brian Christian, The Alignment Problem, 269 (#complicates, via linked claim)

Categorization

• XREF: connects to: Palantir’s software predicts individual violence likelihood for police — Shoshana Zuboff, The Age of Surveillance Capitalism, 1236; Machine predictions can cut crime by a quarter without growing jail populations — Joseph Avery, Jim Sidanius, Joel Cooper, Gregory Davis, Amanda N., Bias in the Law: A Definitive Look at…, 1172; Untested risk models can detain people on factors having no real predictive value — Brian Christian, The Alignment Problem, 269 (+2 more)

Broader Connections

• Machine predictions can cut crime by a quarter without growing jail populations #expands (0.97) — Joseph Avery, Jim Sidanius, Joel Cooper, Gregory Davis, Amanda N., Bias in the Law, 1172
• Predictive policing algorithms tagged Uyghur women as terrorists before arrest. #complicates (0.86) — Khaled A. Beydoun, The New Crusades_ Islamophobia and the…, 531
• Untested risk models can detain people on factors having no real predictive value #complicates (0.85) — Brian Christian, The Alignment Problem, 269
• Utilitarian risk arithmetic justifies sacrificing innocent convicts for public benefit #complicates (0.76)