Allegheny County, the city of Pittsburgh and the University of Pittsburgh's Institute for Cyber Law, Policy and Security are teaming up to address bias in the region's automated decision making tools. The 22-person Pittsburgh Task Force on Public Algorithms was announced Wednesday, and is charged with forming guidelines and best practices for municipal algorithms.
Allegheny County currently uses automated decision systems for three purposes: to determine whether an allegation of child mistreatment should be investigated; whether to confine defendents ahead of trial; and where police should be deployed. According to Pitt Cyber, the county is also looking to deploy automated systems for predicting opioid overdoses and screening newborns for risk of neglect.
Algorithms make decisions based on historical data: in the case of policing, the system will look at an area's past crime rates to determine where police may be beneficial. But Pitt Cyber director David Hickton said this data is often biased.
"We would be allocating the police to a community based on historical numbers, and reinforce those numbers," said Hickton, a former U.S. Attorney for the Western District of Pennsylvania. "And we would not get a fair and accurate assessment because the police would be in a community that had previously been over-policed, and they would ignore communities that might need policing."
In 2018, New York City formed an Automated Decision Systems Task Force to oversee use of algorithms. Its report was published in November, but advocates, and some task force members, expressed concern that the process was not transparent and had limited engagement with the public.
Angel Diaz, counsel at the New York City-based Brennan Center for Justice and a member of the Pittsburgh Task Force on Public Algorithms, said it's important for the task force to engage in a sustained way with the public.
"That will require a lot of opportunities for the public to come and learn about how these systems are being used in ways that affect their lives," Diaz said. "But also have an opportunity to raise their issues and concerns so the task force can appropriately and accurately reflect their concerns into the recommendations."
A full report with best practices and guidelines for algorithms is expected in the summer of 2021. Public meetings will be held on March 10th at the Homewood-Brushton Branch YMCA and March 19th at the Beltzhoover Consensus Group monthly meeting.