Taking on AI bias
Connecting state and local government leaders
Policy experts in Pittsburgh are working to ensure that municipal decision-making algorithms improve services delivery but do not discriminate against residents based on race or demographics.
Bias that often unwittingly creeps into artificial intelligence models can have broad and damaging impacts. Facial recognition systems misidentify people of color, and judicial risk assessments that have been demonstrated by ProPublica as "remarkably unreliable" in flagging people likely to commit violent crimes.
Policy experts at the University of Pittsburgh have launched a task form that will establish best practices and practical guidelines for the use of municipal decision-making algorithms. Those recommendations are intended to ensure the models used in human service and criminal justice settings In Allegheny County improve government services but do not discriminate against people based on race or demographics.
Local government has been using AI for several years. In 2016, Allegheny County's Department of Human Services began using predictive analytics to inform child welfare decisions with a system that rated the severity of child abuse complaints. The system tapped into databases related to criminal justice, social services and drug and alcohol treatment programs to predict the risks to a child based on factors present in the household.
The county was also an early adopter of a risk assessment tool that used administrative data to calculate a risk score to aid judges in making their pretrial decisions in determining bail.
The “Hello Baby” program, set to launch this month, uses a predictive risk model that accesses available data -- such as birth records, child welfare records, homelessness, jail/juvenile probation records -- to determine the appropriate outreach and support to offer families with newborns.
The 22-member Pittsburgh Task Force on Public Algorithms is sponsored by Pitt’s Institute for Cyber Law, Policy and Security, commonly known as Pitt Cyber.
“Increasingly, algorithms are being used to facilitate efficient government. We need to ensure that historical discrimination and existing inequities are not reinforced,” said Pitt Cyber Founding Director and Task Force Chair David Hickton, in a statement. “Pittsburgh should lead the way in effective and fair oversight of these systems. We can be a national model, ensuring algorithmic accountability and equity for all residents.”
Task force members are drawn from local and national experts and community leaders. The task force is also served by a government advisory panel with designees from Allegheny County and the City of Pittsburgh.
The group will use a combination of community outreach meetings and public comments to assess county residents’ major concerns with municipal decision-making algorithms. In summer 2021, it plans to publish a full report of its research and recommendations for best practices for the technology.