Policy & Practice | Fall 2024

from our partners

By Molly Tierney

GenAI in Child Welfare Redux: Making the Knowable Known

I n a recent article on generative arti ficial intelligence (GenAI) in Child Welfare, 1 I shared my thinking about the urgent need to put new tools to use for caseworkers who, without them, enter into decision making poorly informed and, too often, are driven by confirmation bias related to the allega tion they have in hand. I hope for a turn of events in which caseworkers have a host of information that allows them to see parents as whole people with complex lives that include magnificent capabilities and enormous challenges. Providing caseworkers with a balanced view of a family and their circumstances might help them avoid the plethora of negative errors that plague human services: children coming into foster care when they could have been left safely at home with their family of origin. A well-respected colleague made a wise observation of the risks of deploying GenAI to help. It could tip the opposite way, increasing the power imbalance between caseworkers and families and invading the privacy of the families we serve. These are very good points that deserve thorough review. It’s important that we are honest about the balance of power in child welfare. There isn’t one. In a safety assessment, the caseworker has all the power, and the parent has none. Naming this is important because if we can admit it, we might get the caseworker to wield their power more gently. That being said, using GenAI or not using GenAI will not alter that dynamic. Only humans can, and they might be better positioned to do so if

they are more informed and prepared to share information that they have. Privacy is a noun that Merriam Webster 2 defines as “the quality or state of being apart from company or observation” and “freedom from unauthorized intrusion.” Presuming the presence of this “freedom” in child welfare is a misunderstanding about child protection. When an agency believes a child may be in harm’s way, we—literally, all of us who live in the United States—authorize them to take action. No member of the family should have an expectation of privacy once child protection is in the room. But we all know child protection arrives too late in the cycle. If we want to help families, the trick will be

knowing someone needs help as early as possible. It reminds me of an experience I had when I was leading an urban child welfare jurisdiction. Among the challenges we faced were the deaths of infants. Like all jurisdictions, part of our post-mortem process was to have a Child Fatality Review . The com mittee that convened for these reviews included child welfare, public safety, public health and leadership from every hospital in town. The purpose of the review was epidemiology: could we determine actions that might prevent future deaths? In one particular annual review, a pattern leaped off the page with statistical clarity: a substantial portion of the infants who died were

Policy & Practice Fall 2024 30