AI Tool for Adoption Matching Falls Short for Vulnerable Foster Kids
Child protective services agencies have long struggled to find lasting homes for vulnerable children and teens. Thea Ramirez, CEO of Adoption-Share, introduced an artificial intelligence-powered tool called Family-Match to address this issue. Ramirez claimed that the algorithm, designed by former researchers at an online dating service, could predict which adoptive families would stay together and improve adoption success rates.
However, an investigation by the Associated Press revealed that the AI tool has produced limited results in the states where it has been used. Virginia and Georgia discontinued the use of the algorithm after trial runs, citing its inability to produce adoptions. Tennessee scrapped the program before implementation due to compatibility issues, and social workers in Florida reported mixed experiences with Family-Match.
State officials expressed concerns about the lack of transparency regarding the algorithm’s inner workings and the ownership of sensitive data collected by Adoption-Share. The experiences with Family-Match highlight the challenges of deploying predictive analytics in social service agencies without a complete understanding of the technology’s limitations.
Ramirez, who has a background in social work and is the wife of a Georgia pastor, has been a proponent of adoption as an alternative to abortion. She collaborated with Gian Gonzaga, a former research scientist at eharmony, to develop the adoption matchmaking tool. Gonzaga referred questions to Ramirez and eharmony denied any affiliation with Family-Match.
Despite the shortcomings of the AI tool, Ramirez has garnered support from conservative groups and individuals, including first lady Melania Trump. Adoption-Share has secured contracts and grants, generating millions in revenue since 2016.
In Virginia’s two-year test of Family-Match, only one known adoption was facilitated by the algorithm. Social workers in the state found the tool to be unhelpful and questioned its effectiveness. Similarly, Georgia ended its pilot program after only two adoptions, and social workers raised concerns about the tool’s matching recommendations.
Overall, the experiences with Family-Match serve as a lesson for social service agencies attempting to implement predictive analytics in addressing the challenges of finding homes for vulnerable children.