Family-Match adoption algorithm fails to live up to promises

Occurred: November 2023

Can you improve this page?
Share your insights with us

An AI-powered tool introduced to increase the likelihood of orphans and adoptive families being a good match in the USA has had little effect in the states where it has been used.

Developed by former social worker Thea Ramirez, Family-Match provides an algorithmically-generated 'relational fit' score on the basis of information about a child submitted by foster parents or social workers, and by people looking to adopt. It then presents a list of the most suitable potential parents for every child.

However, an AP investigation found that two states had dropped the Family-Match after initial pilots, and that social workers in Florida, Georgia and Virginia complained that it was not useful, and that it pairs foster kids with unwilling families. 

According to AP, Ramirez had 'overstated the capabilities of the proprietary algorithm to government officials as she has sought to expand its reach', and that Adoption-Share, the non-profit that runs Family-Match, provides little transparency about how its algorithm works.

AIAAIC view

AP's findings raises significant questions about the effectiveness and value of Family-Match, and more generally regarding the use of artificial intelligence in solving adoption challenges.

Adoption-Share's reluctance to provide access to information and data about how Family-Match works fits in with a broader pattern of developers and operators exploiting the opacity of their systems so they can be unduly hyped with little risk of recourse.

Databank

Operator: Florida Department of Health; Georgia Department of Public Health; Virginia Department of Health
Developer: Adoption-Share
Country: USA
Sector: Govt - welfare
Purpose: Predict adoption effectiveness
Technology: Prediction algorithm; Machine learning
Issue: Accuracy/reliability; Value/effectiveness
Transparency: Governance; Marketing

Page info
Type: Incident
Published: November 2023