AI Matchmaking Fails to Find Forever Homes for Vulnerable Foster Kids

A portion of these individuals are without parents, while others have been taken away from their guardians. A significant number of them are older and face numerous challenges or disabilities. The majority carry the emotional wounds of being moved frequently between foster homes, separated from their siblings, or subjected to sexual and physical abuse.

For decades, child welfare agencies have struggled with how to find sustainable homes for such vulnerable children and youth—a challenge so enormous that social workers can never guarantee a perfect fit.

Thea Ramirez stepped into this war with a tool she promoted as a technical solution, powered by artificial intelligence, which can eventually predict which Adoptive Families will stay together. Ramirez argued that this algorithm, designed by former researchers at an online dating service, could promote successful adoptions across the United States and make cash-strapped child welfare agencies more efficient.

“We’re using science — not just preferences — to create scores that can predict long-term success,” Ramirez said in an April 2021 YouTube video in which he described his goals to flip “the script on how America connects children and families.” Family-Match algorithm.

However, an Associated Press investigation found that the AI tool — among the few deployment algorithms on the market — has produced limited results in states where it has been used, according to Family-Match’s self-reported data obtained from public records by the AP. at the request of state and local agencies.

Ramirez has also overstated the proprietary algorithm’s capabilities to state officials as he has sought to expand its reach, though social workers told the AP the tool was not helpful and often led them to unwilling families.

Virginia and Georgia abandoned the algorithm after trials, citing its inability to produce adoptions, though both states have continued to do business with Ramirez’s nonprofit, Adoption-Share, according to an AP review of hundreds of pages of documents.

Tennessee scrapped the program before it was implemented, saying it didn’t work in their internal system even after state officials tried to create it for more than two years, and social workers reported mixed experiences with Family-Match in Florida, where it was used. has expanded.

State officials told the AP that the organization run by Ramirez as CEO owns some of the sensitive data collected by Family-Match. They also noted that the nonprofit offered little transparency about how the algorithm worked.

According to the AP, these experiences provide lessons for social service agencies looking to deploy predictive analytics without fully understanding the technology’s limitations, especially when trying to address such enduring human challenges as finding homes for children the judges describe as “least adoptable.” .”

“There’s never going to be a foolproof way to predict people’s behavior,” said Bonni Goodwin, a child welfare data expert at the University of Oklahoma. “Nothing is more unpredictable than puberty.”

Ramirez, of Brunswick, Georgia, where his nonprofit is also based, declined to provide details about the inner workings of the algorithm and declined interview requests. In an email, she said the tool was a starting point for social workers and did not decide whether to adopt a child. He also disputed child welfare directors’ accounts of Family-Match’s performance.

“User satisfaction surveys and check-ins with end users in our office show that Family-Match is a valuable tool and useful to users who actively use it to support recruiting,” Ramirez wrote.

EXCITED FOR MACHINE TURNING

Ramirez, a former social worker and Georgia pastor’s wife, has long pushed to promote adoption as a way to reduce abortions, according to her public statements, newsletters and blog posts.

More than ten years ago, she launched a website that connects pregnant women with potential adoptive parents. He marketed it as “the ONLY online community dedicated exclusively to networking crisis pregnancy centers” and promised to donate 10% of membership fees to anti-abortion counseling centers whose goal is to get women to terminate their pregnancies. Ramirez said in an email that Family-Match is not affiliated with such centers.

Next, she focused on helping children in foster care who have no family members to raise them. Most of the 50,000 children adopted nationwide in 2021 ended up with relatives, federal statistics show, while about 5,000 ended up with people they didn’t know before. Such recruitment-based adoptions are the most difficult to implement, say social workers.

Ramirez has said he called Gian Gonzaga, a researcher who had mastered the algorithms at eharmony, a dating site with Christian roots that promises users “true love” for marriage seekers. He asked Gonzaga if she could work with him to create an adoptive couple.

Gonzaga, who worked with his wife Heather Setrakian on harmony and then the Family-Match algorithm, relayed questions to Ramirez. Setrakian said he is very proud of his years of work developing the Family-Match model.

Eharmony spokeswoman Kristen Berry said the dating site is “not affiliated with Family-Match.” Berry described Gonzaga and Setrakian as “just former employees.”

NOT “PERFECTLY USEFUL”

Ramirez later began touring the country promoting Family-Match to government officials. His work and religious beliefs drew support primarily from conservatives, including first lady Melania Trump, who highlighted Ramirez’s efforts at a foster care event in the White House Situation Room. Ramirez has written reports and given a high-profile presentation at the American Enterprise Institute, benefited from high-profile fundraising events and used connections to get state officials to pilot his tool.

Social workers say Family-Match works like this: Adults seeking adoption submit survey responses through the algorithm’s online platform, and foster parents or social workers enter each child’s information.

After the algorithm has generated a score that measures “relative fit,” Family-Match displays a list of the best possible parents for each child. Social workers then check the candidates.

In the best case scenario, the child is matched and placed at home for a trial; the parents then submit the legal papers to formalize the adoption.

Family-Match began reuniting families in Florida and Virginia in 2018. Then-Virginia Gov. Terry McAuliffe, a Democrat, ordered the pilot at the urging of a campaign donor whom he named the state’s “adoption champion.” In Florida, which has a privatized child welfare system, regional foster care organizations soon signed up to the algorithm for free—thanks to a grant from a foundation founded by the then-CEO of Patrón tequila and his wife.

As philanthropic dollars dried up in Florida, the state government picked up the tab and last month awarded Adoptio-Share a $350,000 contract for its services.

Pilots in Tennessee and Georgia followed.

Adoptio-Share has generated $4.2 million in revenue since 2016; it will report about $1.2 million in 2022, according to tax filings.

In Virginia’s two-year Family-Match test, the algorithm produced only one known adoption, officials said.

“Local staff indicated they did not find the tool particularly helpful,” the Virginia Department of Social Services said in a statement, noting that Family-Match “had not been proven effective” in the state.

Social workers in Virginia were also confused that the algorithm seemed to match all the children to the same parent, said Traci Jones, deputy director of the state’s Department of Social Services.

“We didn’t have access to the algorithm even after we requested it,” Jones said.

By 2022, Virginia had awarded Adoption-Share an even larger contract for a different foster care initiative that the nonprofit says will “leverage” the Family-Match app.

Georgia officials said they ended their first pilot in October 2022 because the tool didn’t work as planned, ultimately resulting in only two uses during their year-long trial.

Social workers said the tool’s compatibility recommendations often led them to unwilling parents, prompting them to question whether the algorithm correctly assessed the adults’ ability to adopt these children.

Ramirez met with the governor’s office and also lobbied the state committee on direct assets, saying the tool was an “incredible accomplishment.” By July, the Georgia Department of Human Services signed a new agreement with Adoption-Share to use Family-Match again — this time for free, agency spokeswoman Kylie Winton said.

Florida’s privatized child welfare system works with more than a dozen regional agencies that provide foster care and adoption services. When the AP requested public information about its Family-Match cases, many of these agencies gave the tool mixed reviews and were unable to explain Family-Match’s self-reported data, making it difficult to assess the algorithm’s supposed success rate.

In the state of Florida, Family-Match sought reimbursement for 603 placements that resulted in 431 adoptions over a five-year period. According to Adoption-Share’s fiscal year 2023 third quarter report, AP received from a Pensacola-based child welfare organization.

Scott Stevens, an attorney representing the FamiliesFirst Network, told the AP in June that only three trial placements recommended by Family-Match had failed since the agency began using the algorithm in 2019. But Adoption-Share data, which Stevens provided to the AP, shows that her office made 76 other Family-Match placements that did not show the children had been formally adopted. Asked by the AP for clarification, Stevens could not say what happened in those 76 cases and referred additional questions to Family-Match.

Ramirez declined to discuss the split, but acknowledged in an email that not all matches work.

“Transitions can take time on the road to adoption,” Ramirez said in an email, adding that “the decision to finalize an adoption is ultimately the agency’s responsibility” for input from children and judges. Adoption-Share posted on their Facebook page on Sunday that the organization had reached 500 adoptions in Florida!

Jenn Petion, president and CEO of an adoption organization in Jacksonville, said she likes how the algorithm allows her team to tap into a statewide pool of prospective parents. Petion has also credited Family-Match for helping him find his adopted daughter, whom he described in an Adoption-Share annual report as a “100 percent match.”

Family-Match helps social workers make “better decisions, better matches,” Petion said, but his agency, Family Support Services, declined to provide statistics on Family-Match.

Fort Myers-based Southwest Florida Children’s Network said the Family-Match tool has resulted in 22 matches and eight adoptions over the past five years, compared to hundreds of matches and hundreds of adoptions its social workers made without the tool.

Bree Bofill, adoption program director for Miami-based Citrus Family Care Network, said social workers found the tool didn’t work very well and often suggested potential families that weren’t right.

“It’s frustrating to be told the kids are a match, but in reality the families aren’t interested in them,” Bofill said of the algorithm.

Bofill also said it was difficult to assess the utility of the tool because Family-Match officials sometimes asked social workers who found potential parents to tell adults to sign up for the tool even if it didn’t matter in the adoption, allowing the algorithm to claim credit for the match.

Winton, a spokesman for the Georgia agency, told the AP a similar point — Family-Match could claim reimbursement for pairings if the child and parent were already in its system, even if the program didn’t create a match. Family-Match, in a “confidential” user guide posted online in April 2023, told social workers not to delete cases found outside the tool. Instead, they were told to document the match in the system so that Adoption-Share could refine its algorithm and track families.

Ramirez did not address Bofill’s claim, but said in an email that Family-Match’s reports reflect social workers’ input into the system.

“Children as Guinea Pigs”

Officials in Virginia, Georgia and Florida said they weren’t sure how the tool scored families on the highly sensitive variables the algorithm uses.

In Georgia, Family-Match still collects information on whether foster youth have been sexually abused, their gender, and whether they have a criminal record or “identify as LGBTQIA.” Such information is typically limited to the strictly protected case files of child protection authorities.

In Tennessee, a version of the algorithm’s questionnaire for prospective parents asked about their household income and their estimates of how “conventional” or “uncreative” they were. They were also asked whether they agreed – or disagreed – with the statement, whether they were asking for God’s help, according to information obtained by the AP.

When the Tennessee Department of Children’s Services reviewed the proposed Family-Match assessment, they questioned some of the data that Family-Match wanted to collect. Tennessee officials questioned why Family-Match needed certain sensitive data points and how that information affected the outcome of the match, according to an internal document in which state employees noted questions and feedback about the algorithm. Ramirez said the agency did not question the validity of the study and said the discussions were part of a streamlining process.

Virginia officials said that when families’ information was entered into the tool, “Adoption Share owned the information.”

In Florida, two agencies admitted to using Family-Match unofficially without a contract, but did not say how the children’s information was protected.

Ramirez would not say whether Family-Match has removed the pilot data from its servers, but said his organization maintains a compliance audit and adheres to the terms of the contract.

Social Security advocates and data security experts have raised alarms about government agencies’ growing reliance on predictive analytics to help them do their jobs. These researchers and advocates say such tools can exacerbate racial disparities and discriminate against families based on characteristics they cannot change.

Adoption-Share is part of a small group of organizations that say their algorithms can help social workers place children with foster or adoptive families.

“We’re basically using children as guinea pigs in these tools. They’re crash test dummies,” said Suresh Venkatasubramanian, a former deputy director of Biden’s White House Office of Science and Technology Policy who now works at Brown University. “That’s a big problem there.”

SEARCH FOR EXPANSION

Adoptio-Share is still looking to expand its business in places like New York, Delaware and Missouri, where child welfare officials are evaluating its pitch. Ramirez also said he saw an opportunity last year to present Family-Match to the U.S. Department of Health and Human Services “to showcase our tool and how it can be a useful resource.”

This year, Adoption-Share contracted with the Florida Department of Health to use Family-Match to build an algorithm to “increase the pool of families willing to foster and/or adopt medically complex children,” according to state contracts. Health officials did not respond to repeated requests for comment.

Connie Going, a longtime Florida social worker whose own viral adoption story has been credited by Ramirez as the inspiration for Family-Match, said she didn’t think the tool would help such vulnerable children. Going said the algorithm gives expectant parents false hope by failing to deliver successful matches and ultimately makes her job harder.

“We’ve been relying on something that’s not 100% useful,” Going said. “It’s wasted time for social workers and wasted emotional experiences for children.”

One more thing! ReturnByte is now on WhatsApp channels! Click here to join so that you never miss any updates from the world of technology.