Article content

From a new Beatles song, to the “word of the year,” AI is everywhere. The tentacles of artificial intelligence seem to reach around every corner — writing travel guides, picking stocks, and serving up custom graphics like “vampire unicorn in a Ramones T-shirt.”

Of all the ways AI is poised to make life easier and/or ruin everything, here’s one I did not see coming: adoption matchmaking.

Article content

Article content

The gears of adoption tend to move slowly. In Ontario, the public adoption process can take at least a year and sometimes longer, the Ministry of Children, Community and Social Services says. Matching a child’s needs with the right family takes time and care.

If it feels like a long wait for prospective parents, it’s much harder on the kids. According to the Children’s Aid Foundation of Canada, about 59,000 children and youth are currently in permanent government care — placed with foster families, extended family, or in group homes.

To make the process more efficient, some tech entrepreneurs looked to AI.

The Adoption Council of Ontario, the agency that runs the province’s adoption intake service, notes finding the right fit “means families who can affirm important parts of a child’s or youth’s identity, including their race, gender, ability, religion and culture, and meet their needs.” Some have complex medical requirements, or bear the scars of trauma or abuse.

Article content

Running through these tick-boxes for a match sound a little like a dating app, which is the model that inspired Family-Match. The U.S.-based app was conceived by a pastor’s wife, looking to promote adoption as an alternative to abortion, the Associated Press reported this week. She teamed with a researcher who managed algorithms at the dating site eHarmony. Their AI platform claimed to predict whether adoptive families would remain together.

Even with the involvement of social workers, it’s intuitively uncomfortable to hand over such delicate, person-centred work to an algorithm. The AP investigation raises concerns about the information collected by adoption apps, and who owns it. Case files with sensitive information — such as whether youth have a criminal record, or a history of sexual abuse — normally would be restricted to trained caseworkers, bound by confidentiality and ethics.

The trouble is, AI is only as good as its learning inputs. The more information it gathers, and the better the quality of that information, the more reliable its results.

Virginia and Georgia abandoned the Family-Match app after dismal trial runs, AP reports. A pilot project in Tennessee was grounded before it could launch. The app is still used in Florida, but the testimonials aren’t glowing. A program manager in Miami noted parents often had “no interest” in a child they were matched with. In Fort Myers, Fla., social workers found hundreds more matches and adoptions working without the app, than with it.

Article content

AI may be capable of good, but it’s really still in its experimental phase. We haven’t worked out all the kinks. Sometimes, generative AI pulls together reliable information and other times, it lies. It’s even been caught making up news stories as sources, to cover its tracks.

These deviations from reality are  called “hallucinations.” Nobody’s certain why they happen. Maybe the learning algorithm prioritizes coming up with a response, even when it doesn’t know the right one.

So far, AI in adoption matching may not be much of a time-saver, but that’s a trivial flaw. The bigger mistake is beta-testing with vulnerable kids’ futures.

[email protected]

Article content