An AI should never make a decision for child welfare. It simply isn't black and white and can just be tied to logic. Child welfare involves a lot more than just behavior, it involves emotions, something AI currently lack and probably will for a long time. The fact it is already in use is terrible. You can't just be like "Parent has X, so their children needs Y", there are maybe hundreds of factors in play on their children's welfare and a disability doesn't mean a child can't thrive nor does it mean a child's welfare is in danger. People who has no disabilities can be terrible parents too and a disability doesn't necessarily make one a bad parent. There are simply too many factors in a child's upbringing that their welfare isn't solely based on treatment from a parent. While parents play a large role in the welfare of children, then disabilities of the parent can be treated and almost be of no harm to a child's welfare.
Do AI's necessarily output based on logic?
It's making choices everywhere. Almost every choice in public life has the same argument and modern ai has the same problems