NYC lets AI gamble with Child Welfare

The Markup revealed in its reporting last month that New York City’s Administration for Children’s Services (ACS) has been quietly deploying an algorithmic tool to categorize families as “high risk”. Using a grab-bag of factors like neighborhood and mother’s age, this AI tool can put families under intensified scrutiny without proper justification and oversight.

ACS knocking on your door is a nightmare for any parent, with the risk that any mistakes can break up your family and have your children sent to the foster care system. Putting a family under such scrutiny shouldn’t be taken lightly and shouldn’t be a testing ground for  automated decision-making by the government.

 This “AI” tool, developed internally by ACS’s Office of Research Analytics, scores families for “risk” using 279 variables and subjects those deemed highest-risk to intensified scrutiny. The lack of transparency, accountability, or due process protections demonstrates that ACS has learned nothing from the failures of similar products in the realm of child services.

The algorithm operates in complete secrecy and the harms from this opaque “AI theater” are not theoretical. The 279 variables are derived only from cases back in 2013 and 2014 where children were seriously harmed. However, it is unclear how many cases were analyzed, what, if any, kind of auditing and testing was conducted, and whether including of data from other years would have altered the scoring.

What we do know is disturbing: Black families in NYC face ACS investigations at seven times the rate of white families and ACS staff has admitted that the agency is more punitive towards Black families, with parents and advocates calling its practices “predatory.” It is likely that the algorithm effectively automates and amplifies this discrimination.

Despite the disturbing lack of transparency and accountability, ACS’s usage of this system has subjected families that this system ranks as “highest risk” to additional scrutiny, including possible home visits, calls to teachers and family, or consultations with outside experts. But those families, their attorneys, and even caseworkers don’t know when and why the system flags a case, making it difficult to challenge the circumstances or process that leads to this intensified scrutiny.

This is not the only incidence in which usage of AI tools in the child services system has encountered issues with systemic biases. Back in 2022, the Associated Press reported that Carnegie Mellon researchers found that from August 2016 to May 2018, Allegheny County in Pennsylvania used an algorithmic tool that flagged 32.5% of Black children for “mandatory” investigation compared to just 20.8% of white, all while social workers disagreed with the algorithm’s risk scores about one-third of the time.

The Allegheny system operates with the same toxic combination of secrecy and bias now plaguing NYC. Families and their attorneys can never know their algorithmic scores, making it impossible to challenge decisions that could destroy their lives. When a judge asked to see a family’s score in court, the county resisted, claiming it didn’t want to influence legal proceedings with algorithmic numbers, which suggests that the scores are too unreliable for judicial scrutiny yet acceptable for targeting families.

Elsewhere these biased systems were successfully challenged. The developers of the Allegheny tool had already had their product rejected in New Zealand, where researchers correctly identified that the tool would likely result in more Māori families being tagged for investigation. Meanwhile, California spent $195,273 developing a similar tool before abandoning it in 2019 due in part to concerns about racial equity.

Governmental deployment of automated and algorithmic decision making not only perpetuates social inequalities, but removes mechanisms for accountability when agencies make mistakes. The state should not be using these tools for rights-determining decisions and any other uses must be subject to vigorous scrutiny and independent auditing to ensure the public’s trust in the government’s actions.