Technology
Algorithmic Decision-Making and Discretionary Power in Social Services
Quick fact
In many U.S. states, predictive risk models now flag families for child welfare investigations—sometimes without the caseworker being told why a family was selected, shifting the power of judgment from the caseworker to an opaque formula.
Why this is interesting
A caseworker must decide whether to remove a child from a home. An algorithm says yes. But the caseworker has a hunch—a small clue that might change everything. Who should have the final say?
Read the full explanation
Understanding Algorithmic Decision-Making and Discretionary Power in Social Services
Think of a social worker as a detective and a caregiver combined. They gather information, talk to people, and use their experience and intuition to decide how to help each family. That freedom to adapt the rules to the person is called 'discretion.' Now, imagine a new colleague—an algorithm—that also gives recommendations. It processes large amounts of data to predict outcomes, like the likelihood of a child being harmed. This algorithm doesn't replace the caseworker entirely, but it changes the situation. At first, it seems helpful: it can handle huge caseloads and ensure consistency. But as agencies begin to rely on it, the algorithm's ranking can steer caseworkers' attention, framing which cases are 'high risk.' The caseworker might feel pressured to follow the machine's suggestion, because deviating requires extra justification. Over time, the algorithm effectively sets the boundaries of decision-making, narrowing the space for human judgment.
A deeper explanation
The mechanism works through three steps: first, the algorithm encodes a definition of risk (e.g., from historical data about previous cases). Second, it produces recommendations that are seen as neutral and objective, because they are data-driven. Third, institutions embed these recommendations into workflows—screening, prioritization, resource allocation—so that deviating from the recommendation becomes costly or difficult. This shifts discretionary power in two directions: at the frontline, the caseworker's latitude is reduced because they must justify any departure from a system recommendation, effectively moving discretion to the system's designers and the historical data that trained it. Moreover, the algorithm may not weigh factors that a human would find crucial, like a parent's untapped motivation to change. The result is a redefinition of what 'good judgment' means: instead of being a human skill, it becomes alignment with a score. Accountability also suffers, as the logic behind a recommendation may be invisible, making it hard to contest. Thus, algorithmic systems can be powerful tools, but they fundamentally redistribute discretion, raising questions about fairness, transparency, and human dignity.