Designing Human-Centered Algorithms for the Public Sector: A Case Study of the U.S. Child-Welfare System

TL;DR

This paper explores how to design human-centered algorithms for the U.S. Child Welfare System, focusing on improving decision-making, understanding systemic disparities, and supporting caseworkers through empirical studies and practical insights.

Contribution

It offers a comprehensive analysis of human-algorithm interaction in public sector decision-making and proposes actionable guidelines for designing more effective, context-aware algorithms.

Findings

Caseworkers' interactions with algorithms influence decision quality.

Casenotes reveal systemic constraints and invisible labor.

Human-centered design improves algorithmic support for discretionary work.

Abstract

The U.S. Child Welfare System (CWS) is increasingly seeking to emulate business models of the private sector centered in efficiency, cost reduction, and innovation through the adoption of algorithms. These data-driven systems purportedly improve decision-making, however, the public sector poses its own set of challenges with respect to the technical, theoretical, cultural, and societal implications of algorithmic decision-making. To fill these gaps, my dissertation comprises four studies that examine: 1) how caseworkers interact with algorithms in their day-to-day discretionary work, 2) the impact of algorithmic decision-making on the nature of practice, organization, and street-level decision-making, 3) how casenotes can help unpack patterns of invisible labor and contextualize decision-making processes, and 4) how casenotes can help uncover deeper systemic constraints and risk factors that are hard to quantify but directly impact families and street-level decision-making. My goal for this research is to investigate systemic disparities and design and develop algorithmic systems that are centered in the theory of practice and improve the quality of human discretionary work. These studies have provided actionable steps for human-centered algorithm design in the public sector.