The Promises and Perils of using LLMs for Effective Public Services

TL;DR

This paper evaluates the potential and limitations of using Large Language Models and topic modeling to assist public child welfare agencies, highlighting both promising applications and critical challenges in decision-making support.

Contribution

It introduces a novel application of LocalLLM and BERTopic models in tracking child welfare case progress and discusses their effectiveness and shortcomings in supporting social workers.

Findings

Models can signal case progress and deviations effectively.

They struggle with detecting complex, discretionary case trajectories.

The paper outlines future participatory design directions for AI tools in public services.

Abstract

Governments are the primary providers of essential public services and are responsible for delivering them effectively. In high-stakes decision-making domains such as child welfare (CW), agencies must protect children without unnecessarily prolonging a family's engagement with the system. With growing optimism around AI, governments are pushing for its integration but concerns regarding feasibility and harms remain. Through collaborations with a large Canadian CW agency, we examined how LocalLLM and BERTopic models can track CW case progress. We demonstrate how the tools can potentially assist workers in opportunistically addressing gaps in their work by signaling case progress/deviations. And yet, we also show how they fail to detect case trajectories that require discretionary judgments grounded in social work training, areas where practitioners would actually want support to pre-emptively address substantive case concerns. We also provide a roadmap of future participatory directions to co-design language tools for/with the public sector.