“I Want to Be Pushed, I Want to Grow”: Enabling Social Workers to Design Evaluations of LLM Augmentation in Their Work (Extended Abstract)

Abstract

Workers are increasingly asked to adopt AI systems to assist their work, yet are rarely given a voice in defining what meaningful AI augmentation should look like in their work or how to evaluate for it. In this paper, we propose worker-driven AI measurement---a bottom-up approach to AI evaluation where workers collaboratively shape decisions about which tasks AI augments, what ``successful’’ augmentation means, and how it should be measured. We explore how to support this through a case study with 19 workers from a local school social work organization. Through an iterative series of eight workshops, workers identify and operationalize their measuremet goals, which are largely absent in existing industry evaluations. Specifically, workers develop a benchmark to capture how well an LLM serves as a ``people challenger’’ by challenging them to reflect on their own possible assumptions and biases in their work. Through iterative operationalization, workers design and refine an LLM-as-a-judge rubric with elaborations, examples, and edge cases that build upon their professional and lived expertise as workers in their specific organizational and work context. In validations of the worker-created benchmark, we find strong agreement between worker and LLM judge ratings and that the benchmark successfully differentiates performance across six state-of-the-art LLMs. Based on our case study, we discuss opportunities for future work to support worker-driven AI measurement as a complementary approach to existing top-down AI evaluation approaches.