AI systems are increasingly used to assist socially complex forms of human work, centered on reasoning about and responding to the needs, behaviors, and intentions of other people. For example, AI-based tools are designed to support care workers in providing socio-emotional support or to support social service workers in making decisions about how best to allocate scarce resources. However, in practice, such systems often fail to meaningfully augment worker abilities. How can we design towards more pro-worker futures for AI in socially complex work, where AI deployments truly complement and enhance worker expertise and capabilities? In my research, I explore this question in close collaboration with organizations and workers in feminized labor contexts such as social work and education, where worker expertise has historically been devalued and misunderstood.

In the first thread of my research, I conduct field observations, design research, experimental studies, and case study analyses to advance our understanding of why AI for socially complex work often fails to meaningfully support workers. Across a series of studies, I adopt a social ecological approach to explore this question at multiple levels, investigating worker-level, organization-level, and societal-level challenges, and I identify opportunities for future work to address these challenges. In the second thread of my research, I explore the design of new processes and tools to begin addressing these opportunities. First, I explore whether the design of new training interfaces might help social workers learn to use AI tools more critically. Second, I collaborate with public sector agencies and community groups across the US to develop a new toolkit to support multi-stakeholder deliberations around proposed AI projects. Finally, I collaborate with a local school social work organization to explore how to support worker participation in the design of AI measurement instruments (e.g., benchmarks and metrics) that capture workers' beliefs about what it means to meaningfully augment their work.

In sum, this thesis provides conceptual and empirical foundations to guide towards pro-worker futures for AI in the workplace–specifically by identifying challenges and opportunities across multiple levels of the social ecology that interacts with workplace AI.

343 pages

Thesis Committee:
Kenneth Holstein (Co-chair)
Haiyi Zhu (Co-chair)
Jodi Forlizzi
Alexandra Chouldechova (Abridge)
Krzysztof Gajos (Harvard University)

Brad A. Myers, Head, Human-Computer Interaction Institute
Martial Hebert, Dean, School of Computer Science

Return to: SCS Technical Report Collection
School of Computer Science homepage

This page maintained by reports@cs.cmu.edu