Strengthening Communities: Designing AI for Knowledge Synthesis and Transfer in Social Services

The welfare of vulnerable and marginalized individuals, including those facing poverty, food insecurity, and housing instability, is significantly influenced by their ability to access essential social services. Community social service organizations work tirelessly to meet these needs, yet their effectiveness is shaped not only by available resources but also by how knowledge is managed, shared, and retained across their networks. Unlike corporate environments with structured knowledge management systems, community social service organizations rely heavily on informal processes, including conversations, ad-hoc emails, and staff memory, leading to inefficiencies, knowledge silos, and challenges in maintaining service continuity. To investigate these challenges and explore the role of artificial intelligence (AI) in supporting knowledge management, this dissertation presents three empirical studies. First, we conducted an ethnographic study with community service workers (CSWs) across two organizations to understand how they manage and transfer knowledge in their daily work. Our findings reveal that CSWs navigate high caseloads, shifting service landscapes, and documentation burdens, often relying on personal networks and tacit knowledge rather than formalized systems. The study also highlight the need for accessible, sustainable knowledge management solutions that fit the informal knowledge-sharing practices of CSWs while enhancing knowledge retention, knowledge transfer, and collaboration. Second, we engaged CSWs in participatory design workshops across four organizations, co-designing AI-driven interventions to support knowledge management and knowledge transfer. Participants recognized the potential of AI and Large Language Models (LLMs) to improve information synthesis, reduce administrative overhead, and enhance inter-agency coordination. However, they also voiced concerns about privacy, data bias, and the risk of AI undermining the human-centered values of their work. These discussions informed key design considerations for AI integration, ensuring that automation complements rather than replaces human expertise. Finally, we explored AI-assisted case note management through co-design workshops with caseworkers. The study revealed significant challenges in writing and reading case notes, including uncertainty in documentation and difficulties in retrieving past case information. Caseworkers saw the potential of LLMs to facilitate structured documentation and enhance knowledge transfer but also raised concerns about automation reducing personalization and the need for privacy safeguards. These insights informed design guidelines for LLM-driven case note management and design considerations to align with social service workflows. This dissertation is grounded in the following thesis statement, which shaped both the empirical investigation of knowledge management in community social services and the broader examination of AI design in resource-limited environments: AI is promising for improving knowledge management in community social services, but its implementation must be approached with care. If designed without consideration of the social, ethical, and practical realities of service work, AI systems risk amplifying biases and disrupting the trust-based relationships that define social services. By centering the voices of community service workers and prioritizing AI as a collaborative tool rather than a substitute for human expertise, we can design technology that strengthens, rather than disrupts, the relational and human-centered nature of social service work.