Designing Intelligent Interactive Systems for Vulnerable Populations

Abstract

Technology alone cannot drive social change for vulnerable populations—it amplifies existing human intentions and social forces, sometimes exacerbating inequalities rather than reducing them. This dissertation shows that well-intentioned technological interventions do not inherently benefit marginalized communities and examines how technology can effectively support vulnerable populations when designed to amplify the right forces. My qualitative analysis of LLM-based mental health chatbots revealed that while these systems offer on-demand, non-judgmental support that boosts user confidence, they face significant limitations in addressing the complex needs of vulnerable users. For LGBTQ+ individuals specifically, these chatbots frequently fail to grasp the nuances of their experiences, providing empathetic but ultimately inadequate support. This demonstrates how technology that merely attempts to ``fix'' missing institutional components without addressing underlying societal factors can widen rather than bridge existing disparities. To effectively support vulnerable populations, technology must amplify intentional positive forces. Through an experiment on dating interfaces, I demonstrated that supposedly ``neutral'' designs like swipe interfaces perpetuate racial bias, even among users who explicitly claimed not to consider race in their decision-making. By redesigning the interface to display substantive profile information before race information such as photos and names, I significantly reduced biased choices, showing how technology can meaningfully engage willing participants' stated values rather than amplifying implicit biases. Finally, my work with humanitarian frontline negotiators illustrates how technology can strengthen positive social forces by supporting existing expert practices. Rather than replacing human judgment with AI recommendations, I designed tools that enhance the negotiation process by contextualizing cases and exploring options with associated risks. This approach builds on practitioners' validation practices while augmenting their capabilities to secure essential resources for people in conflict zones. This dissertation argues that effective technological interventions must identify and strengthen underlying positive social forces while meaningfully engaging willing participants' stated values. By designing systems that amplify these specific elements rather than simply introducing technology as a solution, we can create intelligent systems that genuinely serve vulnerable populations and reduce, rather than reinforce, existing inequalities.