Artificial intelligence (AI) is increasingly integrating into everyday life, with significant potential to transform healthcare and scientific research. These fields offer unique opportunities for societal benefit—including improved health outcomes, climate adaptation, and accelerated scientific discovery—but successful integration of AI requires a sociotechnical approach that takes into account human norms and practices. This thesis explores how AI can be designed and deployed in human-centered and inclusive ways, particularly for historically marginalized communities. Through three case studies, I investigate AI’s role in real-world settings: (1) examining how Black older adults socially relate to an AI voice assistant for home exercise, (2) analyzing the adoption of the first generative AI chatbot in a U.S. national laboratory from an organizational perspective, and (3) developing a participatory AI system for climate adaptation that integrates community-driven data. This work contributes to understanding how to design effective and inclusive AI systems in science and healthcare settings.