Design Principles and Measurement Frameworks for Human-Centered AI and Human-AI Interaction: A Systematic Review

Human-centered AI (HCAI) and human-AI interaction (HAX) are increasingly important as AI adoption expands across high-stakes domains. However, empirical evidence remains fragmented across disciplines, limiting the development of transferable design guidance. This systematic review synthesized empirical research on HCAI and HAX and was reported in accordance with PRISMA guidance. Eight databases were searched, and additional preprint records were identified through arXiv. These searches yielded 9,718 records, of which 44 articles were included in the final review. Findings were organized around three dimensions: outcome metrics, collaboration frameworks and AI roles, and domain-specific patterns. Across studies, AI support was often associated with improved objective task performance, including accuracy, task completion time, and decision quality. Subjective outcomes such as trust, satisfaction, usability, and workload varied with transparency, explanation format, cognitive demand, and perceived control. Results suggest that HCAI and HAX frameworks should support task performance while preserving human autonomy and manageable workload. They should also support appropriate trust and reflect domain-specific needs.