Human-Centered Artificial Intelligence for Healthcare, Education, Business, and Assistive Technologies

Abstract

Artificial intelligence increasingly mediates consequential human decisions across healthcare, education, business, assistive technologies, digital health, and organizational management. Yet the dominant evaluation paradigm—predictive accuracy—inadequately captures the properties most relevant to human welfare: interpretability, usability, accessibility, fairness, privacy, and the quality of human-AI collaboration. Human-centered AI reframes AI development as a socio-technical design discipline in which model outputs must support—rather than supplant—human expertise, judgment, and agency. This structured critical review synthesizes an eight-axis human-centered taxonomy covering domain, function, data modality, architecture family, design concern, deployment pathway, evidence role, and evidence maturity. Seven human-centered domains are examined: healthcare and biomedical decision support, education and adaptive learning, assistive technologies and accessibility, neuro-affective and mental-health AI, business and organizational decision-making, human-facing IoT and smart infrastructure, and cybersecurity and trustworthy digital systems. Synthesis reveals that while AI architectures have advanced substantially across vision transformers, hybrid ensembles, graph neural networks, and federated systems, human-centered design properties—validated explainability, inclusive design, privacy-preserving deployment, affective ambiguity management, and governance accountability—remain inconsistently addressed. An eleven-direction research agenda emphasizes human-centered evaluation, user-driven validation, fairness-aware design, accessible deployment, and governance-ready AI systems across all critical domains.