Aligning AI with human values: Design principles for human-centered AI

Human-Centered Artificial Intelligence (HCAI) is gaining traction as organizations and policymakers strive to align AI systems with human values and societal needs. While academic research has articulated key HCAI dimensions such as explainability and fairness, industry has operationalized these concepts in practice-oriented guidelines (e.g., Microsoft HAX, Google’s People + AI Guidebook, IBM’s Design for AI Toolkit). Yet a consolidated, research-driven synthesis is still missing. Our study responds to this gap by systematically drawing on peer-reviewed scholarship to distil a set of actionable design principles for AI applications. We conducted a systematic literature review across four major databases (Scopus, Web of Science, IEEE Xplore, ACM), yielding in 178 peer-reviewed papers. From these papers, we inductively coded relevant requirements and principles. This process revealed three core requirements: (1) user perception, (2) functional, and (3) ethical requirements with 22 subdimensions. Thereafter, we conducted 11 semi-structured interviews with industry practitioners and academic experts to critique, refine, and prioritise the identified principles. Using, thematic analysis of the interview data, we mapped expert feedback to the three clusters, revealing both alignment and tension between academic ideals and practical realities. The resulting framework contains 27 design principles anchored in verifiable requirements, which we, based on our interview data, organized in a prioritization matrix. To implement such principles, we also discuss approaches such as interactive explanation, human-in-the-loop governance and value-oriented transparency demonstrating how developers can translate abstract principles into concrete design decisions. By embedding our principles into existing software and MLOps lifecycles, firms can accelerate responsible AI delivery, enhance user trust and show readiness for audits. For researchers, the results contain a consolidated vocabulary and an empirically grounded agenda for evaluating HCAI outcomes.