Katerina Katsamba and Waldemar Pfoertsch, University of Limassol (CIIM), Cyprus

Human-centered artificial intelligence is usually framed as a design intention: build AI that respects human agency, oversight and wellbeing. This paper argues that intention is not sufficient. AI systems are always deployed inside social conditions — professional hierarchies, institutional rules, time pressure, bias and conflicting stakeholder expectations — that shape whether human oversight is actually exercised. We introduce Socially Embedded Human-Centered AI as an extension of the Human Filter, the ethical mechanism through which humans interpret, question and evaluate AI outputs before they become consequential decisions. We map the three Human Filter functions — ethical sensitivity, value alignment and reflective elevation — against five embeddedness conditions, illustrate the framework across healthcare, law and business, and derive design implications for trustworthy, human-centered AI systems that remain accountable under real social pressure.

Human-Centered AI, Human Filter, Socially Embedded AI, Trustworthy AI, Human-AI Collaboration