Beyond Procedural Compliance: Human Oversight as a Dimension of Well-being Efficacy in AI Governance

Human oversight requires more than following rules. Current AI regulations like the EU AI Act mandate human oversight but don't explain what it actually means or how to develop it. This paper argues oversight is a learnable skill—a "well-being capacity"—rather than just a compliance checkbox, requiring people to understand AI, make ethical judgments, and recognize human needs.

People's biases and desires get built into AI systems. The paper emphasizes that humans inevitably project their own fears, interests, and desires onto AI systems. Effective oversight means developing the competence to recognize when our demands of AI might be problematic and having the ability to restrain them—a critical but often overlooked responsibility.

AI literacy alone isn't enough for good oversight. The framework integrates three elements: AI literacy (understanding how systems work), ethical discernment (making value judgments), and awareness of human needs (including recognizing when needs conflict or could cause harm). This multifaceted approach suggests that technical knowledge without ethical reasoning is insufficient.

Education at all levels is essential for sustainable AI governance. Rather than relying on specialized experts alone, the authors argue that developing human oversight capacity should be integrated into professional training, school curricula, and lifelong learning programs. This democratization of AI oversight responsibility is presented as both more sustainable and cost-effective than top-down approaches.

This framework bridges the gap between policy and practice. By reframing human oversight as a developable capacity, the paper offers a practical pathway for translating high-level regulatory goals into concrete educational and training programs that cultivate human agency and responsibility.