Putting AI Ethics into Practice: The Hourglass Model of Organizational AI Governance

Ethics principles alone are insufficient: While many AI ethics principles have been proposed to address risks like bias and discrimination, the paper emphasizes that organizations need concrete processes and practices to actually implement these principles—highlighting a significant gap between stating ethical values and operationalizing them.

AI governance requires a multi-level approach: The hourglass model operates across three interconnected levels (environmental, organizational, and AI system levels), suggesting that responsible AI cannot be achieved through isolated efforts but requires coordination across an organization's entire structure and throughout an AI system's complete lifecycle.

Governance must span the entire AI system lifecycle: Rather than focusing on single points in development, the framework connects governance requirements to all stages of an AI system's life, implying that responsibility for managing risks like bias and discrimination must be maintained from initial design through deployment and beyond.

Regulatory alignment is becoming essential: The paper notes the framework is designed to help organizations align with the forthcoming European AI Act, indicating that responsible AI governance is increasingly becoming a legal and compliance requirement, not just an ethical aspiration.

Systemic coordination between stakeholders matters: The model highlights "the dynamics between the AI governance actors," suggesting that effective AI governance depends on clear communication and coordination among different teams and decision-makers within organizations, rather than siloed responsibility.