The AI Evaluability Gap: The Missing Layer for Managing Risk and Sustaining Value

The article introduces the concept of the "AI Evaluability Gap," highlighting the lack of sufficient evidence for organizations to make high-confidence governance decisions regarding AI risk and value. It proposes a framework for "Evaluability," which ensures that AI systems can generate and maintain the necessary evidence over time, identifying six properties of evaluable evidence: observability, attributability, intervenability, verifiability, calibration, and temporal validity. This framework differentiates between Operational Certification and Investment Certification, emphasizing that addressing the Evaluability Gap is essential for effective AI governance and resource management in organizations.