"Death by a thousand taxonomies?": AI Risk Classification In Practice

A recent study highlights that while the classification of AI risks through Sociotechnical Outcome Taxonomies (SOT) is considered fundamental for effective AI governance, these taxonomies are currently weakly integrated into practical AI governance processes. The research, based on 25 interviews across various sectors, attributes this deficiency to specific features of SOT design and their application.

Why it matters

The effective governance of Artificial Intelligence is critical for mitigating diverse risks and fostering responsible adoption. The weak integration of risk classification frameworks into AI governance processes indicates a significant gap that could lead to unaddressed risks and undermine regulatory efforts, impacting strategic planning and operational stability across sectors.

Key insights

• AI risks encompass a broad spectrum, from user interactions to societal-level impacts.
• Classification of AI risks is foundational for AI governance across regulators, firms, and policymakers.
• Numerous Sociotechnical Outcome Taxonomies (SOT) have been developed by researchers and practitioners.
• SOTs are currently weakly integrated into existing AI governance processes.
• The weak integration of SOTs is attributed to issues in their design and use, though further specifics are not detailed in the provided abstract.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2608.06831