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

An empirical study based on interviews with experts across industry, academia, civil society, and government reveals that Sociotechnical Outcome Taxonomies (SOTs), designed to classify diverse AI risks, are currently weakly integrated into practical AI governance processes. This weak integration is attributed to specific features of SOT design and usage.

Why it matters

The effective classification of AI risks is critical for developing robust governance frameworks and mitigating potential harms across various sectors. The current weak integration of existing risk taxonomies poses a significant challenge to proactive risk management and the responsible deployment of AI technologies.

Key insights

• AI is implicated in a broad spectrum of harms, ranging from immediate user interactions to extensive societal consequences.
• The classification of these diverse AI risks is fundamental to effective AI governance, informing regulators, technology firms, and policymakers.
• Numerous Sociotechnical Outcome Taxonomies (SOTs) have been developed by researchers and practitioners to provide structured accounts of AI risk.
• Despite their development, SOTs are found to be weakly integrated into existing AI governance processes.
• The study identifies two specific features of SOT design and use that contribute to this weak integration.

Source

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