Introduction to the Taxonomy
Central to the PRA for AI framework, the Aspect-Oriented Taxonomy of AI Hazards provides the essential structure for systematically exploring the hazard space and identifying consequential threats. This avoids reliance on ad-hoc lists, ensuring more comprehensive hazard discovery tailored to the AI system under study.
Developed from first principles and grounded in systems thinking, this taxonomy organizes potential risks around the four fundamental aspects as shown in the figure below: Capabilities, Domain Knowledge, Affordances, and Impact Domains. It conceptually links the system's inherent properties ('source aspects': Capabilities, Domain Knowledge, Affordances) to the contexts where harms ultimately manifest ('terminal aspects' within Impact Domains).
This structured approach guides the identification of specific 'aspect-adjacent hazards', whether viewing them as potential harms originating directly from or being enabled by the system's source aspects (Capabilities, Knowledge, Affordances) or as vulnerabilities within the terminal aspects (Impact Domains) through which risks manifest before final impact. This enables assessors using bottleneck and risk pathway analysis to systematically probe potential failure modes from both ends of the causal chain.
Taxonomy Structure
The taxonomy is organized hierarchically into the following Taxonomy Levels (TLs):
- TL0: Aspect Category - The highest-level classification, representing the primary dimensions of AI system analysis in sociotechnical context
- TL1: Aspect Group - Major subdivisions within each Aspect Category, providing a more granular framework for analysis
- TL2: Aspect - Specific elements or characteristics within each Aspect Group, offering detailed points of consideration for hazard identification
- TL3: Hazard Cluster - Groupings of related hazards that may span multiple Aspects, allowing for flexible categorization and cross-cutting analysis
- TL4: AI Hazard (Aspect-derived) - Individual, specific hazards derived from the analysis of various Aspects and their interactions, representing concrete hazards within the AI system's sociotechnical context
Taxonomy Overview
The taxonomy of aspect-oriented AI hazards is presented below. In this visualization:
- Dark-colored boxes represent Aspect Categories (TL0)
- Medium-colored boxes indicate Aspect Groups (TL1)
- Light-colored boxes show individual Aspects (TL2)
You can scroll through the visualization to explore the detailed structure and content of the taxonomy. The Hazard Clusters and Aspect-Derived Hazards will be released on this webpage soon.