Pattern-based AI Risk Assessment: A Taxonomy Expansion Use Case

As artificial intelligence (AI) is increasingly integrated into systems deployed in a wide range of application domains, the need to assess and mitigate the risks of these systems in diverse contexts has become a critical concern. Existing frameworks and methodologies for AI risk assessment support this process, but they often only provide general guidance disconnected from technical decisions. Furthermore, when an AI-based system is deployed in a new application context, they typically require a complete reassessment from scratch, which is a labour-intensive process that may miss potentially relevant risks. To tackle this challenge, this paper suggests a pattern-based approach to AI risk assessment that leverages semantic models of interlinked design and risk patterns to enable efficient and effective risk assessment across application contexts. We illustrate the effectiveness of our approach in a case study on a taxonomy expansion system in (i) a medical diagnosis application, and (ii) an e-commerce recommender application context and demonstrate how abstract risk patterns can support both architectural design decisions on the system level and structured risk assessments in a given application context. Our initial experiences suggest that the approach offers a promising and scalable method for assessing risks across application contexts.