Balancing privacy and explainability in AI: Differential privacy and graph theory as governance tools

Abstract

Anonymisation (particularly the differential privacy method) stands as the most valuable technique for safeguarding individuals’ privacy, especially in an organisational context. Combining graph theory with the differential privacy method ensures that while data remains protected, the explainability of artificial intelligence (AI) models is not compromised, thereby achieving the recommended state of explainable AI. This paper synthesises technical and regulatory analysis to tackle the problem of achieving an optimal relation between privacy protection in AI systems and explainability in computational intelligence, focusing specifically on anonymisation techniques. It concludes that while differential privacy effectively safeguards data subjects, its integration with graph theory can enhance the level of explainability in AI systems, making it a viable solution for AI developers and privacy practitioners. By analysing current regulatory frameworks, including the General Data Protection Regulation (GDPR) and the European Union (EU) AI Act, alongside practical anonymisation methodologies, the study demonstrates how an innovative combination of privacy-enhancing technologies (PETs) and graph theory can align with regulatory compliance and ensure the recommended level of explainability in AI systems. This article is also included in The Business & Management Collection which can be accessed at https://hstalks.com/business/.