The impact of artificial intelligence on data privacy: a risk management perspective

Abstract

Purpose The purpose of this paper is to increase the artificial intelligence (AI) ethical literacy of information governance professionals by explaining how AI intensifies familiar data privacy issues by virtue of its dependency on data and ability to create new personal information, to explicate emerging privacy enhancing methods and to show their continuity with existing privacy and information governance principles. Design/methodology/approach The paper uses an interdisciplinary design research methodology that extends current governance frameworks by combining information science, information governance and applied ethics concepts. Three information sources were referenced: 1) academic papers; 2) standards and best practices published by governmental and nongovernmental organizations and professional associations; and 3) white papers, market research and vendor reports. The literature review was informed by real-world implementation knowledge and anecdotal evidence to identify privacy risks when using AI. Useful tools, techniques and governance approaches to manage and mitigate the risks associated with digital privacy and ethics when using AI are identified and discussed. Findings The paper analyzes the relationship between different approaches to AI (e.g. symbolic-deductive, machine learning and deep learning) and levels of privacy risks. It identifies risk reduction methods (e.g. differential privacy) and relates these to extant privacy principles such as data minimization. Finally, the paper shows the continuity between information governance practices and newly emerging AI governance and risk frameworks. Originality/value The authors present useful tools and techniques and discuss them from a business perspective, using the lens of information governance to mitigate AI-related privacy risks. The authors also discuss how design techniques and technologies can help minimize data collection of sensitive information and can be used to anonymize sensitive data when training AI models.

Ähnliche Arbeiten