Drivers and Inhibitors for the Adoption of Privacy Preserving Machine Learning in Organizations : Critical Information Infrastructures, Winter Term 21/22

Background: Ever-improving machine learning techniques are leading to its widespread use, bringing benefits to organizations deploying them. The technology has now reached all sectors, leading to mainstream use. Simultaneously, a wide discussion about information security and privacy in machine learning developed, expressing concerns about data access, data storage and data usage. To address these concerns, the field of privacy preserving machine learning (PPML) emerged in recent years.
Objective: Since this research field is still in its early stages, scientific PPML contributions to date have mainly focused on technical aspects and have hardly looked at its implementation and diffusion within organizations. This paper will provide insights to explore the opportunities and risks of using PPML from an organizational perspective.
Methods: This work is based on an explorative and qualitative research approach in the form of semi-structured interviews with experts.
Results: We conducted interviews with 5 experts from the industry. We identified 16 factors, 8 of which drive adoption and 8 of which inhibit it. These factors were divided into technical, organizational, and legal. ... mehrIt can be shown that PPML solutions are not yet widely diffused in the market and that policy makers are lacking awareness about the technology. Increased computational costs, performance losses, and financial uncertainty are major inhibitors of PPML. In contrast, security and privacy guarantees, as well as privacy-compliant collaboration, are important drivers of technology.
Conclusion: We provided an insight into the current state of developments in research, politics and industry. The technical factors are highly complex, but are already receiving attention in research. Organizational aspects are currently neglected, while a heterogeneous legal framework irritates organizations.