Demystifying AI: understanding stakeholder interactions in explaining how AI works

Abstract

This thesis addresses the interaction between consumers and companies in explaining how AI functions, and resolving two research questions. The first investigates how consumer and firm hopes for explainability of AI differ, and how these differences affect consumer trust and engagement. The second investigates how companies can most effectively convey the rationale of AI-based decisions to enhance transparency and establish consumer trust. To elicit the responses to these sorts of questions, the research uses a qualitative approach through conducting semi-structured consumer and company representative interviews followed by thematic analysis of the transcripts using NVivo. From the results, stark differences in stakeholder demands come through: consumers are more interested in offering clear, transparent descriptions that offer a sense of control and create trust, whereas company representatives are more focused on explainability from the domains of operational viabilities, regulatory compliance, and brand image. The paper highlights the importance of customized, context sensitive explanations to support heterogenous user requirements, as well as co-creation practices wherein users collaborate with providers to develop explanation features, practices that help align AI systems with user expectations and foster trust. Academically, the study makes a contribution to explainable AI (XAI) literature through employing a stakeholder interaction view, revealing explainability as a complex construct and integrating stakeholder-specific knowledge into models of trust in AI. Managerially, it offers organizational strategies that can be used to demystify AI, including employing multi-level and user-specific explanation interfaces, proactive transparency in AI communications, and involving users in explanation design, hence fostering increased trust in AI-based services.