Towards responsible design of AI-augmented systems : applications for evidence-informed decision making

Abstract

This dissertation examines the responsible design of artificial intelligence (AI)-augmented systems for evidence-informed decision making in organizational and policy contexts. As AI, and in particular, generative AI and large language models (LLMs), increasingly permeate high-stakes analytical processes, critical questions emerge regarding reliability, trustworthiness, and the appropriate distribution of agency between human actors and computational systems. While AI offers unprecedented capabilities for scaling the detection and measurement of complex phenomena from unstructured data, its deployment raises significant concerns about bias, transparency, accountability, and the legitimacy of AI-generated knowledge claims. Through four empirical publications situated in sustainability and innovation contexts, the research in this dissertation demonstrates concrete applications of responsible AI (RAI) principles through human-in-the-loop (HITL) design. Three core principles for RAI implementation are articulated out of these results: HITL validation should be embedded at critical system junctures where analytical choices materially affect downstream results, transparent flows of data retrieval and output use should enable scrutiny and traceability, and systems should include deliberate alignment of AI-automated tasks with human interpretation and domain context. Methodologically, the dissertation provides concrete guidance for implementing RAI-augmented analysis in strategic management and policy research, demonstrating system designs that preserve human agency while extending analytical capabilities. Theoretically, it advances understanding of legitimacy, epistemic authority, and agency distribution in AI-enabled knowledge work. Practically, it offers organizations and policymakers empirically grounded insight for integrating AI into analytical processes that support high-stakes decision making while maintaining contextualization and accountability.