Cadre théorique pour l’évaluation éthique des IA en entreprise

Thesis Abstract

In contemporary enterprises, the rapid deployment of artificial intelligence (AI) systems raises complex ethical challenges that threaten stakeholder trust, regulatory compliance, and organizational legitimacy. This study addresses the gap in theoretical tools to assess and govern the ethics of AI applications within corporate settings, focusing on developing a coherent framework that integrates normative ethics, governance mechanisms, and practical assessment procedures. The aim is to construct a theoretically grounded cadre capable of guiding both policy formulation and operational evaluation of AI ethics across diverse organizational contexts. Specific objectives are (1) to synthesize key ethical theories and governance principles relevant to corporate AI, (2) to identify dimensions of ethical risk and accountability that are salient for enterprise AI deployments, (3) to formulate a modular evaluative framework that maps ethical considerations to decision-making processes, data practices, and human–AI interaction, (4) to validate the framework through empirical application in a multi-sector sample, and (5) to delineate actionable indicators and thresholds for ongoing monitoring and improvement. The study adopts a mixed-methods design combining theory-driven framework development with empirical validation. The theoretical component builds on deontological and virtue ethics, utilitarianism, algorithmic fairness, transparency, and accountability theories, together with organizational governance models such as CSR, risk management, and ethics-by-design. The empirical component uses a purposive sample of 28 AI-intensive firms across finance, healthcare, manufacturing, and retail sectors, complemented by 12 interviews with ethics officers and AI engineers to illuminate governance practice. Data collection employs a structured survey instrument to quantify perceived ethical risk across five dimensions—data integrity, autonomy and control, fairness and non-discrimination, transparency and explainability, and accountability—and semi-structured interviews to capture contextual factors shaping ethical evaluation. Validity and reliability are ensured through pilot testing (n=62), Cronbach’s alpha evaluation (>0.80 for all scales), and triangulation of quantitative and qualitative data. The analysis proceeds in three stages (i) exploratory factor analysis to identify latent ethical risk dimensions; (ii) regression analysis to test associations between governance maturity (as a composite index) and ethical risk scores, controlling for firm size and sector; and (iii) thematic analysis of interview transcripts to refine the framework’s constructs and generate practical indicators. Model specification integrates structural equation modeling to assess the relationships among governance mechanisms, perceived ethical risk, and reported decision-making quality. The study also tests the robustness of the framework using sensitivity analyses across sector-specific subsamples. Expected findings include (a) a parsimonious set of core ethical risk dimensions that reliably predict governance needs, (b) evidential support for the positive impact of ethics-by-design and transparent data practices on perceived ethical compliance, (c) identification of sector-specific moderating effects, notably heightened sensitivity to fairness in healthcare and data governance in finance, and (d) a validated framework with modular components—ethics principles, governance processes, assessment indicators, and decision-support tools—that can be deployed adaptively. The contribution to knowledge lies in operationalizing an integrative theoretical model that connects normative ethics with corporate governance and practical evaluation of AI ethics, filling a methodological gap between abstract ethics discourse and day-to-day managerial practice. The framework advances theory by articulating how ethical theories translate into measurable organizational practices and how these, in turn, influence ethical outcomes in real-world AI deployments. The study concludes that a modular, theory-informed cadre—comprising ethics principles, governance mechanisms, and performance indicators—enables continuous ethical appraisal of AI systems and supports accountability across stakeholders. Recommendations include embedding the framework within corporate risk management, aligning it with regulatory reporting requirements, and investing in leadership training to cultivate an ethics-aware organizational culture. The paper also offers a roadmap for longitudinal validation, suggesting annual refresh cycles of indicators and periodic recalibration of the framework in response to regulatory developments, technological evolution, and shifting societal expectations.