Design, implement, and evaluate a virtue-based AI ethics framework for decision-making

Thesis Abstract

Design, implement, and evaluate a virtue-based AI ethics framework for decision-making addresses the critical need for normative guidance in autonomous and semi-autonomous systems operating under complex social and moral stakes. The study investigates how virtue ethics can inform decision-making processes in AI to promote trustworthy and morally coherent outcomes beyond rule-based or utilitarian approaches. The aim is to develop a practical framework that operationalizes virtue concepts—such as prudence, courage, temperance, and justice—within AI systems and to evaluate its impact on decision quality, stakeholder trust, and perceived fairness. Specific objectives include (1) to translate core virtue ethics tenets into computational requirements and decision policies; (2) to implement a modular software framework that integrates virtue-informed modules with existing AI governance mechanisms; (3) to design an evaluation protocol comprising technical performance metrics, ethical alignment indicators, and stakeholder perception measures; (4) to assess the framework’s effect on decision outcomes across domain scenarios including healthcare triage, financial lending, and automated hiring; and (5) to propose guidelines for deployment and governance that account for cultural variability and normative pluralism. The methodology adopts a mixed-methods design combining constructionist and pragmatist philosophical assumptions. The population comprises AI systems and human-in-the-loop processes in three domain simulations healthcare triage (n=150 simulated decisions), credit underwriting (n=200 applicants), and recruitment screening (n=180 applications). A purposive sampling approach selects scenarios with ethically salient trade-offs. Data collection instruments include (a) a virtue-informed decision engine implemented as an extension to a standard rule-based AI platform, (b) expert panels of 12 ethicists and domain professionals who provide iterative feedback, (c) standardized scenario scripts for user studies, and (d) surveys and interview guides to capture trust, perceived fairness, and acceptability. Validity and reliability are established through triangulation, pilot testing (pilot sample sizes of 20 healthcare decisions, 30 loan applications, and 25 recruitment cases), and inter-rater reliability checks (Cohen’s kappa > 0.80 for coding schemes). The primary data analysis employs a multi-layered approach (i) quantitative evaluation using regression analysis to examine the relationship between virtue-aligned decision policies and objective performance metrics (e.g., decision accuracy, error rates) as well as trust indices; (ii) multivariate ANOVA to compare differences across domains and user groups; (iii) thematic analysis of expert feedback and stakeholder interviews to extract normative themes and potential tensions; and (iv) model checking and sensitivity analyses to assess robustness under varying cultural and organizational contexts. The conceptual development draws on virtue ethics (Aristotelian prudence and moral virtue) integrated with contemporary AI governance theories, including value-alignment and corrigibility, and is guided by a formalized virtue taxonomy mapped to computational primitives (perceptual assessment, deliberative weighing, and action selection). A conceptual model situates virtue-informed modules within a governance layer, an ethical reasoner, and an outcome allocator, with feedback loops to ensure continuous alignment with stakeholder values. Key expected findings include that virtue-informed AI demonstrates improved ethical alignment in boundary cases, yielding higher perceived fairness and trust without measurable losses in task performance. The study anticipates domain-specific differences in acceptable virtue expressions, highlighting the role of cultural and organizational context in shaping virtue enactment. The contribution to knowledge comprises a concrete design blueprint for integrating virtue ethics into AI decision-making, an operationalized framework linking moral theory to computational mechanisms, and an empirical evaluation that informs governance practices for ethically sensitive AI deployment. The study concludes that virtue-based AI ethics frameworks can enhance normative alignment and stakeholder trust when integrated with explicit governance structures and transparent disclosure of virtue-based reasoning. Recommendations include adopting modular virtue-aware components in high-stakes AI systems, developing cultural adaptation protocols for virtue expression, and embedding ongoing ethical auditing to monitor virtue performance over time. Further research is advised to refine the virtue taxonomy, explore cross-cultural validity, and extend evaluation to real-world deployments with longitudinal monitoring.