Abstract
This thesis develops and evaluates an evidence-based Ethical Risk Classification Framework (ERCF) for analyzing substantive ethical engagement in AI policy documents. The study addresses the gap between high-level AI ethics principles and reproducible computational methods for comparative governance analysis. A secondary dataset of AI ethics policy recordswas used as the starting point. After retrieval, text extraction, and validation, a corpus of 44 policy documents was constructed from an initial 112 records (110 link-bearing entries). Supervised classification was implemented for four core dimensions: Accountability, Bias & Fairness, Privacy, and Transparency. A TF-IDF + Logistic Regression baseline was evaluated using stratified 5-fold cross-validation and compared with SVM, BERT, and BiLSTM models. Unsupervised topic modeling (LDA) was applied to provide complementary thematic structure beyond predefined labels. Findings show that classification is feasible across all four dimensions, with strongest baseline performance on Transparency and comparatively lower performance on Privacy. Across this dataset, interpretable baseline modeling remained competitive and stable relative to more complex architectures. ERCF probability outputs enabled document-level and grouped comparative indicators, showing variation in ethical emphasis across sectors and country groupings. LDA outputs identified recurring governance, privacy, technical, and social-impact discourse patterns. The thesis contributes a transparent end-to-end pipeline, empirical evidence on ethicstopic variation in policy texts, and a practical scoring layer (ERCF) for governance-oriented interpretation. Results are bounded by sample size, archival temporal coverage, and the absence of external ground-truth labeled validation data; therefore, conclusions are interpreted within these scope constraints.
Publication Date
5-2026
Document Type
Thesis
Student Type
Graduate
Degree Name
Professional Studies (MS)
Department, Program, or Center
Graduate Programs & Research
Advisor
Ioannis Karamitsos
Recommended Citation
Almarri, Ahmad, "An Evidence-Based Ethical Risk Classification Framework (ERCF) for AI Policies" (2026). Thesis. Rochester Institute of Technology. Accessed from
https://repository.rit.edu/theses/12600
Campus
RIT Dubai
