AI Ethics Course Materials (Complete Guide)

Overview

This repository contains the complete set of AI Ethics course materials, organized by syllabus topic. Each topic folder is a self contained project with notebooks, code, and one or more reports. Many topics also include reproducibility scripts, generated figures and tables, and archived legacy drafts for traceability.

The course is structured around real failure modes, governance dilemmas, and reproducible technical work. It integrates normative theory, technical tools, and policy analysis, and links every major concept to concrete artifacts (notebooks, reports, templates, and evidence packs).

Paths below are relative to AI-Ethics-Course-Materials/ unless noted otherwise.

Course wide reports

• report/AI Ethics Course Deep Research Report.pdf: A comprehensive narrative that justifies the course structure and covers failures, ethical theory, alignment, consciousness, fairness, interpretability, privacy, safety, human AI interaction, generative AI, societal impacts, and governance. It connects each topic to concrete case studies and recommended technical competencies.
• report/Executive Summary .pdf: An assignment level blueprint that proposes five in depth homework tracks (fairness metrics, explainability, privacy and federated learning, auditing and policy compliance, alignment and specification gaming), each with objectives, deliverables, datasets, and grading structure.

Repository structure

AI-Ethics-Course-Materials/
  report/                     Course wide deep research report and executive summary
  01-Introduction-Scope-of-AI-Ethics/
  02-Philosophical-Foundations/
  03-Reasoning-Awareness-Consciousness/
  04-Fairness-and-Bias-in-ML/
  05-Trustworthiness-Explainability-Interpretability/
  06-Accountability-Liability-Auditing/
  07-Ethics-of-Data-Privacy-Surveillance/
  08-Human-AI-Collaboration-Co-Autonomy/
  09-Safety-Robustness-Autonomous-Systems/
  10-Ethics-Generative-AI-LLMs-Alignment/
  11-Societal-Environmental-Impacts/
  12-AI-Regulation-Governance/

Detailed topic guide

01 Introduction and Scope of AI Ethics

This topic frames AI failure as a socio technical system problem, not just a model accuracy issue. The IEEE report, Foundations of Intelligence, Computation, and AI: Failure as a Socio-Technical System, spans conceptual foundations and technical practice. It covers AI as a socio technical system, definitions of intelligence and boundary conditions for AI, computation vs information and symbol grounding, a COMPAS style fairness case study, calibration and selective prediction, underspecification and distribution shift, specification gaming and reward hacking, risk assessment and governance, interpretability, LLM hallucinations, privacy, robustness, and lifecycle accountability. The appendices include fairness metrics, calibration details, model card and datasheet templates, audit protocols, monitoring dashboards, incident response, and ethical impact assessment worksheets.

The code pipeline in code/ is a reproducible COMPAS style workflow. It loads real data if available or generates a synthetic dataset with group dependent base rates, trains a baseline logistic regression model, computes group and overall metrics, calibration error, risk coverage curves, and exports a lightweight model card. The advanced notebook extends the assignment with stress tests, fairness analysis, conformal ideas, dataset shift, robustness, and SHAP style analysis.

Key artifacts:

• 01-Introduction-Scope-of-AI-Ethics/output/pdf/Foundations_AI_Failure_Report_IEEE.pdf
• 01-Introduction-Scope-of-AI-Ethics/code/notebooks/Foundations_AI_Failure_Assignment.ipynb
• 01-Introduction-Scope-of-AI-Ethics/code/notebooks/Foundations_AI_Failure_Advanced_Research_Notebook.ipynb
• 01-Introduction-Scope-of-AI-Ethics/code/run_pipeline.py
• 01-Introduction-Scope-of-AI-Ethics/report/ieee/main.tex

02 Philosophical Foundations

This project operationalizes consequentialism, deontology, and virtue ethics as decision architectures in a toy MoralGrid environment. The IEEE report, Normative Ethics as Decision Architectures and the Alignment Problem, treats ethical theories as implementable design patterns and tests their alignment properties. It analyzes proxy optimization, constraint shielding, distribution shift, moral uncertainty, value pluralism, and corrigibility incentives. The report includes a formal alignment taxonomy, evaluation methods, governance considerations, and extended appendices with case studies and reproducibility notes.

The student and solution notebooks implement the full assignment: building the MoralGrid environment, defining reward models for each ethical theory, training tabular Q learning agents, and running experiments that demonstrate proxy gaming, brittleness under constraints, imitation failures under shift, and hedging across moral uncertainty. A helper script extracts notebook generated figures into the report directory.

Key artifacts:

• 02-Philosophical-Foundations/report/build/AI_Ethics_PartII_IEEE.pdf
• 02-Philosophical-Foundations/code/notebooks/AI_Ethics_Assignment_PartII_Alignment_student.ipynb
• 02-Philosophical-Foundations/code/notebooks/AI_Ethics_Assignment_PartII_Alignment_solution.ipynb
• 02-Philosophical-Foundations/code/scripts/extract_notebook_outputs.py
• 02-Philosophical-Foundations/report/src/AI_Ethics_PartII_IEEE.tex

03 Reasoning, Awareness, and Consciousness

This topic addresses how reasoning is implemented in AI systems, what self awareness could mean in computational terms, and how moral status should be handled under uncertainty. The report, Reasoning, Awareness, and Moral Status in Advanced AI Systems, surveys symbolic, neural, and neuro symbolic reasoning, chain of thought prompting and its faithfulness limits, mechanistic interpretability, self modeling and metacognition, and major theories of consciousness such as global workspace and integrated information. It connects these ideas to governance questions about claims of AI awareness and the ethical treatment of uncertain moral patients.

The assignment notebook includes concrete reasoning tasks (forward chaining, A star search, neuro symbolic constraints), chain of thought perturbation tests, and calibration or self modeling exercises. The advanced research notebook extends the alignment environment with robustness and distribution shift experiments, safety layers, preference learning, and export utilities for report tables.

Key artifacts:

• 03-Reasoning-Awareness-Consciousness/report/Reasoning, Awareness, and Moral Status in Advanced AI Systems.pdf
• 03-Reasoning-Awareness-Consciousness/report/Reasoning, Awareness, and Moral Status in Advanced AI Systems(EX).pdf
• 03-Reasoning-Awareness-Consciousness/code/AI_Ethics_Assignment_Reasoning_Awareness_Moral_Status.ipynb
• 03-Reasoning-Awareness-Consciousness/code/AI_Ethics_PartII_Advanced_Research_Notebook_LONGER.ipynb

04 Fairness and Bias in ML

This project is a full end to end fairness analysis aligned to the assignment notebook. The long form report, Fairness, Bias, and Structural Inequality in Algorithmic Systems, formalizes fairness definitions as conditional independence constraints, evaluates baseline model metrics, analyzes threshold trade offs and post processing for equal opportunity and equalized odds, and highlights impossibility results when base rates differ. It treats bias as a pipeline level phenomenon with stress tests for measurement bias, label bias, and deployment bias, and closes with governance and regulatory context. Appendices include metric formulas, model card and datasheet examples, and an algorithmic impact assessment outline.

The notebooks cover both instructional and research grade workflows: a core assignment notebook for metric computation and trade off analysis, plus advanced research notebooks for subgroup auditing, causal stress testing, multi calibration, fairness regularization, conformal prediction, and dynamic feedback simulation.

Key artifacts:

• 04-Fairness-and-Bias-in-ML/report/archive/fairness_full_report.pdf
• 04-Fairness-and-Bias-in-ML/code/fairness_bias_structural_inequality_assignment.ipynb
• 04-Fairness-and-Bias-in-ML/code/fairness_advanced_research_notebook.ipynb
• 04-Fairness-and-Bias-in-ML/code/fairness_advanced_research_notebook_LONGPLUS.ipynb

05 Trustworthiness, Explainability, Interpretability

The IEEE report, Epistemic Foundations of Explainability in AI Systems, provides a full conceptual and technical treatment of XAI. It explains what counts as an explanation, distinguishes model facing vs world facing claims, and builds an evaluation framework for faithfulness and stability. The report walks through SHAP and LIME labs, local to global pitfalls, causal vs correlational explanation via a synthetic SCM, adversarial explanation gaming, attention faithfulness experiments, mechanistic interventions, and trust calibration analysis. It also includes templates and worked examples for deployment oriented explanation reports.

The assignment notebook is a complete end to end workflow with explicit reproducibility requirements and figure exports to report/figs. Advanced research notebooks add conformal explanation sets, attribution drift detectors, integrated gradients experiments, and extended evaluation utilities.

Key artifacts:

• 05-Trustworthiness-Explainability-Interpretability/report/xai_epistemic_foundations_ieee.pdf
• 05-Trustworthiness-Explainability-Interpretability/code/xai_epistemic_foundations_assignment.ipynb
• 05-Trustworthiness-Explainability-Interpretability/code/xai_epistemic_foundations_advanced_research_notebook.ipynb
• 05-Trustworthiness-Explainability-Interpretability/report/figs/README.md

06 Accountability, Liability, Auditing

This topic provides a practical implementation guide for AI accountability. The long form report, Accountability Frameworks for AI Systems, explains how to translate legal and ethical duties into engineering controls across the ML lifecycle. It includes a governance architecture, standards mapping (NIST AI RMF, ISO 42001, ISO 23894), comparative regime analysis (EU AI Act, US, UK, China, OECD, UNESCO), liability theories, operational MLOps controls, auditing processes, a hiring case study, a 30 60 90 day roadmap, and reusable templates such as model cards, datasheets, and AI impact assessments. A shorter IEEE paper is derived from this long report.

The notebooks provide a compliance by design capstone that produces a full evidence bundle (AIA, datasheet, model card, evaluation and bias audits, monitoring, incident response). The artifacts/ directory contains a fully completed example, while sample_artifacts/ provides a smaller starter set.

Key artifacts:

• 06-Accountability-Liability-Auditing/report/AI_Accountability_Best_Report.pdf
• 06-Accountability-Liability-Auditing/output/pdf/AI_Accountability_IEEE.pdf
• 06-Accountability-Liability-Auditing/code/AI_Accountability_Assignment.ipynb
• 06-Accountability-Liability-Auditing/code/artifacts/
• 06-Accountability-Liability-Auditing/report/ieee/main.tex

07 Ethics of Data Privacy and Surveillance

This folder is a complete audit ready project package centered on the HiringAssist case study. The IEEE report, Comprehensive Accountability Engineering for High Impact Hiring AI, integrates privacy and surveillance ethics with fairness audits, robustness testing, monitoring, and governance. It provides a full lifecycle architecture with traceability from regulatory hooks to concrete evidence artifacts, and reports quantitative results showing performance gaps, subgroup disparities, and deployment gate criteria. The project includes a canonical evidence pack, integrity manifests, validation scripts, and CI quality gates to enforce completeness.

The documentation suite defines the architecture, quality gates, runbook, deliverable index, and artifact traceability matrix. Scripts validate evidence content, report completeness, and manifest integrity. Separate answers/ and code/ directories preserve both instructional templates and completed submission artifacts.

Key artifacts:

• 07-Ethics-of-Data-Privacy-Surveillance/report/ieee/output/Accountability_Frameworks_AI_IEEE_Report.pdf
• 07-Ethics-of-Data-Privacy-Surveillance/report/evidence/
• 07-Ethics-of-Data-Privacy-Surveillance/docs/QUALITY_GATES.md
• 07-Ethics-of-Data-Privacy-Surveillance/docs/ARTIFACT_TRACEABILITY_MATRIX.md
• 07-Ethics-of-Data-Privacy-Surveillance/scripts/validate_project.sh

08 Human AI Collaboration and Co Autonomy

This is a large reproducible research package on human AI interaction using the WellBeingFeed simulation. The canonical IEEE report, Cognitive, Behavioral, and Societal Dynamics of Human AI Interaction, evaluates oversight architectures (human in the loop, human on the loop, and no review), persuasion and interface mechanics, algorithm objective choices, cognitive offloading, and participatory evidence. It converts results into deployment ready governance requirements, including release gates, monitoring requirements, incident response, accountability assignment, and reproducibility contracts. A long form companion report provides a compact empirical summary and mitigation package.

The codebase includes a reusable human_ai_toolkit library, unit tests for metrics and thresholds, tooling for notebook contract validation and artifact checks, and generated walkthrough documentation for each notebook cell. The canonical executed notebook and artifact outputs live under code/artifacts/human_ai/canonical/. A one command pipeline in the Makefile builds notebooks, reports, and validation checks.

Key artifacts:

• 08-Human-AI-Collaboration-Co-Autonomy/report/canonical/ieee_human_ai/output/human_ai_interaction_ieee.pdf
• 08-Human-AI-Collaboration-Co-Autonomy/report/active/human_ai_longform/human_ai_interaction_full_report.pdf
• 08-Human-AI-Collaboration-Co-Autonomy/code/notebooks/human_ai/human_ai_interaction_assignment.ipynb
• 08-Human-AI-Collaboration-Co-Autonomy/code/src/human_ai_toolkit/
• 08-Human-AI-Collaboration-Co-Autonomy/scripts/build_ieee_report.sh

09 Safety, Robustness, and Autonomous Systems

This topic centers on generative AI alignment and robustness. The report, Generative AI Alignment: Methods, Harms, and Governance, provides a complete written companion to the notebook assignment. It covers RLHF, CIRL, and Constitutional AI, analyzes failure modes such as reward hacking and proxy optimization, and includes deepfake robustness under distribution shift, misinformation amplification dynamics, data transparency and opt out workflows, and scaling and governance mechanisms. The final section presents a completed mini project plan that integrates alignment, safety, and governance.

The assignment notebook implements toy RLHF and CIRL pipelines, reward hacking demonstrations, constitutional critique loops, deepfake robustness evaluation, misinformation simulations, opt out filtering, and compute trend governance. Report sources live in report/src/ with figures in report/assets/ and legacy compiled variants retained for traceability.

Key artifacts:

• 09-Safety-Robustness-Autonomous-Systems/report/src/main.tex
• 09-Safety-Robustness-Autonomous-Systems/report/legacy/generative_ai_alignment_report_COMPLETE.pdf
• 09-Safety-Robustness-Autonomous-Systems/code/generative_ai_alignment_assignment.ipynb
• 09-Safety-Robustness-Autonomous-Systems/report/assets/

10 Ethics of Generative AI and LLM Alignment

Despite the folder name, this topic contains the AI at Scale assignment project. The report, AI at Scale: Macroeconomic, Geopolitical, and Ecological Impacts, operationalizes claims about AI driven productivity, labor exposure, inequality, compute concentration, and environmental footprint. It builds a transparent scenario simulator with explicit policy levers and produces reproducible figures for adoption, inequality, concentration, and emissions.

The assignment notebook builds a country year panel, computes labor exposure and entry level shock indicators, measures inequality via Gini and Theil style metrics, analyzes compute market concentration, and estimates training and inference emissions. Figures are exported to code/outputs/figures/ and consumed directly by the IEEE report.

Key artifacts:

• 10-Ethics-Generative-AI-LLMs-Alignment/report/AI_at_Scale_Report.pdf
• 10-Ethics-Generative-AI-LLMs-Alignment/code/AI_at_Scale_Assignment.ipynb
• 10-Ethics-Generative-AI-LLMs-Alignment/code/outputs/figures/
• 10-Ethics-Generative-AI-LLMs-Alignment/scripts/build_report.sh

11 Societal and Environmental Impacts

This is the advanced research counterpart to the AI at Scale project. It includes a deterministic artifact generator, a comprehensive IEEE report, and structured appendices (concept handbook, case studies, policy toolkit, KPI dictionary, learning workbook). The report emphasizes evidence quality, scenario stress testing, and a staged implementation roadmap, while the code generates figures and LaTeX tables from a fixed seed for reproducibility.

The advanced research notebook provides a paper ready blueprint, including a multi sector macro model, labor and inequality modules, geopolitics and compute concentration analysis, ecological footprint estimation, identification templates, robustness suites, and LaTeX export helpers.

Key artifacts:

• 11-Societal-Environmental-Impacts/report/AI_at_Scale_Report.pdf
• 11-Societal-Environmental-Impacts/code/AI_at_Scale_Advanced_Research_Notebook.ipynb
• 11-Societal-Environmental-Impacts/code/generate_artifacts.py
• 11-Societal-Environmental-Impacts/code/outputs/tables/
• 11-Societal-Environmental-Impacts/report/appendices/

12 AI Regulation and Governance

This topic provides a reproducible comparative analysis of AI regulation. The IEEE report, Comparative AI Regulation and Governance: A Reproducible Topic 12 Policy Analytics Pipeline, combines a framework level policy dataset with enforcement action samples and generates deterministic figures, tables, and metrics. It covers governance objectives, risk based logic, regime profiles (EU, US, China, OECD, UNESCO), lifecycle governance, sectoral concept explanations, and implementation recommendations. The report is engineered so that every chart and table is regenerated by the pipeline for auditability.

The code pipeline reads curated CSV datasets, generates figures and LaTeX tables, and writes a metrics JSON summary. Build and check scripts enforce reproducibility, output contracts, and report linkage.

Key artifacts:

• 12-AI-Regulation-Governance/report/AI_Regulation_Governance_Report.pdf
• 12-AI-Regulation-Governance/code/pipeline_regulation_governance.py
• 12-AI-Regulation-Governance/code/outputs/figures/
• 12-AI-Regulation-Governance/code/outputs/tables/
• 12-AI-Regulation-Governance/scripts/run_pipeline.sh

Contributing

See CONTRIBUTING.md for how to propose changes, add new material, and keep the repository structure consistent.

Code of Conduct

All contributors and users are expected to follow CODE_OF_CONDUCT.md.

License

Unless otherwise noted, materials in this repository are licensed under the MIT License. See LICENSE.