Description

This curriculum engages learners in the same depth and structure as a multi-workshop organizational initiative addressing AI ethics, spanning operational, governance, and strategic levels across diverse technical and socio-political contexts.

Module 1: Defining Social Justice Frameworks in AI Development

Selecting normative justice models (e.g., Rawlsian, utilitarian, intersectional) to guide algorithmic fairness criteria in hiring systems.
Mapping marginalized community input into AI design requirements during stakeholder analysis for public sector deployments.
Deciding whether to prioritize distributive, procedural, or recognition justice in automated welfare eligibility systems.
Integrating international human rights standards into AI ethics charters for multinational corporations.
Resolving conflicts between privacy rights and transparency demands in community-led AI audits.
Establishing criteria for when algorithmic impact assessments must include historically oppressed groups as co-evaluators.
Designing feedback loops that allow affected populations to contest AI-driven decisions in real time.
Choosing between consensus-based and representative models for ethics review boards in AI projects.

Module 2: Bias Auditing and Mitigation at Scale

Implementing stratified evaluation metrics across intersectional demographic categories in credit scoring models.
Selecting between pre-processing, in-processing, and post-processing bias mitigation techniques based on system latency constraints.
Managing trade-offs between model accuracy and fairness when reweighting underrepresented classes in training data.
Documenting bias mitigation choices in model cards for regulatory compliance under EU AI Act requirements.
Conducting third-party adversarial testing to uncover emergent discriminatory patterns in multilingual NLP systems.
Deciding whether to exclude sensitive attributes (e.g., race, gender) or use them for active debiasing in risk assessment tools.
Designing continuous monitoring pipelines to detect bias drift in production systems with evolving user demographics.
Allocating budget for ongoing bias testing versus feature development in resource-constrained AI teams.

Module 3: Data Sovereignty and Inclusive Data Governance

Negotiating data licensing agreements with Indigenous communities using data trust frameworks.
Implementing differential privacy mechanisms while preserving statistical utility for low-population subgroups.
Establishing data access committees to enforce community-specific data usage restrictions in health AI projects.
Designing opt-in consent architectures that support granular control over data reuse in federated learning systems.
Choosing between centralized and decentralized data storage to balance security, compliance, and community control.
Enforcing data expiration policies for sensitive behavioral datasets collected from vulnerable populations.
Creating audit trails that log data access and transformations for accountability in cross-border AI collaborations.
Integrating local data protection laws (e.g., GDPR, POPIA) into global AI data pipelines with heterogeneous jurisdictions.

Module 4: Algorithmic Transparency and Explainability Trade-offs

Selecting explanation methods (e.g., SHAP, LIME, counterfactuals) based on end-user technical literacy in legal aid chatbots.
Deciding which model components to open-source versus protect as trade secrets in public interest AI applications.
Designing layered disclosure systems that provide simplified explanations to users and technical details to regulators.
Managing disclosure risks when explaining decisions could reveal training data membership or model vulnerabilities.
Implementing real-time explanation APIs without degrading system performance in high-throughput environments.
Validating the accuracy of explanations against ground-truth causal mechanisms in complex ensemble models.
Documenting known limitations of explainability methods in user-facing documentation for automated decision systems.
Allocating engineering resources to develop custom interpretability tools for domain-specific models.

Module 5: Power Distribution in AI Development Teams

Structuring team composition to include domain experts from historically excluded communities in core development roles.
Implementing decision rights matrices to clarify who can approve model changes affecting vulnerable populations.
Conducting power mapping exercises to identify and mitigate dominance patterns in AI design workshops.
Establishing escalation protocols for ethical concerns raised by junior team members without retaliation risk.
Rotating leadership roles in AI sprints to distribute influence and prevent epistemic gatekeeping.
Designing compensation structures that equitably value community knowledge contributions in co-design processes.
Enforcing inclusive meeting practices to ensure non-dominant language speakers can contribute to model specification.
Creating shadow review boards to evaluate whether project priorities reflect community needs or corporate interests.

Module 6: Regulatory Compliance and Cross-Jurisdictional Enforcement

Mapping conflicting AI regulations (e.g., EU AI Act, U.S. Executive Order, China’s algorithm rules) onto a single product.
Implementing geofencing and jurisdiction-aware routing to apply region-specific compliance rules in global AI services.
Designing audit-ready documentation systems that satisfy both GDPR and Algorithmic Accountability Act requirements.
Conducting gap analyses between self-regulatory AI ethics frameworks and enforceable legal standards.
Establishing compliance escalation paths for AI incidents involving multiple national regulators.
Allocating liability reserves for potential fines under high-risk AI classification systems.
Integrating real-time regulatory change monitoring into AI governance dashboards.
Deciding whether to withdraw AI services from jurisdictions with human rights-violating enforcement practices.

Module 7: Long-Term Alignment and Superintelligence Governance

Designing constitutional AI constraints that preserve human oversight in autonomous scientific discovery systems.
Implementing corrigibility mechanisms to allow shutdown or modification of recursive self-improving models.
Specifying value learning protocols that incorporate evolving societal norms into long-horizon AI planning.
Establishing multi-stakeholder review boards for approving capability thresholds in large-scale AI training runs.
Creating containment protocols for AI systems that exceed human-level performance in strategic reasoning.
Developing verification methods to ensure AI alignment claims are testable and falsifiable.
Allocating compute governance to prevent concentration of superintelligence development in unaccountable entities.
Designing international treaties for AI capability thresholds analogous to nuclear non-proliferation agreements.

Module 8: Community-Led AI and Decentralized Control Models

Implementing blockchain-based voting systems for community governance of shared AI models.
Designing data cooperatives that allow members to collectively negotiate AI partnership terms.
Deploying edge AI systems that process sensitive data locally to minimize centralized control.
Establishing community review panels with veto power over AI deployment in local public services.
Creating open governance forums with binding decision authority in municipal AI initiatives.
Developing technical interfaces that allow non-technical users to modify AI behavior rules.
Allocating model ownership shares to data contributors in participatory AI projects.
Building dispute resolution mechanisms for conflicts between community governance bodies and technical teams.

Module 9: Crisis Response and AI Harm Mitigation

Activating emergency rollback protocols for AI systems causing real-world harm during deployment.
Establishing rapid response teams with authority to suspend models without executive approval.
Designing compensation frameworks for individuals harmed by automated decision errors.
Implementing forensic logging to reconstruct AI decision chains during incident investigations.
Coordinating public communication strategies that acknowledge AI failures without triggering mass distrust.
Creating safe harbors for whistleblowers reporting AI risks within regulated industries.
Developing redress portals that allow affected users to appeal and receive remedies for AI harms.
Conducting root cause analyses that distinguish between technical failure, governance gaps, and design intent.