Transparency, Explainability, Security

This unit examines transparency and explainability as core components of trustworthy AI, and explores their limits, stakeholder implications, and practical steps to operationalize them. It also introduces how AI systems can be compromised and exploited so students can connect explainability and robustness to security and governance practices.

Learning Objectives

• analyze trade-offs between transparency, privacy, and reliability in AI systems and identify the impacts of transparency decisions on different stakeholders
• apply technical and procedural approaches (e.g., model cards, feature attribution, documentation practices) to assess and improve the explainability of AI components
• demonstrate methods to identify common AI compromise vectors and describe how insecure or poorly explained models can be exploited in real-world scenarios
• evaluate organizational processes and leading practices for operationalizing explainability and transparency and recommend targeted improvements to policy, documentation, or workflows