AI AUDITS: BLACK BOX VS. WHITE BOX PERSPECTIVES
summary
Argues that GDPR Art.22/15 and Art.35 map onto concrete practices: counterfactual explanations (computeable, deployed in TensorBoard and Google Cloud, adopted by IBM and Vodafone) give actionable “what to change” answers without revealing source code; an algorithmic impact assessment or DPIA becomes the practical regulator’s tool (Canada offers an online DPIA, EU AI Act favors risk-based lists). Also stresses system-level audits — data provenance, versioning and monitoring — because deep‑learning models prove brittle in COVID‑19 image and other real‑world use cases.
AI-written summary · about
chapters (4)
- 0:00Start
- 8:17Multi-layered Explanation
- 42:43Impact Assessment can also be a tool for individual rights
- 48:27The AIA and EU institutions policy-making
related talks
01h 06mDatabases and AI Training
39mThe Search for Transparency and Accountability in the Age of...
24mTechnical Tracks
26mAssessing Risks: Perspectives on the Data Protection Impact
24mFrom Black-Box to Guard-Box in Financial AI Security
39mTrust Cards For AI
01h 11mThe Intersection of AI and Regulation How Organisations and Regulators Should...
08mWhy AI Governance Matters More Than Ever