Responsible AI

Responsible AI

Building AI systems that are fair, transparent, and accountable.

Learning Outcomes

By completing this topic, you will:

• Identify sources of bias in ML systems
• Apply fairness metrics and mitigation strategies
• Use SHAP for model explanations
• Design for accountability and transparency

Visual Guides

Fairness Metrics

SHAP Feature Importance

Ethics Assessment

Prerequisites

• Classification and Supervised Learning
• Understanding of model evaluation
• Basic ethics concepts

Key Concepts

Bias in ML Systems

Sources of unfairness:

• Data bias: Unrepresentative training data
• Algorithmic bias: Model amplifies existing patterns
• Measurement bias: Flawed outcome definitions

Fairness Metrics

• Demographic parity: Equal positive rates across groups
• Equalized odds: Equal TPR and FPR across groups
• Individual fairness: Similar people get similar predictions

Explainability with SHAP

• Feature importance at individual and global levels
• Additive explanations based on game theory
• Visualization of feature contributions

When to Use

Responsible AI practices are essential when:

• Decisions affect people’s lives
• Protected attributes are involved
• Regulatory compliance is required
• Building trust is important

Common Pitfalls

• Treating fairness as a one-time check
• Optimizing for single fairness metric
• Ignoring intersectionality
• Conflating correlation with discrimination
• Not involving stakeholders in design

(c) Joerg Osterrieder 2025