Workbook

Explainable AI

This workbook provides a hands-on introduction to explainable, interpretable, and transparent artificial intelligence. Readers will learn how to understand, evaluate, explain, secure, and align modern AI systems.

Units

  1. 0

    Why Explainable AI?

    AI already filters spam, routes traffic, detects fraud, recommends what we watch, and helps people make consequential decisions. This unit asks what happens when the model is accurate for the wrong reason, who needs to understand it, and what useful understanding looks like in practice.

    →
  2. 1

    Interpretable Machine Learning

    Build models whose prediction process can be inspected directly, from regression coefficients and decision rules to prototype and monotonic neural networks.

    →
  3. 2

    Explainable Machine Learning

    Explain individual predictions, global behavior, deep neural networks, language-model pipelines, and the quality of the explanations themselves.

    →
  4. 3

    Mechanistic Interpretability

    Reverse engineer neural networks from concepts and circuit evidence through superposition, sparse autoencoders, feature steering, and large-scale model analysis.

    →
  5. 4

    Adversarial AI

    AI systems can be accurate on ordinary test data and still fail on inputs chosen to make them fail. In this unit, you will attack small systems, trace how the attacks work, compare defenses, and build a practical red-team plan.

    →
  6. 5

    AI Alignment

    How do we build AI systems that pursue intended goals, remain reliable as conditions change, and stay accountable to the people affected by them? This unit moves from reward design and preference learning to assurance, values, and governance.

    →
  7. 6

    Transparent AI

    Transparency turns AI systems into inspectable operations: people can see what data was used, how decisions were made, who is responsible, and what evidence exists when something goes wrong. This unit covers traceability, provenance, auditability, failure analysis, and transparency patterns for foundation models, RAG, agents, and multi-agent systems.

    →

About

This workbook was created and is maintained by Dr. Brinnae Bent at Duke University for AIPI 590: Emerging Trends in Explainable AI.


This course requires a basic understanding of data science, machine learning, and neural networks. For a refresher on Data Science fundamentals, check out my Data Science Workbook.
For a refresher on Deep Learning, check out my Deep Learning Workbook.


Licensed under CC BY-SA 4.0.