Explore the practical aspects of AI explainability in this 57-minute workshop from the Alan Turing Institute's AI UK 2022 series. Delve into four key principles, process-based vs. outcome-based explanations, and explanation-aware design. Examine model choice considerations through a case study, addressing issues like unconscious and statistical biases, bias mitigation, and historical bias. Investigate impact explanations, data subject dignity, and fairness explanations in AI algorithms. Discuss legal and ethical issues, problem formulation, social trust, and human bias in AI implementation. Conclude with insights on deep learning models and sufficient interpretability, providing a comprehensive overview of AI explainability challenges and solutions in real-world applications.
Overview
Syllabus
Introduction
Overview
Four Principles
Processbased vs Outcomebased explanation
Explanation aware design
Model choice considerations
Case study
Motivation
Case
Features
Mural
Link issues
Link working
Sharing the screen
Requesting video access
Unconscious biases
Statistical biases
Bias mitigation
Historical bias
Impact explanation
Data subject dignity
Recruitment activity
Fairness explanation
AI algorithm
Legal issues
Ethical issues
Problem formulation
Should we use technology
Social trust
Human bias
Implementation fairness
Bias
One last question
Deep learning models
Sufficient interpretability
Outro
Taught by
Alan Turing Institute
