Ethics in Artificial Intelligence: Bias, Fairness and Beyond (Studies in Computational Intelligence, 1123)

✍ Scribed by Animesh Mukherjee (editor), Juhi Kulshrestha (editor), Abhijnan Chakraborty (editor), Srijan Kumar (editor)

• Publisher: Springer
• Year: 2024
• Tongue: English
• Leaves: 150
• Category: Library

✦ Synopsis

This book is a collection of chapters in the newly developing area of ethics in artificial intelligence. The book comprises chapters written by leading experts in this area which makes it a one of its kind collections. Some key features of the book are its unique combination of chapters on both theoretical and practical aspects of integrating ethics into artificial intelligence. The book touches upon all the important concepts in this area including bias, discrimination, fairness, and interpretability. Integral components can be broadly divided into two segments – the first segment includes empirical identification of biases, discrimination, and the ethical concerns thereof in impact assessment, advertising and personalization, computational social science, and information retrieval. The second segment includes operationalizing the notions of fairness, identifying the importance of fairness in allocation, clustering and time series problems, and applications of fairness in softwaretesting/debugging and in multi stakeholder platforms. This segment ends with a chapter on interpretability of machine learning models which is another very important and emerging topic in this area.

✦ Table of Contents

Foreword
Preface
Contents
About the Editors
Testing, Debugging, and Repairing Individual Discrimination in Machine Learning Models
1 Introduction
2 Background and Notation
2.1 Fairness
2.2 Decision Tree
3 Testing
3.1 Problem Setup
3.2 Maximizing Path Coverage
3.3 Maximizing Effectiveness of Discrimination Detection
4 Debugging
4.1 RID
4.2 Determining RIDs
4.3 On-Demand Sample Generation/Active Learning
5 Repairing
5.1 Setup
5.2 Objective
5.3 Iterative Algorithm
6 Experimental Results
6.1 Benchmarks
6.2 Setup and Configuration
6.3 Research Questions
6.4 Metrics
6.5 Results
7 Related Works
8 Conclusion
References
Group and Individual Fairness in Clustering Algorithms
1 Introduction
2 Preliminaries
3 Group Fairness
4 Individual Fairness
5 Relationships Between Fairness Levels and Their Notions
5.1 Relationship Between Group Fairness Notions
5.2 Relationship Between Individual Fair Notions
5.3 Relationship Between Group and Individual Fairness
6 Algorithms and Theoretical Guarantees
6.1 Group Fairness
6.2 Individual Fairness
6.3 Extension to Multiple Protected Attributes
6.4 Fair Algorithms Under Different Setting
6.5 Deep Fair Clustering
7 Discussion and Open Problems
References
Temporal Fairness in Online Decision-Making
1 Introduction
2 Fairness in Static Decision-Making
3 Ensuring Fairness in Dynamic Settings
3.1 Temporal Fairness and Memory
3.2 Temporal Fairness and Learnability
4 Example I: Partial Memory Comparative Fairness
5 Example II: Full Memory Relaxed Comparative Fairness
5.1 Algorithm Design for a Single Context
5.2 Algorithm Design for N Contexts
5.3 Improvements and Open Directions
6 Connections with Law and Policy
References
No AI After Auschwitz? Bridging AI and Memory Ethics in the Context of Information Retrieval of Genocide-Related Information
1 Introduction
2 AI-driven IR Systems and Genocide-Related Information
3 Memory Ethics and Human Curation of Genocide-Related Information
4 Bridging Memory Ethics and AI-driven IR System Design
5 Discussion
References
Algorithmic Fairness in Multi-stakeholder Platforms
1 Introduction
2 Algorithmic Fairness and Its Importance
3 Algorithmic Fairness in Online Platforms
4 The Case of Multi-stakeholder Platforms
4.1 Platforms with Two Types of Stakeholders
4.2 Platforms with Three or More Types of Stakeholders
5 Multi-stakeholder Fairness
6 Conclusion
References
Biases and Ethical Considerations for Machine Learning Pipelines in the Computational Social Sciences
1 Introduction
2 Dataset Creation and Collection Bias
2.1 Sampling Bias
2.2 Negative Set Bias
2.3 Label Bias
2.4 Apprehension Bias
3 ML Model and Data Analysis Bias
3.1 Confounding Bias
3.2 Chronological Bias
3.3 Algorithm Bias
4 Data and Model Evaluation Bias
4.1 Human Evaluation Bias
4.2 Validation and Test Set Bias
5 Responsible Research for CSS ML Pipelines
References
The Theory of Fair Allocation Under Structured Set Constraints
1 Introduction
2 Preliminaries and Notations
3 Matroid-Constrained Fair Allocation
4 Connectivity Constraints on Goods
5 Connectivity on Agents
6 Conclusion
References
Interpretability of Deep Neural Models
1 Introduction
2 Background and Related Work
3 Method
3.1 Feature Group Attribution Problem
3.2 Solution Axioms
3.3 Our Method: Integrated Directional Gradients
4 Evaluation
4.1 Setup
4.2 ``CHECKLIST'' Tests
4.3 Correctness
4.4 Capturing Semantic Interaction
5 Conclusion
References