The Societal Impacts of Algorithmic Decision-Making/ Manish Raghavan.

Format:

Book

Author/Creator:

Raghavan, Manish, author.

Series:

ACM books - Collection 3 ; #53.

ACM books, 2374-6777 ; #53

Language:

English

Subjects (All):

The Societal Impacts of Algorithmic Decision-Making (Computer Science).

Genre:

Electronic books.

Physical Description:

1 online resource (xxiv, 340pages) LuaTEX

Edition:

First Edition

Place of Publication:

[New York, NY, USA] : Association for Computing Machinery; [2023].

System Details:

Mode of access: World Wide Web

System requirements: Adobe Acrobat Reader

Contents:

Introduction

Overview of the Book

PART I THEORETICAL FOUNDATIONS FOR FAIRNESS IN ALGORITHMIC DECISION-MAKING

Overview of Part I

1 Inherent Tradeoffs in the Fair Determination of Risk Scores

1.1 A Formal Model of Risk Assessment

1.2 The Characterization Theorems

1.3 The Approximate Theorem

1.4 Reducing Loss with Equal Base Rates

1.5 Conclusion

2 On Fairness and Calibration

2.1 Problem Setup

2.2 Relaxing Equalized Odds to Preserve Calibration

2.3 Experiments

2.4 Discussion and Conclusion

3 The Externalities of Exploration and How Data Diversity Helps Exploitation

3.1 Preliminaries

3.2 Group Externality of Exploration

3.3 Greedy Algorithms and LinUCB with Perturbed Contexts

3.4 Analysis: LinUCB with Perturbed Contexts

3.5 Analysis: Greedy Algorithms with Perturbed Contexts

PART II MODELS OF BEHAVIOR

Overview of Part II

4 Selection Problems in the Presence of Implicit Bias

4.1 Overview and Summary of Results

4.2 Biased Selection with Power Law Distributions

4.3 Biased Selection with Bounded Distributions

4.4 Conclusion

5 How Do Classifiers Induce Agents to Behave Strategically?

5.1 Model and Overview of Results

5.2 Incentivizing Particular Effort Profiles

5.3 Optimizing Other Objectives

5.4 The Structure of the Space of Linear Mechanisms

5.5 Conclusion

6 Algorithmic Monoculture and Social Welfare

6.1 Algorithmic Hiring as a Case Study

6.2 Instantiating with Ranking Models

6.3 Models with Multiple Firms

6.4 Conclusion

PART III APPLICATION DOMAINS

Overview of Part III

7 Mitigating Bias in Algorithmic Hiring: Evaluating Claims and Practices

7.1 Background

7.2 Empirical Findings

7.3 Analysis of Technical Concerns

7.4 Algorithmic De-biasing

7.5 Discussion and Recommendations

8 The Hidden Assumptions Behind Counterfactual Explanations and Principal Reasons

8.1 What Are Feature-highlighting Explanations?

8.2 Feature-highlighting Explanations in Practice

8.3 Unavoidable Tensions

8.4 Conclusion

PART IV CONCLUSION AND FUTURE WORK

9 Future Directions

9.1 Fairness in Machine Learning and Mechanism Design

9.2 Algorithmic Discrimination

9.3 Transparent and Meaningful Explanations

PART V APPENDICES

10 Appendix A Inherent Tradeoffs in the Fair Determination of Risk Scores

A.1 NP-completeness of Non-trivial Integral Fair Risk Assignments

11 Appendix B On Fairness and Calibration

B.1 Linearity of Calibrated Classifiers

B.2 Cost Functions

B.3 Relationship Between Cost and Error

B.4 Proof of Algorithm 2.1 Optimality and Approximate Optimality

B.5 Proof of Impossibility and Approximate Impossibility

B.6 Details on Experiments

12 Appendix C The Externalities of Exploration and How Data Diversity Helps Exploitation

C.1 (Sub)gaussians and Concentration

C.2 KL-divergence

C.3 Linear Algebra

C.4 Logarithms

13 Appendix D Selection Problems in the Presence of Implicit Bias

D.1 Missing Proofs for Section 4.2

D.2 Additional Theorems for Power Laws

D.3 Lemmas for the Equivalence Definition

D.4 Lemmas for Section D.2

D.5 Lemmas and Proofs for Section 4.3

14 Appendix E How Do Classifiers Induce Agents to Behave Strategically?

E.1 Characterizing the Agent's Response to a Linear Mechanism

15 Appendix F Algorithmic Monoculture and Social Welfare

F.1 Random Utility Models Satisfying Definition 6.1

F.2 3-candidate RUM Counterexamples

F.3 Proof of Theorem 6.2

F.4 Verifying that the Mallows Model Satisfies Definition 6.1

F.5 Proof of Theorem 6.3

F.6 Supplementary Lemmas for the Mallows Model

16 Appendix G Mitigating Bias in Algorithmic Decision-making: Evaluating Claims and Practices

G.1 Administrative Information on Vendors

Bibliography

Authors' Biographies

Index

Other Format:

Print version:

ISBN:

3603195

9798400708626

9798400708602

Access Restriction:

Restricted for use by site license.