Identifying bias in human and machine decisions
Abstract/Contents
- Abstract
- Many suspect that decisions in employment, college admissions, lending, and the criminal justice system are systematically biased against certain groups. To date, efforts to provide statistical support for these concerns are often hampered by subtle statistical flaws that are known to affect popular tests for discrimination. The first part of this dissertation introduces these tests and their flaws, before proposing two new statistical tests for discrimination. The threshold test uses a hierarchical Bayesian model to identify the standards that decision-makers apply to different protected groups, while risk-adjusted regression aims to determine whether observed differences in treatment are justified by legitimate policy goals. Decisions aided by machine learning algorithms have also come under scrutiny for bias. As humans use machines to inform more and more consequential decisions, it is becoming crucial to understand how bias might occur in algorithmic systems. The study of identifying bias in algorithmic decisions is young, and many of the initial approaches suffer from the same statistical limitations as tests for bias in human decisions. The second part of this dissertation explores these approaches to algorithmic fairness, demonstrates their deficiencies, and provides a framework for the development of fair algorithmic decisions. Throughout this dissertation we apply our methods to large datasets from the US criminal justice system, studying citizens' interactions with the police and defendants' interactions with the courts. Despite this criminal justice focus, our work is applicable to studying bias in any domain, whether the decisions are made by humans or algorithms.
Description
Type of resource | text |
---|---|
Form | electronic resource; remote; computer; online resource |
Extent | 1 online resource. |
Place | California |
Place | [Stanford, California] |
Publisher | [Stanford University] |
Copyright date | 2018; ©2018 |
Publication date | 2018; 2018 |
Issuance | monographic |
Language | English |
Creators/Contributors
Author | Corbett-Davies, Samuel James |
---|---|
Degree supervisor | Bernstein, Michael S, 1984- |
Degree supervisor | Goel, Sharad, 1977- |
Thesis advisor | Bernstein, Michael S, 1984- |
Thesis advisor | Goel, Sharad, 1977- |
Thesis advisor | Jurafsky, Dan, 1962- |
Degree committee member | Jurafsky, Dan, 1962- |
Associated with | Stanford University, Computer Science Department. |
Subjects
Genre | Theses |
---|---|
Genre | Text |
Bibliographic information
Statement of responsibility | Samuel James Corbett-Davies. |
---|---|
Note | Submitted to the Computer Science Department. |
Thesis | Thesis Ph.D. Stanford University 2018. |
Location | electronic resource |
Access conditions
- Copyright
- © 2018 by Samuel James Corbett-Davies
Also listed in
Loading usage metrics...