[Paper Review] A Framework for Fairness: A Systematic Review of Existing Fair AI Solutions

Brianna Richardson, Juan E. Gilbert|arXiv (Cornell University)|Dec 10, 2021

Ethics and Social Impacts of AI30 citations

TL;DR

This paper systematically reviews algorithmic bias, surveys the fairness solution space (toolkits and checklists), analyzes practical shortcomings, and offers recommendations to bridge researchers and practitioners.

ABSTRACT

In a world of daily emerging scientific inquisition and discovery, the prolific launch of machine learning across industries comes to little surprise for those familiar with the potential of ML. Neither so should the congruent expansion of ethics-focused research that emerged as a response to issues of bias and unfairness that stemmed from those very same applications. Fairness research, which focuses on techniques to combat algorithmic bias, is now more supported than ever before. A large portion of fairness research has gone to producing tools that machine learning practitioners can use to audit for bias while designing their algorithms. Nonetheless, there is a lack of application of these fairness solutions in practice. This systematic review provides an in-depth summary of the algorithmic bias issues that have been defined and the fairness solution space that has been proposed. Moreover, this review provides an in-depth breakdown of the caveats to the solution space that have arisen since their release and a taxonomy of needs that have been proposed by machine learning practitioners, fairness researchers, and institutional stakeholders. These needs have been organized and addressed to the parties most influential to their implementation, which includes fairness researchers, organizations that produce ML algorithms, and the machine learning practitioners themselves. These findings can be used in the future to bridge the gap between practitioners and fairness experts and inform the creation of usable fair ML toolkits.

Motivation & Objective

• Define and categorize algorithmic bias sources across the ML pipeline (pre-existing, technical, emerging).
• Survey existing fairness solution spaces, including software toolkits and checklists.
• Assess how fairness tools are used in practice and identify caveats and design flaws.
• Provide recommendations to align fairness research with industry needs and practitioner practices.

Proposed method

• Literature synthesis and taxonomy development of bias types (pre-existing, technical, emerging).
• Catalog and describe major fairness toolkits and checklists from industry and academia.
• Analyze practicality, usability, and design gaps of fairness solutions.
• Discuss caveats, such as metric conflicts, ethics washing, and socio-technical considerations.
• Propose actionable recommendations for fairness researchers, organizations, and ML practitioners.

Experimental results

Research questions

• RQ1 What are the major entry points and forms of algorithmic bias driving fairness concerns?
• RQ2 What fairness toolkits and checklists exist, and what features do they offer or lack?
• RQ3 How are current fairness solutions performing in practice, and what are their caveats?
• RQ4 What recommendations can bridge the gap between fairness research and industry practice?

Key findings

• Identifies a taxonomy of biases: pre-existing, technical, and emerging (deployment) biases.
• Documents a diverse solution space dominated by software toolkits and lifecycle checklists.
• Highlights a gap between fairness researchers and ML practitioners in real-world application.
• Notes challenges such as conflicting fairness metrics, robustness issues, and the ethics/technical balance.
• Recommends human-centered design and domain-specific, usable fair AI toolkits to improve adoption.