[Paper Review] Fair and Unbiased Algorithmic Decision Making: Current State and Future Challenges

Songül Tolan|arXiv (Cornell University)|Jan 15, 2019

Ethics and Social Impacts of AI41 citations

TL;DR

A critical overview of fairness in algorithmic decision making, highlighting the complexities of defining fairness, the limitations of statistical criteria, and the need for domain-specific constraints, transparency, and audits.

ABSTRACT

Machine learning algorithms are now frequently used in sensitive contexts that substantially affect the course of human lives, such as credit lending or criminal justice. This is driven by the idea that `objective' machines base their decisions solely on facts and remain unaffected by human cognitive biases, discriminatory tendencies or emotions. Yet, there is overwhelming evidence showing that algorithms can inherit or even perpetuate human biases in their decision making when they are based on data that contains biased human decisions. This has led to a call for fairness-aware machine learning. However, fairness is a complex concept which is also reflected in the attempts to formalize fairness for algorithmic decision making. Statistical formalizations of fairness lead to a long list of criteria that are each flawed (or harmful even) in different contexts. Moreover, inherent tradeoffs in these criteria make it impossible to unify them in one general framework. Thus, fairness constraints in algorithms have to be specific to the domains to which the algorithms are applied. In the future, research in algorithmic decision making systems should be aware of data and developer biases and add a focus on transparency to facilitate regular fairness audits.

Motivation & Objective

• Explain why machine learning in sensitive domains can inherit biases from data and decisions.
• Survey existing fairness criteria and their limitations across contexts.
• Argue that fairness constraints must be tailored to specific domains rather than unified into a single framework.
• Highlight the importance of transparency and regular fairness audits to address data and developer biases.

Proposed method

• Review and synthesize literature on fairness-aware machine learning and its formalizations.
• Discuss limitations and tradeoffs of statistical fairness criteria in different contexts.
• Present arguments for domain-specific fairness constraints and transparency-based governance.

Experimental results

Research questions

• RQ1 What are the limitations of standard statistical notions of fairness in various application domains?
• RQ2 How do data and developer biases impact algorithmic decision making, and how can transparency enable audits?
• RQ3 Why must fairness constraints be context-specific rather than universally unified across all settings?

Key findings

• Fairness criteria are numerous and often flawed, with tradeoffs that prevent a single universal framework.
• Algorithms can encode and amplify human biases when trained on biased data or decisions.
• Fairness constraints should be tailored to the application domain to be meaningful and effective.
• Future work should emphasize transparency to facilitate regular fairness audits.
• Awareness of both data and developer biases is essential for advancing fair decision making.