Fair Enough? A map of the current limitations of the requirements to have fair algorithms

TL;DR

The paper argues that the demand for fair algorithms in automated decision-making is incomplete without addressing broader social, legal, and ethical choices. It critiques the reliance on purely observational fairness metrics and advocates incorporating causal reasoning to distinguish direct, indirect, and spurious discrimination, while acknowledging substantial limitations such as identifiability and the meaningful use of attributes as causal nodes. It surveys how protected attribute identification, data collection, and aggregation complicate fairness assessments, and it discusses business necessity and long-term societal impacts as sources of justification for discrimination in some contexts. The authors call for multidisciplinary collaboration to establish a shared vocabulary and frameworks, using examples like the Equality Act to illustrate how legal principles can guide practical fairness work, and they acknowledge unresolved challenges beyond technical solutions, including bias in unstructured data and complex system modularity.

Abstract

In recent years, the increase in the usage and efficiency of Artificial Intelligence and, more in general, of Automated Decision-Making systems has brought with it an increasing and welcome awareness of the risks associated with such systems. One of such risks is that of perpetuating or even amplifying bias and unjust disparities present in the data from which many of these systems learn to adjust and optimise their decisions. This awareness has on the one hand encouraged several scientific communities to come up with more and more appropriate ways and methods to assess, quantify, and possibly mitigate such biases and disparities. On the other hand, it has prompted more and more layers of society, including policy makers, to call for fair algorithms. We believe that while many excellent and multidisciplinary research is currently being conducted, what is still fundamentally missing is the awareness that having fair algorithms is per se a nearly meaningless requirement that needs to be complemented with many additional social choices to become actionable. Namely, there is a hiatus between what the society is demanding from Automated Decision-Making systems, and what this demand actually means in real-world scenarios. In this work, we outline the key features of such a hiatus and pinpoint a set of crucial open points that we as a society must address in order to give a concrete meaning to the increasing demand of fairness in Automated Decision-Making systems.

Figures (2)

• Figure 1: 1973 Berkeley graduate admission.$G$ represents gender, $D$ the chosen Department and $Y$ the admission outcome. The dashed line represents the possible direct effect of gender on the admission choice.
• Figure 2: Standard Fairness Modelplecko2022fairness. The bidirected dashed edge represents the possible presence of hidden confounding.