Algorithm-assisted decision procedures—including some of the most high-profile ones, such as COMPAS—have been described as unfair because they compound injustice. The complaint is that in such procedures a decision disadvantaging members of a certain group is based on information reflecting the fact that the members of the group have already been unjustly disadvantaged. I assess this reasoning. First, I distinguish the anti-compounding duty from a related but distinct duty—the proportionality duty—from which at least some of the intuitive appeal of the (...) former illegitimately derives. Second, I distinguish between different versions of the anti-compounding duty, arguing that, on some versions, uses of algorithm-assisted decision procedures rarely clash with the anti-compounding duty. Third, drawing on examples of algorithm-assisted decision procedures, I present three objections to the idea that there is a reason not to compound injustice. The most important of these is that one can compound injustice in a non-disrespectful way, and that the wrongfulness of non-disrespectfully compounding injustice is fully explained by the proportionality duty. (shrink)

Here, we provide an ethical analysis of discrimination in private insurance to guide the application of non-discriminatory algorithms for risk prediction in the insurance context. This addresses the need for ethical guidance of data-science experts, business managers, and regulators, proposing a framework of moral reasoning behind the choice of fairness goals for prediction-based decisions in the insurance domain. The reference to private insurance as a business practice is essential in our approach, because the consequences of discrimination and predictive inaccuracy (...) in underwriting are different from those of using predictive algorithms in other sectors. Here we focus on the trade-off in the extent to which one can pursue indirect non-discrimination versus predictive accuracy. The moral assessment of this trade-off is related to the context of application—to the consequences of inaccurate risk predictions in the insurance domain. (shrink)

Algorithms can now identify patterns and correlations in the (big) datasets, and predict outcomes based on those identified patterns and correlations with the use of machine learning techniques and big data, decisions can then be made by algorithms themselves in accordance with the predicted outcomes. Yet, algorithms can inherit questionable values from the datasets and acquire biases in the course of (machine) learning, and automated algorithmic decision-making makes it more difficult for people to see algorithms as (...) biased. While researchers have taken the problem of algorithmic bias seriously, but the current discussion on algorithmic fairness tends to conceptualize ‘fairness’ in algorithmic fairness primarily as a technical issue and attempts to implement pre-existing ideas of ‘fairness’ into algorithms. In this paper, I show that such a view of algorithmic fairness as technical issue is unsatisfactory for the type of problem algorithmic fairness presents. Since decisions on fairness measure and the related techniques for algorithms essentially involve choices between competing values, ‘fairness’ in algorithmic fairness should be conceptualized first and foremost as a political issue, and it should be (re)solved by democratic communication. The aim of this paper, therefore, is to explicitly reconceptualize algorithmic fairness as a political question and suggest the current discussion of algorithmic fairness can be strengthened by adopting the accountability for reasonableness framework. (shrink)

Machine learning algorithms are increasingly used to shape high-stake allocations, sparking research efforts to orient algorithm design towards ideals of justice and fairness. In this research on algorithmic fairness, normative theorizing has primarily focused on identification of “ideally fair” target states. In this paper, we argue that this preoccupation with target states in abstraction from the situated dynamics of deployment is misguided. We propose a framework that takes dynamic trajectories as direct objects of moral appraisal, highlighting three respects (...) in which such trajectories can be subject to evaluation in relation to their temporal dynamics, robustness, and representation. (shrink)

While rapid advances in artificial intelligence hiring tools promise to transform the workplace, these algorithms risk exacerbating existing biases against marginalized groups. In light of these ethical issues, AI vendors have sought to translate normative concepts such as fairness into measurable, mathematical criteria that can be optimized for. However, questions of disability and access often are omitted from these ongoing discussions about algorithmic bias. In this paper, I argue that the multiplicity of different kinds and intensities of people’s disabilities (...) and the fluid, contextual ways in which they manifest point to the limits of algorithmic fairness initiatives. In particular, existing de-biasing measures tend to flatten variance within and among disabled people and abstract away information in ways that reinforce pathologization. While fair machine learning methods can help mitigate certain disparities, I argue that fairness alone is insufficient to secure accessible, inclusive AI. I then outline a disability justice approach, which provides a framework for centering disabled people’s experiences and attending to the structures and norms that underpin algorithmic bias. (shrink)

The application of machine-learning technologies to medical practice promises to enhance the capabilities of healthcare professionals in the assessment, diagnosis, and treatment, of medical conditions. However, there is growing concern that algorithmic bias may perpetuate or exacerbate existing health inequalities. Hence, it matters that we make precise the different respects in which algorithmic bias can arise in medicine, and also make clear the normative relevance of these different kinds of algorithmic bias for broader questions about justice and fairness in healthcare. (...) In this paper, we provide the building blocks for an account of algorithmic bias and its normative relevance in medicine. (shrink)

Holm (2022) argues that a class of algorithmic fairness measures, that he refers to as the ‘performance parity criteria’, can be understood as applications of John Broome’s Fairness Principle. We argue that the performance parity criteria cannot be read this way. This is because in the relevant context, the Fairness Principle requires the equalization of actual individuals’ individual-level chances of obtaining some good (such as an accurate prediction from a predictive system), but the performance parity criteria do not guarantee any (...) such thing: the measures merely ensure that certain population-level ratios hold. (shrink)

Inspired by recent breakthroughs in predictive modeling, practitioners in both industry and government have turned to machine learning with hopes of operationalizing predictions to drive automated decisions. Unfortunately, many social desiderata concerning consequential decisions, such as justice or fairness, have no natural formulation within a purely predictive framework. In efforts to mitigate these problems, researchers have proposed a variety of metrics for quantifying deviations from various statistical parities that we might expect to observe in a fair world and offered (...) a variety of algorithms in attempts to satisfy subsets of these parities or to trade o the degree to which they are satised against utility. In this paper, we connect this approach to fair machine learning to the literature on ideal and non-ideal methodological approaches in political philosophy. The ideal approach requires positing the principles according to which a just world would operate. In the most straightforward application of ideal theory, one supports a proposed policy by arguing that it closes a discrepancy between the real and the perfectly just world. However, by failing to account for the mechanisms by which our non-ideal world arose, the responsibilities of various decision-makers, and the impacts of proposed policies, naive applications of ideal thinking can lead to misguided interventions. In this paper, we demonstrate a connection between the fair machine learning literature and the ideal approach in political philosophy, and argue that the increasingly apparent shortcomings of proposed fair machine learning algorithms reflect broader troubles faced by the ideal approach. We conclude with a critical discussion of the harms of misguided solutions, a reinterpretation of impossibility results, and directions for future research. (shrink)

Predictive algorithms are playing an increasingly prominent role in society, being used to predict recidivism, loan repayment, job performance, and so on. With this increasing influence has come an increasing concern with the ways in which they might be unfair or biased against individuals in virtue of their race, gender, or, more generally, their group membership. Many purported criteria of algorithmic fairness concern statistical relationships between the algorithm’s predictions and the actual outcomes, for instance requiring that the rate of (...) false positives be equal across the relevant groups. We might seek to ensure that algorithms satisfy all of these purported fairness criteria. But a series of impossibility results shows that this is impossible, unless base rates are equal across the relevant groups. What are we to make of these pessimistic results? I argue that none of the purported criteria, except for a calibration criterion, are necessary conditions for fairness, on the grounds that they can all be simultaneously violated by a manifestly fair and uniquely optimal predictive algorithm, even when base rates are equal. I conclude with some general reflections on algorithmic fairness. (shrink)

This article examines the complaint that arbitrary algorithmic decisions wrong those whom they affect. It makes three contributions. First, it provides an analysis of what arbitrariness means in this context. Second, it argues that arbitrariness is not of moral concern except when special circumstances apply. However, when the same algorithm or different algorithms based on the same data are used in multiple contexts, a person may be arbitrarily excluded from a broad range of opportunities. The third contribution is to (...) explain why this systemic exclusion is of moral concern and to offer a solution to address it. (shrink)

The combination of increased availability of large amounts of fine-grained human behavioral data and advances in machine learning is presiding over a growing reliance on algorithms to address complex societal problems. Algorithmic decision-making processes might lead to more objective and thus potentially fairer decisions than those made by humans who may be influenced by greed, prejudice, fatigue, or hunger. However, algorithmic decision-making has been criticized for its potential to enhance discrimination, information and power asymmetry, and opacity. In this paper, (...) we provide an overview of available technical solutions to enhance fairness, accountability, and transparency in algorithmic decision-making. We also highlight the criticality and urgency to engage multi-disciplinary teams of researchers, practitioners, policy-makers, and citizens to co-develop, deploy, and evaluate in the real-world algorithmic decision-making processes designed to maximize fairness and transparency. In doing so, we describe the Open Algortihms project as a step towards realizing the vision of a world where data and algorithms are used as lenses and levers in support of democracy and development. (shrink)

With the increasing use of algorithms in high-stakes areas such as criminal justice and health has come a significant concern about the fairness of prediction-based decision procedures. In this article I argue that a prominent class of mathematically incompatible performance parity criteria can all be understood as applications of John Broome’s account of fairness as the proportional satisfaction of claims. On this interpretation these criteria do not disagree on what it means for an algorithm to be _fair_. Rather they (...) express different understandings of what grounds a claim to a good being allocated by an algorithmic decision procedure. I then argue that an important implication of the Broomean interpretation is that it strengthens the case for outcome-based criteria. Finally, I consider how a version of the levelling-down objection to performance parity criteria arises within the Broomean account. (shrink)

In this paper we develop a general theory of algorithmic fairness. Drawing on Johnson King and Babic’s work on moral encroachment, on Gary Becker’s work on labor market discrimination, and on Strawson’s idea of resentment and indignation as responses to violations of the demand for goodwill toward oneself and others, we locate attitudes to fairness in an agent’s utility function. In particular, we first argue that fairness is a matter of a decision-maker’s relative concern for the plight of people from (...) different groups, rather than of the outcomes produced for different groups. We then show how an agent’s preferences, including in particular their attitudes to error, give rise to their decision thresholds. Tying these points together, we argue that the agent’s relative degrees of concern for different groups manifest in a difference in decision thresholds applied to these groups. (shrink)

Predictive policing, the practice of using of algorithmic systems to forecast crime, is heralded by police departments as the new frontier of crime analysis. At the same time, it is opposed by civil rights groups, academics, and media outlets for being ‘biased’ and therefore discriminatory against communities of color. This paper argues that the prevailing focus on racial bias has overshadowed two normative factors that are essential to a full assessment of the moral permissibility of predictive policing: fairness in the (...) social distribution of the benefits and burdens of policing as well as the distinctive role of consent in determining fair distribution. When these normative factors are given their due attention, several requirements emerge for the fair implementation of predictive policing. Among these requirements are that police departments inform and solicit buy-in from affected communities about strategic decision-making and that departments favor non-enforcement-oriented interventions. (shrink)

To address the rising concern that algorithmic decision-making may reinforce discriminatory biases, researchers have proposed many notions of fairness and corresponding mathematical formalizations. Each of these notions is often presented as a one-size-fits-all, absolute condition; however, in reality, the practical and ethical trade-offs are unavoidable and more complex. We introduce a new approach that considers fairness—not as a binary, absolute mathematical condition—but rather, as a relational notion in comparison to alternative decisionmaking processes. Using US mortgage lending as an example use (...) case, we discuss the ethical foundations of each definition of fairness and demonstrate that our proposed methodology more closely captures the ethical trade-offs of the decision-maker, as well as forcing a more explicit representation of which values and objectives are prioritised. (shrink)

Data-driven predictive algorithms are widely used to automate and guide high-stake decision making such as bail and parole recommendation, medical resource distribution, and mortgage allocation. Nevertheless, harmful outcomes biased against vulnerable groups have been reported. The growing research field known as 'algorithmic fairness' aims to mitigate these harmful biases. Its primary methodology consists in proposing mathematical metrics to address the social harms resulting from an algorithm's biased outputs. The metrics are typically motivated by -- or substantively rooted in -- (...) ideals of distributive justice, as formulated by political and legal philosophers. The perspectives of feminist political philosophers on social justice, by contrast, have been largely neglected. Some feminist philosophers have criticized the local scope of the paradigm of distributive justice and have proposed corrective amendments to surmount its limitations. The present paper brings some key insights of feminist political philosophy to algorithmic fairness. The paper has three goals. First, I show that algorithmic fairness does not accommodate structural injustices in its current scope. Second, I defend the relevance of structural injustices -- as pioneered in the contemporary philosophical literature by Iris Marion Young -- to algorithmic fairness. Third, I take some steps in developing the paradigm of 'responsible algorithmic fairness' to correct for errors in the current scope and implementation of algorithmic fairness. I close by some reflections of directions for future research. (shrink)

This paper presents an intuitive explanation about why and how Rawlsian Theory of Justice (Rawls in A theory of justice, Harvard University Press, Harvard, 1971) provides the foundations to a solution for algorithmic bias. The contribution of the paper is to discuss and show why Rawlsian ideas in their original form (e.g. the veil of ignorance, original position, and allowing inequalities that serve the worst-off) are relevant to operationalize fairness for algorithmic decision making. The paper also explains how this leads (...) to a specific MinMaxfairness solution, which addresses the basic challenges of algorithmic justice. We combine substantive elements of Rawlsian perspective with an intuitive explanation in order to provide accessible and practical insights. The goal is to propose and motivate why and how the MinMaxfairness solution derived from Rawlsian principles overcomes some of the current challenges for algorithmic bias and highlight the benefits provided when compared to other approaches. The paper presents and discusses the solution by building a bridge between the qualitative theoretical aspects and the quantitative technical approach. (shrink)

Over the past decade, scholars, institutions, and activists have voiced strong concerns about the potential of automated decision systems to indirectly discriminate against vulnerable groups. This article analyses the ethics of algorithmic indirect discrimination, and argues that we can explain what is morally bad about such discrimination by reference to the fact that it causes harm. The article first sketches certain elements of the technical and conceptual background, including definitions of direct and indirect algorithmic differential treatment. It next introduces three (...) prominent accounts of fairness as potential explanations if the badness of algorithmic indirect discrimination, but argues that all three are vulnerable to powerful levelling-down-style objections. Instead, the article demonstrates how proper attention to the way differences in decision-scenarios affect the distribution of harms can help us account for intuitions in prominent cases. Finally, the article considers a potential objection based on the fact that certain forms of algorithmic indirect discrimination appear to distribute rather than cause harm, and notes that we can explain how such distributions cause harm by attending to differences in individual and group vulnerability. (shrink)

Algorithmic decision-making increasingly shapes people's daily lives. Given that such autonomous systems can cause severe harm to individuals and social groups, fairness concerns have arisen. A human-centric approach demanded by scholars and policymakers requires considering people's fairness perceptions when designing and implementing algorithmic decision-making. We provide a comprehensive, systematic literature review synthesizing the existing empirical insights on perceptions of algorithmic fairness from 58 empirical studies spanning multiple domains and scientific disciplines. Through thorough coding, we systemize the current empirical literature along (...) four dimensions: (1) algorithmic predictors, (2) human predictors, (3) comparative effects (human decision-making vs. algorithmic decision-making), and (4) consequences of algorithmic decision-making. While we identify much heterogeneity around the theoretical concepts and empirical measurements of algorithmic fairness, the insights come almost exclusively from Western-democratic contexts. By advocating for more interdisciplinary research adopting a society-in-the-loop framework, we hope our work will contribute to fairer and more responsible algorithmic decision-making. (shrink)

Philosophy & Public Affairs, Volume 50, Issue 2, Page 239-266, Spring 2022.

Philosophy &Public Affairs, Volume 51, Issue 2, Page 166-190, Spring 2023.

Despite the rapid adoption of technology in human resource departments, there is little empirical work that examines the potential challenges of algorithmic decision-making in the recruitment process. In this paper, we take the perspective of job applicants and examine how they perceive the use of algorithms in selection and recruitment. Across four studies on Amazon Mechanical Turk, we show that people in the role of a job applicant perceive algorithm-driven recruitment processes as less fair compared to human only (...) or algorithm-assisted human processes. This effect persists regardless of whether the outcome is favorable to the applicant or not. A potential mechanism underlying algorithm resistance is the belief that algorithms will not be able to recognize their uniqueness as a candidate. Although the use of algorithms has several benefits for organizations such as improved efficiency and bias reduction, our results highlight a potential cost of using them to screen potential employees during recruitment. (shrink)

Fairness in machine learning (ML) is an ever-growing field of research due to the manifold potential for harm from algorithmic discrimination. To prevent such harm, a large body of literature develops new approaches to quantify fairness. Here, we investigate how one can divert the quantification of fairness by describing a practice we call “fairness hacking” for the purpose of shrouding unfairness in algorithms. This impacts end-users who rely on learning algorithms, as well as the broader community interested in (...) fair AI practices. We introduce two different categories of fairness hacking in reference to the established concept of p-hacking. The first category, intra-metric fairness hacking, describes the misuse of a particular metric by adding or removing sensitive attributes from the analysis. In this context, countermeasures that have been developed to prevent or reduce p-hacking can be applied to similarly prevent or reduce fairness hacking. The second category of fairness hacking is inter-metric fairness hacking. Inter-metric fairness hacking is the search for a specific fair metric with given attributes. We argue that countermeasures to prevent or reduce inter-metric fairness hacking are still in their infancy. Finally, we demonstrate both types of fairness hacking using real datasets. Our paper intends to serve as a guidance for discussions within the fair ML community to prevent or reduce the misuse of fairness metrics, and thus reduce overall harm from ML applications. (shrink)

Algorithmic fairness is a new interdisciplinary field of study focused on how to measure whether a process, or algorithm, may unintentionally produce unfair outcomes, as well as whether or how the potential unfairness of such processes can be mitigated. Statistical discrimination describes a set of informational issues that can induce rational (i.e., Bayesian) decision-making to lead to unfair outcomes even in the absence of discriminatory intent. In this article, we provide overviews of these two related literatures and draw connections between (...) them. The comparison illustrates both the conflict between rationality and fairness and the importance of endogeneity (e.g., “rational expectations” and “self-fulfilling prophecies”) in defining and pursuing fairness. Taken in concert, we argue that the two traditions suggest a value for considering new fairness notions that explicitly account for how the individual characteristics an algorithm intends to measure may change in response to the algorithm. (shrink)

To address the rising concern that algorithmic decision-making may reinforce discriminatory biases, researchers have proposed many notions of fairness and corresponding mathematical formalizations. Each of these notions is often presented as a one-size-fits-all, absolute condition; however, in reality, the practical and ethical trade-offs are unavoidable and more complex. We introduce a new approach that considers fairness—not as a binary, absolute mathematical condition—but rather, as a relational notion in comparison to alternative decision-making processes. Using U.S. mortgage lending as an example use (...) case, we discuss the ethical foundations of each definition of fairness and demonstrate that our proposed methodology more closely captures the ethical trade-offs of the decision-maker. (shrink)

Data-driven algorithms are currently deployed in several fields, leading to a rapid increase in the importance algorithms have in decision-making processes. Over the last years, several instances of discrimination by algorithms were observed. A new branch of research emerged to examine the concept of “algorithmic fairness.” No consensus currently exists on a single operationalization of fairness, although causal-based definitions are arguably more aligned with the human conception of fairness. The aim of this article is to investigate the (...) degree of this alignment in a case study inspired by a recent ruling of an Italian court on the reputational-ranking algorithm used by a food delivery platform. I relied on the documentation of the legal dispute to discuss the applicability, intuitiveness and appropriateness of causal models in evaluating fairness, with a specific focus on a causal-based fairness definition called “counterfactual fairness.” I first describe the details of the dispute and the arguments presented to the court, as well as the court’s final decision, to establish the context of the case study. Then, I translate the dispute into a formal simplified problem using a causal diagram, which represents the main aspects of the data generation process in the case study. I identify the criteria used by the court in ruling that the algorithm was unfair and compare them with the counterfactual fairness definition. The definition of counterfactual fairness was found to be well aligned with the human conception of fairness in this case study, using the court order rationale as a gold standard. (shrink)

Algorithms increasingly make managerial decisions that people used to make. Perceptions of algorithms, regardless of the algorithms' actual performance, can significantly influence their adoption, yet we do not fully understand how people perceive decisions made by algorithms as compared with decisions made by humans. To explore perceptions of algorithmic management, we conducted an online experiment using four managerial decisions that required either mechanical or human skills. We manipulated the decision-maker, and measured perceived fairness, trust, and emotional (...) response. With the mechanical tasks, algorithmic and human-made decisions were perceived as equally fair and trustworthy and evoked similar emotions; however, human managers' fairness and trustworthiness were attributed to the manager's authority, whereas algorithms' fairness and trustworthiness were attributed to their perceived efficiency and objectivity. Human decisions evoked some positive emotion due to the possibility of social recognition, whereas algorithmic decisions generated a more mixed response – algorithms were seen as helpful tools but also possible tracking mechanisms. With the human tasks, algorithmic decisions were perceived as less fair and trustworthy and evoked more negative emotion than human decisions. Algorithms' perceived lack of intuition and subjective judgment capabilities contributed to the lower fairness and trustworthiness judgments. Positive emotion from human decisions was attributed to social recognition, while negative emotion from algorithmic decisions was attributed to the dehumanizing experience of being evaluated by machines. This work reveals people's lay concepts of algorithmic versus human decisions in a management context and suggests that task characteristics matter in understanding people's experiences with algorithmic technologies. (shrink)

This paper examines racial discrimination and algorithmic bias in predictive policing algorithms (PPAs), an emerging technology designed to predict threats and suggest solutions in law enforcement. We first describe what discrimination is in a case study of Chicago’s PPA. We then explain their causes with Broadbent’s contrastive model of causation and causal diagrams. Based on the cognitive science literature, we also explain why fairness is not an objective truth discoverable in laboratories but has context-sensitive social meanings that need to (...) be negotiated through democratic processes. With the above analysis, we next predict why some recommendations given in the bias reduction literature are not as effective as expected. Unlike the cliché highlighting equal participation for all stakeholders in predictive policing, we emphasize power structures to avoid hermeneutical lacunae. Finally, we aim to control PPA discrimination by proposing a governance solution—a framework of a social safety net. (shrink)

With increasing use of automated algorithmic decision-making, issues of algorithmic fairness have attracted much attention lately. In this growing literature, existing concepts from ethics and political philosophy are often applied to new contexts. The reverse—that novel insights from the algorithmic fairness literature are fed back into ethics and political philosophy—is far less established. However, this short commentary on Baumann and Loi (Philosophy & Technology, 36(3), 45 2023) aims to do precisely this. Baumann and Loi argue that among algorithmic group fairness (...) measures proposed, one—sufficiency (well-calibration) is morally defensible for insurers to use, whereas independence (statistical parity or demographic parity) and separation (equalized odds) are not normatively appropriate in the insurance context. Such a result may seem to be of relatively narrow interest to insurers and insurance scholars only. We argue, however, that arguments such as that offered by Baumann and Loi have an important but so far overlooked connection to the derivation of the minimal state offered by Nozick (1974) and thus to political philosophy at large. (shrink)

In this position paper, we introduce a new epistemic lens for analyzing algorithmic harm. We argue that the epistemic lens we propose herein has two key contributions to help reframe and address some of the assumptions underlying inquiries into algorithmic fairness. First, we argue that using the framework of epistemic injustice helps to identify the root causes of harms currently framed as instances of representational harm. We suggest that the epistemic lens offers a theoretical foundation for expanding approaches to algorithmic (...) fairness in order to address a wider range of harms not recognized by existing technical or legal definitions. Second, we argue that the epistemic lens helps to identify the epistemic goals of inquiries into algorithmic fairness. There are two distinct contexts within which we examine algorithmic harm: at times, we seek to understand and describe the world as it is, and, at other times, we seek to build a more just future. The epistemic lens can serve to direct our attention to the epistemic frameworks that shape our interpretations of the world as it is and the ways we envision possible futures. Clarity with respect to which epistemic context is relevant in a given inquiry can further help inform choices among the different ways of measuring and addressing algorithmic harms. We introduce this framework with the goal of initiating new research directions bridging philosophical, legal, and technical approaches to understanding and mitigating algorithmic harms. (shrink)

Machine learning classifiers are increasingly used to inform, or even make, decisions significantly affecting human lives. Fairness concerns have spawned a number of contributions aimed at both identifying and addressing unfairness in algorithmic decision-making. This paper critically discusses the adoption of group-parity criteria (e.g., demographic parity, equality of opportunity, treatment equality) as fairness standards. To this end, we evaluate the use of machine learning methods relative to different steps of the decision-making process: assigning a predictive score, linking a classification to (...) the score, and adopting decisions based on the classification. Throughout our inquiry we use the COMPAS system, complemented by a radical simplification of it (our SAPMOC I and SAPMOC II models), as our running examples. Through these examples, we show how a system that is equally accurate for different groups may fail to comply with group-parity standards, owing to different base rates in the population. We discuss the general properties of the statistics determining the satisfaction of group-parity criteria and levels of accuracy. Using the distinction between scoring, classifying, and deciding, we argue that equalisation of classifications/decisions between groups can be achieved thorough group-dependent thresholding. We discuss contexts in which this approach may be meaningful and useful in pursuing policy objectives. We claim that the implementation of group-parity standards should be left to competent human decision-makers, under appropriate scrutiny, since it involves discretionary value-based political choices. Accordingly, predictive systems should be designed in such a way that relevant policy goals can be transparently implemented. Our paper presents three main contributions: (1) it addresses a complex predictive system through the lens of simplified toy models; (2) it argues for selective policy interventions on the different steps of automated decision-making; (3) it points to the limited significance of statistical notions of fairness to achieve social goals. (shrink)

We are currently witnessing a sharp rise in the use of algorithmic decision-making tools. In these instances, a new wave of policy concerns is set forth. This article strives to map out these issues, separating the wheat from the chaff. It aims to provide policy makers and scholars with a comprehensive framework for approaching these thorny issues in their various capacities. To achieve this objective, this article focuses its attention on a general analytical framework, which will be applied to a (...) specific subset of the overall discussion. The analytical framework will reduce the discussion to two dimensions, every one of which addressing two central elements. These four factors call for a distinct discussion, which is at times absent in the existing literature. The two dimensions are the specific and novel problems the process assumedly generates and the specific attributes which exacerbate them. While the problems are articulated in a variety of ways, they most likely could be reduced to two broad categories: efficiency and fairness-based concerns. In the context of this discussion, such problems are usually linked to two salient attributes the algorithmic processes feature—its opaque and automated nature. (shrink)

In light of the widespread use of algorithmic (intelligent) systems across numerous domains, there is an increasing awareness about the need to explain their underlying decision-making process and resulting outcomes. Since oftentimes these systems are being considered as black boxes, adding explanations to their outcomes may contribute to the perception of their transparency and, as a result, increase users’ trust and fairness perception towards the system, regardless of its actual fairness, which can be measured using various fairness tests and measurements. (...) Different explanation styles may have a different impact on users’ perception of fairness towards the system and on their understanding of the outcome of the system. Hence, there is a need to understand how various explanation styles may impact non-expert users’ perceptions of fairness and understanding of the system’s outcome. In this study we aimed at fulfilling this need. We performed a between-subject user study in order to examine the effect of various explanation styles on users’ fairness perception and understanding of the outcome. In the experiment we examined four known styles of textual explanations (case-based, demographic-based, input influence-based and sensitivity-based) along with a new style (certification-based) that reflect the results of an auditing process of the system. The results suggest that providing some kind of explanation contributes to users’ understanding of the outcome and that some explanation styles are more beneficial than others. Moreover, while explanations provided by the system are important and can indeed enhance users’ perception of fairness, their perception mainly depends on the outcome of the system. The results may shed light on one of the main problems in explainability of algorithmic systems, which is choosing the best explanation to promote users’ fairness perception towards a particular system, with respect to the outcome of the system. The contribution of this study is reflected in the new and realistic case study that was examined, in the creation and evaluation of a new explanation style that can be used as the link between the actual (computational) fairness of the system and users’ fairness perception and in the need of analyzing and evaluating explanations while taking into account the outcome of the system. (shrink)

There is growing concern that decision-making informed by machine learning (ML) algorithms may unfairly discriminate based on personal demographic attributes, such as race and gender. Scholars have responded by introducing numerous mathematical definitions of fairness to test the algorithm, many of which are in conflict with one another. However, these reductionist representations of fairness often bear little resemblance to real-life fairness considerations, which in practice are highly contextual. Moreover, fairness metrics tend to be implemented in narrow and targeted toolkits (...) that are difficult to integrate into an algorithm’s broader ethical assessment. In this paper, we derive lessons from ethical philosophy and welfare economics as they relate to the contextual factors relevant for fairness. In particular we highlight the debate around acceptability of particular inequalities and the inextricable links between fairness, welfare and autonomy. We propose Key Ethics Indicators (KEIs) as a way towards providing a more holistic understanding of whether or not an algorithm is aligned to the decision-maker’s ethical values. (shrink)

Efforts to promote equitable public policy with algorithms appear to be fundamentally constrained by the “impossibility of fairness” (an incompatibility between mathematical definitions of fairness). This technical limitation raises a central question about algorithmic fairness: How can computer scientists and policymakers support equitable policy reforms with algorithms? In this article, I argue that promoting justice with algorithms requires reforming the methodology of algorithmic fairness. First, I diagnose the problems of the current methodology for algorithmic fairness, which I (...) call “formal algorithmic fairness.” Because formal algorithmic fairness restricts analysis to isolated decision-making procedures, it leads to the impossibility of fairness and to models that exacerbate oppression despite appearing “fair.” Second, I draw on theories of substantive equality from law and philosophy to propose an alternative methodology, which I call “substantive algorithmic fairness.” Because substantive algorithmic fairness takes a more expansive scope of analysis, it enables an escape from the impossibility of fairness and provides a rigorous guide for alleviating injustice with algorithms. In sum, substantive algorithmic fairness presents a new direction for algorithmic fairness: away from formal mathematical models of “fair” decision-making and toward substantive evaluations of whether and how algorithms can promote justice in practice. (shrink)

A number of artificial intelligence systems have been proposed to assist users in identifying the issues of algorithmic fairness and transparency. These AI systems use diverse bias detection methods from various perspectives, including exploratory cues, interpretable tools, and revealing algorithms. This study explains the design of AI systems by probing how users make sense of fairness and transparency as they are hypothetical in nature, with no specific ways for evaluation. Focusing on individual perceptions of fairness and transparency, this study (...) examines the roles of normative values in over-the-top platforms by empirically testing their effects on sensemaking processes. A mixed-method design incorporating both qualitative and quantitative approaches was used to discover user heuristics and to test the effects of such normative values on user acceptance. Collectively, a composite concept of transparent fairness emerged around user sensemaking processes and its formative roles regarding their underlying relations to perceived quality and credibility. From a sensemaking perspective, this study discusses the implications of transparent fairness in algorithmic media platforms by clarifying how and what should be done to make algorithmic media more trustable and reliable platforms. Based on the findings, a theoretical model is developed to define transparent fairness as an essential algorithmic attribute in the context of OTT platforms. (shrink)

What does it mean for algorithmic classifications to be fair to different socially salient groups? According to classification parity criteria, what is required is equality across groups with respect to some performance measure such as error rates. Critics of classification parity object that classification parity entails that achieving fairness may require us to choose an algorithm that makes no group better off and some groups worse off than an alternative. In this article, I interpret the problem of algorithmic fairness (...) as a case concerning the ethics of the distribution of algorithmic classifications across groups (as opposed to, e.g., the fairness of data collection). I begin with a short introduction of algorithmic fairness as a problem discussed in machine learning. I then show how the criticism raised against classification parity is a form of leveling down objection, and I interpret the egalitarianism of classification parity as deontic egalitarianism. I then discuss a challenge to this interpretation and suggest a revision. Finally, I examine how my interpretation provides proponents of classification parity with a response to the leveling down criticism and how it relates to a recent suggestion to evaluate fairness for automated decision-making systems based on risk and welfare considerations from behind a veil of ignorance. (shrink)

Fairness is one of the most prominent values in the Ethics and Artificial Intelligence (AI) debate and, specifically, in the discussion on algorithmic decision-making (ADM). However, while the need for fairness in ADM is widely acknowledged, the very concept of fairness has not been sufficiently explored so far. Our paper aims to fill this gap and claims that an ethically informed re-definition of fairness is needed to adequately investigate fairness in ADM. To achieve our goal, after an introductory section aimed (...) at clarifying the aim and structure of the paper, in section “Fairness in algorithmic decision-making” we provide an overview of the state of the art of the discussion on fairness in ADM and show its shortcomings; in section “Fairness as an ethical value”, we pursue an ethical inquiry into the concept of fairness, drawing insights from accounts of fairness developed in moral philosophy, and define fairness as an ethical value. In particular, we argue that fairness is articulated in a distributive and socio-relational dimension; it comprises three main components: fair equality of opportunity, equal right to justification, and fair equality of relationship; these components are grounded in the need to respect persons both as persons and as particular individuals. In section “Fairness in algorithmic decision-making revised”, we analyze the implications of our redefinition of fairness as an ethical value on the discussion of fairness in ADM and show that each component of fairness has profound effects on the criteria that ADM ought to meet. Finally, in section “Concluding remarks”, we sketch some broader implications and conclude. (shrink)

Statistical criteria of fairness are formal measures of how an algorithm performs that aim to help us determine whether an algorithm would be fair to use in decision-making. In this paper, I introduce a new version of the criterion known as “Equalized Odds,” argue that it is a requirement of procedural fairness, and show that it is immune to a number of objections to the standard version.

The literature on algorithmic fairness has examined exogenous sources of biases such as shortcomings in the data and structural injustices in society. It has also examined internal sources of bias as evidenced by a number of impossibility theorems showing that no algorithm can concurrently satisfy multiple criteria of fairness. This paper contributes to the literature stemming from the impossibility theorems by examining how informational richness affects the accuracy and fairness of predictive algorithms. With the aid of a computer simulation, (...) we show that informational richness is the engine that drives improvements in the performance of a predictive algorithm, in terms of both accuracy and fairness. The centrality of informational richness suggests that classification parity, a popular criterion of algorithmic fairness, should be given relatively little weight. But we caution that the centrality of informational richness should be taken with a grain of salt in light of practical limitations, in particular, the so-called bias-variance trade off. (shrink)

This thesis aims to clarify a number of conceptual aspects of the debate surrounding algorithmic fairness. The particular focus here is the role of causal modeling in defining criteria of algorithmic fairness. In Chapter 1, I argue that in the discussion of algorithmic fairness, two fundamentally distinct notions of fairness have been conflated. Subsequently, I propose that what is usually taken to be the problem of algorithmic fairness should be divided into two subproblems, the problem of predictive fairness, and the (...) problem of allocative fairness. At the core of Chapter 2 is the proof of a theorem that establishes that three of the most popular (predictive) fairness criteria are pairwise incompatible. In particular, I show that under certain conditions, a predictive algorithm that satisfies a criterion called counterfactual fairness will with logical necessity violate two other popular predictive fairness criteria called equalized odds and predictive parity. In Chapter 3, a new predictive fairness criterion is developed using a mathematical framework for causal modeling. This fairness criterion, which I call causal relevance fairness, is a relaxation of another popular fairness criterion, counterfactual fairness, but turns out to be more closely in line with philosophical theories of discrimination. In Chapter 4, another infamous impossibility result in algorithmic fairness is analyzed through the lens of causality. I argue that by using a causal inference method called matching, we can modify the two fairness criteria equalized odds and predictive parity in a way that resolves the impossibility. Lastly, Chapter 5 contains an empirical case study. In it, the fairness of a popular recidivism risk prediction tool is analyzed using the criteria of (predictive) fairness developed earlier. (shrink)

A growing number of studies on fairness in artificial intelligence (AI) use the notion of intersectionality to measure AI fairness. Most of these studies take intersectional fairness to be a matter of statistical parity among intersectional subgroups: an AI algorithm is “intersectionally fair” if the probability of the outcome is roughly the same across all subgroups defined by different combinations of the protected attributes. This paper identifies and examines three fundamental problems with this dominant interpretation of intersectional fairness in (...) AI. First, the dominant approach is so preoccupied with the intersection of attributes/categories (e.g., race, gender) that it fails to address the intersection of oppression (e.g., racism, sexism), which is more central to intersectionality as a critical framework. Second, the dominant approach faces a dilemma between infinite regress and fairness gerrymandering: it either keeps splitting groups into smaller subgroups or arbitrarily selects protected groups. Lastly, the dominant view fails to capture what it really means for AI algorithms to be fair, in terms of both distributive and non-distributive fairness. I distinguish a strong sense of AI fairness from a weak sense that is prevalent in the literature, and conclude by envisioning paths towards strong intersectional fairness in AI. (shrink)

Measures of algorithmic bias can be roughly classified into four categories, distinguished by the conditional probabilistic dependencies to which they are sensitive. First, measures of "procedural bias" diagnose bias when the score returned by an algorithm is probabilistically dependent on a sensitive class variable (e.g. race or sex). Second, measures of "outcome bias" capture probabilistic dependence between class variables and the outcome for each subject (e.g. parole granted or loan denied). Third, measures of "behavior-relative error bias" capture probabilistic dependence between (...) class variables and the algorithmic score, conditional on target behaviors (e.g. recidivism or loan default). Fourth, measures of "score-relative error bias" capture probabilistic dependence between class variables and behavior, conditional on score. Several recent discussions have demonstrated a tradeoff between these different measures of algorithmic bias, and at least one recent paper has suggested conditions under which tradeoffs may be minimized. -/- In this paper we use the machinery of causal graphical models to show that, under standard assumptions, the underlying causal relations among variables forces some tradeoffs. We delineate a number of normative considerations that are encoded in different measures of bias, with reference to the philosophical literature on the wrongfulness of disparate treatment and disparate impact. While both kinds of error bias are nominally motivated by concern to avoid disparate impact, we argue that consideration of causal structures shows that these measures are better understood as complicated and unreliable measures of procedural biases (i.e. disparate treatment). Moreover, while procedural bias is indicative of disparate treatment, we show that the measure of procedural bias one ought to adopt is dependent on the account of the wrongfulness of disparate treatment one endorses. Finally, given that neither score-relative nor behavior-relative measures of error bias capture the relevant normative considerations, we suggest that error bias proper is best measured by score-based measures of accuracy, such as the Brier score. (shrink)

The problem of algorithmic fairness is typically framed as the problem of finding a unique formal criterion that guarantees that a given algorithmic decision-making procedure is morally permissible. In this paper, I argue that this is conceptually misguided and that we should replace the problem with two sub-problems. If we examine how most state-of-the-art machine learning systems work, we notice that there are two distinct stages in the decision-making process. First, a prediction of a relevant property is made. Secondly, a (...) decision is taken based (at least partly) on this prediction. These two stages have different aims: the prediction is aimed at accuracy, while the decision is aimed at allocating a given good in a way that maximizes some context-relative utility measure. Correspondingly, two different fairness issues can arise. First, predictions could be biased in discriminatory ways. This means that the predictions contain systematic errors for a specific group of individuals. Secondly, the system’s decisions could result in an allocation of goods that is in tension with the principles of distributive justice. These two fairness issues are distinct problems that require different types of solutions. I here provide a formal framework to address both issues and argue that this way of conceptualizing them resolves some of the paradoxes present in the discussion of algorithmic fairness. (shrink)

The increasing implementation of and reliance on machine-learning (ML) algorithms to perform tasks, deliver services and make decisions in health and healthcare have made the need for fairness in ML, and more specifically in healthcare ML algorithms (HMLA), a very important and urgent task. However, while the debate on fairness in the ethics of artificial intelligence (AI) and in HMLA has grown significantly over the last decade, the very concept of fairness as an ethical value has not yet (...) been sufficiently explored. Our paper aims to fill this gap and address the AI ethics principle of fairness from a conceptual standpoint, drawing insights from accounts of fairness elaborated in moral philosophy and using them to conceptualise fairness as an ethical value and to redefine fairness in HMLA accordingly. To achieve our goal, following a first section aimed at clarifying the background, methodology and structure of the paper, in the second section, we provide an overview of the discussion of the AI ethics principle of fairness in HMLA and show that the concept of fairness underlying this debate is framed in purely distributive terms and overlaps with non-discrimination, which is defined in turn as the absence of biases. After showing that this framing is inadequate, in the third section, we pursue an ethical inquiry into the concept of fairness and argue that fairness ought to be conceived of as an ethical value. Following a clarification of the relationship between fairness and non-discrimination, we show that the two do not overlap and that fairness requires much more than just non-discrimination. Moreover, we highlight that fairness not only has a distributive but also a socio-relational dimension. Finally, we pinpoint the constitutive components of fairness. In doing so, we base our arguments on a renewed reflection on the concept of respect, which goes beyond the idea of equal respect to include respect for individual persons. In the fourth section, we analyse the implications of our conceptual redefinition of fairness as an ethical value in the discussion of fairness in HMLA. Here, we claim that fairness requires more than non-discrimination and the absence of biases as well as more than just distribution; it needs to ensure that HMLA respects persons both as persons and as particular individuals. Finally, in the fifth section, we sketch some broader implications and show how our inquiry can contribute to making HMLA and, more generally, AI promote the social good and a fairer society. (shrink)

Recent advances in machine learning methods have created opportunities to eliminate unfairness from algorithmic decision making. Multiple computational techniques (i.e., algorithmic fairness criteria) have arisen out of this work. Yet, urgent questions remain about the perceived fairness of these criteria and in which situations organizations should use them. In this paper, we seek to gain insight into these questions by exploring fairness perceptions of five algorithmic criteria. We focus on two key dimensions of fairness evaluations: distributive fairness and procedural fairness. (...) We shed light on variation in the potential for different algorithmic criteria to facilitate distributive fairness. Subsequently, we discuss procedural fairness and provide a framework for understanding how algorithmic criteria relate to essential aspects of this construct, which helps to identify when a specific criterion is suitable. From a practical standpoint, we encourage organizations to recognize that managing fairness in machine learning systems is complex, and that adopting a blind or one-size-fits-all mentality toward algorithmic criteria will surely damage people’s attitudes and trust in automated technology. Instead, firms should carefully consider the subtle yet significant differences between these technical solutions. (shrink)