Algorithmic Fairness

In ‘Rawlsian algorithmic fairness and a missing aggregation property of the difference Principle’, the authors argue that there is a false assumption in algorithmic fairness interventions inspired by John Rawls’ theory of justice. They argue that applying the difference principle at the level of a local algorithmic decision-making context (what they term a ‘constituent situation’), is neither necessary nor sufficient for the difference principle to be upheld at the aggregate level of society at large. I find these arguments compelling. They (...) are in line with long-standing concerns about Rawlsian justice in general, and its application in ‘fair’ machine learning in particular. Overall, this article is a very welcome contribution to the literature. In this commentary, I would like to briefly highlight some limitations of these arguments, and some of their broader implications for the field of political philosophy-inspired fair machine learning. (shrink)

When selecting a model from a set of equally performant models, how much unfairness can you really reduce? Is it important to be intentional about fairness when choosing among this set, or is arbitrarily choosing among the set of “good” models good enough? Recent work has highlighted that the phenomenon of model multiplicity—where multiple models with nearly identical predictive accuracy exist for the same task—has both positive and negative implications for fairness, from strengthening the enforcement of civil rights law in (...) AI systems to showcasing arbitrariness in AI decision-making. Despite the enormous implications of model multiplicity, there is little work that explores the properties of sets of equally accurate models, or Rashomon sets, in general. In this paper, we present five main theoretical and methodological contributions which help us to understand the relatively unexplored properties of the Rashomon set, in particular with regards to fairness. Our contributions include methods for efficiently sampling models from this set and techniques for identifying the fairest models according to key fairness metrics such as statistical parity. We also derive the probability that an individual’s prediction will be flipped within the Rashomon set, as well as expressions for the set’s size and the distribution of error tolerance used across models. These results lead to policy-relevant takeaways, such as the importance of intentionally looking for fair models within the Rashomon set, and understanding which individuals or groups may be more susceptible to arbitrary decisions. (shrink)

As artificial intelligence (AI) systems increasingly permeate processes of cultural and epistemic production, there are growing concerns about how their outputs may confine individuals and groups to static or restricted narratives about who or what they could be. In this paper, we advance the discourse surrounding these concerns by making three contributions. First, we introduce the concept of aspirational affordance to describe how culturally shared interpretive resources can shape individual cognition, and in particular exercises practical imagination. We show how this (...) concept can ground productive evaluations of the risks of AI-enabled representations and narratives. Second, we provide three reasons for scrutinizing of AI's influence on aspirational affordances: AI's influence is potentially more potent, but less public than traditional sources; AI's influence is not simply incremental, but ecological, transforming the entire landscape of cultural and epistemic practices that traditionally shaped aspirational affordances; and AI's influence is highly concentrated, with a few corporate-controlled systems mediating a growing portion of aspirational possibilities. Third, to advance such a scrutiny, we introduce the concept of aspirational harm, which, in the context of AI systems, arises when AI-enabled aspirational affordances distort or diminish available interpretive resources in ways that undermine individuals' ability to imagine relevant practical possibilities and alternative futures. Through three case studies, we illustrate how aspirational harms extend the existing discourse on AI-inflicted harms beyond representational and allocative harms, warranting separate attention. Through these conceptual resources and analyses, this paper advances understanding of the psychological and societal stakes of AI's role in shaping individual and collective aspirations. (shrink)

Two goals - improving replicability and accountability of Machine Learning research respectively, have accrued much attention from the AI ethics and the Machine Learning community. Despite sharing the measures of improving transparency, the two goals are discussed in different registers - replicability registers with scientific reasoning whereas accountability registers with ethical reasoning. Given the existing challenge of the Responsibility Gap - holding Machine Learning scientists accountable for Machine Learning harms due to them being far from sites of application, this paper (...) posits that reconceptualizing replicability can help bridge the gap. Through a shift from model performance replicability to claim replicability, Machine Learning scientists can be held accountable for producing non-replicable claims that are prone to eliciting harm due to misuse and misinterpretation. In this paper, I make the following contributions. First, I define and distinguish two forms of replicability for ML research that can aid constructive conversations around replicability. Second, I formulate an argument for claim-replicability's advantage over model performance replicability in justifying assigning accountability to Machine Learning scientists for producing non-replicable claims and show how it enacts a sense of responsibility that is actionable. In addition, I characterize the implementation of claim replicability as more of a social project than a technical one by discussing its competing epistemological principles, practical implications on Circulating Reference, Interpretative Labor, and research communication. (shrink)

The prevailing discourse around AI ethics lacks the language and formalism necessary to capture the diverse ethical concerns that emerge when AI systems interact with individuals. Drawing on Sen and Nussbaum's capability approach, we present a framework formalizing a network of ethical concepts and entitlements necessary for AI systems to confer meaningful benefit or assistance to stakeholders. Such systems enhance stakeholders' ability to advance their life plans and well-being while upholding their fundamental rights. We characterize two necessary conditions for morally (...) permissible interactions between AI systems and those impacted by their functioning, and two sufficient conditions for realizing the ideal of meaningful benefit. We then contrast this ideal with several salient failure modes, namely, forms of social interactions that constitute unjustified paternalism, coercion, deception, exploitation and domination. The proliferation of incidents involving AI in high-stakes domains underscores the gravity of these issues and the imperative to take an ethics-led approach to AI systems from their inception. (shrink)

Algorithms wield increasing control over our lives—over the jobs we get, the loans we're granted, the information we see online. Algorithms can and often do wield their power in a biased way, and much work has been devoted to algorithmic bias. In contrast, algorithmic neutrality has been largely neglected. I investigate algorithmic neutrality, tackling three questions: What is algorithmic neutrality? Is it possible? And when we have it in mind, what can we learn about algorithmic bias?

This paper explores the development of an ethical guardrail framework for AI systems, emphasizing the importance of customizable guardrails that align with diverse user values and underlying ethics. We address the challenges of AI ethics by proposing a structure that integrates rules, policies, and AI assistants to ensure responsible AI behavior, while comparing the proposed framework to the existing state-of-the-art guardrails. By focusing on practical mechanisms for implementing ethical standards, we aim to enhance transparency, user autonomy, and continuous improvement in (...) AI systems. Our approach accommodates ethical pluralism, offering a flexible and adaptable solution for the evolving landscape of AI governance. The paper concludes with strategies for resolving conflicts between ethical directives, underscoring the present and future need for robust, nuanced and context-aware AI systems. (shrink)

Background There is a growing concern about artificial intelligence (AI) applications in healthcare that can disadvantage already under-represented and marginalised groups (eg, based on gender or race). Objectives Our objectives are to canvas the range of strategies stakeholders endorse in attempting to mitigate algorithmic bias, and to consider the ethical question of responsibility for algorithmic bias. Methodology The study involves in-depth, semistructured interviews with healthcare workers, screening programme managers, consumer health representatives, regulators, data scientists and developers. Results Findings reveal considerable (...) divergent views on three key issues. First, views on whether bias is a problem in healthcare AI varied, with most participants agreeing bias is a problem (which we call the bias-critical view), a small number believing the opposite (the bias-denial view), and some arguing that the benefits of AI outweigh any harms or wrongs arising from the bias problem (the bias-apologist view). Second, there was a disagreement on the strategies to mitigate bias, and who is responsible for such strategies. Finally, there were divergent views on whether to include or exclude sociocultural identifiers (eg, race, ethnicity or gender-diverse identities) in the development of AI as a way to mitigate bias. Conclusion/significance Based on the views of participants, we set out responses that stakeholders might pursue, including greater interdisciplinary collaboration, tailored stakeholder engagement activities, empirical studies to understand algorithmic bias and strategies to modify dominant approaches in AI development such as the use of participatory methods, and increased diversity and inclusion in research teams and research participant recruitment and selection. (shrink)

Using artificial intelligence (AI) to make decisions in human resource management (HRM) raises questions of how fair employees perceive these decisions to be and whether they experience respectful treatment (i.e., interactional justice). In this experimental survey study with open-ended qualitative questions, we examine decision making in six HRM functions and manipulate the decision maker (AI or human) and decision valence (positive or negative) to determine their impact on individuals’ experiences of interactional justice, trust, dehumanization, and perceptions of decision-maker role appropriate- (...) ness. In terms of decision makers, the use of human decision makers over AIs generally resulted in better perceptions of respectful treatment. In terms of decision valence, people experiencing positive over negative decisions generally resulted in better perceptions of respectful treatment. In instances where these cases conflict, on some indicators people preferred positive AI decisions over negative human decisions. Qualitative responses show how people identify justice concerns with both AI and human decision making. We outline implications for theory, practice, and 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)

This article offers several contributions to the interdisciplinary project of responsible research and innovation in data science and AI. First, it provides a critical analysis of current efforts to establish practical mechanisms for algorithmic auditing and assessment to identify limitations and gaps with these approaches. Second, it provides a brief introduction to the methodology of argument-based assurance and explores how it is currently being applied in the development of safety cases for autonomous and intelligent systems. Third, it generalises this method (...) to incorporate wider ethical, social, and legal considerations, in turn establishing a novel version of argument-based assurance that we call ‘ethical assurance.’ Ethical assurance is presented as a structured method for unifying the myriad practical mechanisms that have been proposed. It is built on a process-based form of project governance that enlists reflective innovation practices to operationalise normative principles, such as sustainability, accountability, transparency, fairness, and explainability. As a set of interlocutory governance mechanisms that span across the data science and AI lifecycle, ethical assurance supports inclusive and participatory ethical deliberation while also remaining grounded in social and technical realities. Finally, this article sets an agenda for ethical assurance, by detailing current challenges, open questions, and next steps, which serve as a springboard to build an active (and interdisciplinary) research programme as well as contribute to ongoing discussions in policy and governance. (shrink)

The demands of fair machine learning are often expressed in probabilistic terms. Yet, most of the systems of concern are deterministic in the sense that whether a given subject will receive a given score on the basis of their traits is, for all intents and purposes, either zero or one. What, then, can justify this probabilistic talk? We argue that the statistical reference classes used in fairness measures can be understood as defining the probability that hypothetical persons, who are representative (...) of social roles, will receive certain goods. We call these hypothetical persons “representative individuals.” We claim that what we owe to actual, concrete individuals—whose individual chances of receiving the good in the system might be extreme (i.e., either zero or one)—is that their representative individual has an appropriate probability of receiving the good in question. While less immediately intuitive than other approaches, we argue that the representative individual approach has important advantages over other ways of making sense of this probabilistic talk in the context of fair machine learning. (shrink)

Generative artificial intelligence systems based on transformers, including both text-generators like GPT-4 and image generators like DALL-E 3, have recently entered the popular consciousness. These tools, while impressive, are liable to reproduce, exacerbate, and reinforce extant human social biases, such as gender and racial biases. In this paper, we systematically review the extent to which DALL-E Mini suffers from this problem. In line with the Model Card published alongside DALL-E Mini by its creators, we find that the images it produces (...) tend to represent dozens of different occupations as populated either solely by men (e.g., pilot, builder, plumber) or solely by women (e.g., hairdresser, receptionist, dietitian). In addition, the images DALL-E Mini produces tend to represent most occupations as populated primarily or solely by White people (e.g., farmer, painter, prison officer, software engineer) and very few by non-White people (e.g., pastor, rapper). These findings suggest that exciting new AI technologies should be critically scrutinized and perhaps regulated before they are unleashed on society. (shrink)

The rapidly increasing role of automation throughout the economy, culture and our personal lives has generated a large literature on the risks of algorithmic decision-making, particularly in high-stakes legal settings. Algorithmic tools are charged with bias, shrouded in secrecy, and frequently difficult to interpret. However, these criticisms have tended to focus on particular implementations, specific predictive techniques, and the idiosyncrasies of the American legal-regulatory regime. They do not address the more fundamental unease about the prospect that we might one day (...) replace judges with algorithms, no matter how fair, transparent, and intelligible they become. The aim of this paper is to propose an account of the source of that unease, and to evaluate its plausibility. I trace foundational unease with algorithmic decision-making in the law to the powerful intuition that there is a basic moral and legal difference between showing that something is true of many people just like you and showing that it is true of you. Human judgment attends to the exception; automation insists on blindly applying the rule. I show how this intuitive thought is connected to both epistemological arguments about the value of statistical evidence, as well as to court-centered conceptions of the rule of law. Unease with algorithmic decision-making in the law thus draws on an intuitive principle that underpins a disparate range of views in legal philosophy. This suggests the principle is deeply ingrained. Nonetheless, I argue that the powerful intuition is not as decisive as it may seem, and indeed runs into significant epistemological and normative challenges. At an epistemological level, I show how concerns about statistical evidence's ability to track the truth can be resolved by adopting a probabilistic, rather than modal, conception of truth-tracking. At a normative level, commitment to highly individualized decision-making co-exists with equally ingrained and competing principles, such as consistent application of law. This suggests that the “rule of law” may not identify a discrete set of institutional arrangements, as proponents of a court-centric conception would have it, but rather a more loosely defined set of values that could potentially be operationalized in multiple ways, including through some level of algorithmic adjudication. Although the prospect of replacing judges with algorithms is indeed unsettling, it does not necessarily entail unreasonable verdicts or an attack on the rule of law. (shrink)

Disagreements are widespread across the design, evaluation, and alignment pipelines of artificial intelligence (AI) systems. Yet, standard practices in AI development often obscure or eliminate disagreement, resulting in an engineered homogenization that can be epistemically and ethically harmful, particularly for marginalized groups. In this paper, we characterize this risk, and develop a normative framework to guide practical reasoning about disagreement in the AI lifecycle. Our contributions are two-fold. First, we introduce the notion of perspectival homogenization, characterizing it as a coupled (...) ethical-epistemic risk that arises when an aspect of an AI system's development unjustifiably suppresses disagreement and diversity of perspectives. We argue that perspectival homogenization is best understood as a procedural risk, which calls for targeted interventions throughout the AI development pipeline. Second, we propose a normative framework to guide such interventions, grounded in lines of research that explain why disagreement can be epistemically beneficial, and how its benefits can be realized in practice. We apply this framework to key design questions across three stages of AI development tasks: when disagreement is epistemically valuable; whose perspectives should be included and preserved; how to structure tasks and navigate trade-offs; and how disagreement should be documented and communicated. In doing so, we challenge common assumptions in AI practice, offer a principled foundation for emerging participatory and pluralistic approaches, and identify actionable pathways for future work in AI design and governance. (shrink)

Statistical fairness criteria are widely used for diagnosing and ameliorating algorithmic bias. However, these fairness criteria are controversial as their use raises several difficult questions. I argue that the major problems for statistical algorithmic fairness criteria stem from an incorrect understanding of their nature. These criteria are primarily used for two purposes: first, evaluating AI systems for bias, and second constraining machine learning optimization problems in order to ameliorate such bias. The first purpose typically involves treating each criterion as a (...) necessary condition for fairness. The second use involves treating criteria as sufficient conditions for fairness. Since the criteria are used for both roles, some researchers have treated them as both necessary and sufficient conditions, i.e., as definitions of algorithmic fairness. However, serious problems have been raised for the use of these fairness criteria. Under ordinary circumstances, it is impossible to satisfy multiple criteria at the same time. Moreover, there are counterexamples to both the sufficiency and necessity for fairness of each criterion. I argue that we should instead understand fairness criteria as merely providing evidence of fairness. In other words, satisfaction (or violation) of these criteria should be understood as potential evidence of fairness (or bias). Whether a criterion counts as evidence in a particular case will depend on stakeholders' background knowledge and the specific features of the system's task. This evidence account of fairness conditions provides guidance for recognizing both the appropriate uses and the limitations of fairness criteria. (shrink)

Uncovering the world’s ethnic inequalities is hampered by a lack of ethnicity-annotated datasets. Name-ethnicity classifiers (NECs) can help, as they are able to infer people’s ethnicities from their names. However, since the latest generation of NECs rely on machine learning and artificial intelligence (AI), they may suffer from the same racist and sexist biases found in many AIs. Therefore, this paper offers an algorithmic fairness audit of three NECs. It finds that the UK-Census-trained EthnicityEstimator displays large accuracy biases with regards (...) to ethnicity, but relatively less among gender and age groups. In contrast, the Twitter-trained NamePrism and the Wikipedia-trained Ethnicolr are more balanced among ethnicity, but less among gender and age. We relate these biases to global power structures manifested in naming conventions and NECs’ input distribution of names. To improve on the uncovered biases, we program a novel NEC, N2E, using fairness-aware AI techniques. We make N2E freely available at www.name-to-ethnicity.com. (shrink)

It is commonly accepted that clinicians are ethically obligated to disclose their use of medical machine learning systems to patients, and that failure to do so would amount to a moral fault for which clinicians ought to be held accountable. Call this ‘the disclosure thesis.’ Four main arguments have been, or could be, given to support the disclosure thesis in the ethics literature: the risk-based argument, the rights-based argument, the materiality argument and the autonomy argument. In this article, I argue (...) that each of these four arguments are unconvincing, and therefore, that the disclosure thesis ought to be rejected. I suggest that mandating disclosure may also even risk harming patients by providing stakeholders with a way to avoid accountability for harm that results from improper applications or uses of these systems. (shrink)

An algorithm that outputs predictions about the state of the world will almost always be designed with the implicit or explicit goal of outputting accurate predictions (i.e., predictions that are likely to be true). In addition, the rise of increasingly powerful predictive algorithms brought about by the recent revolution in artificial intelligence has led to an emphasis on building predictive algorithms that are fair, in the sense that their predictions do not systematically evince bias or bring about harm to certain (...) individuals or groups. This state of affairs presents two conceptual challenges. First, the goals of accuracy and fairness can sometimes be in tension, and there are no obvious normative guidelines for managing the tradeoffs between these two desiderata when they arise. Second, there are many distinct ways of measuring both the accuracy and fairness of a predictive algorithm; here too, there are no obvious guidelines on how to aggregate our preferences for predictive algorithms that satisfy disparate measures of fairness and accuracy to various extents. The goal of this paper is to address these challenges by arguing that there are good reasons for using a linear combination of accuracy and fairness metrics to measure the all-things-considered value of a predictive algorithm for agents who care about both accuracy and fairness. My argument depends crucially on a classic result in the preference aggregation literature due to Harsanyi. After making this formal argument, I apply my result to an analysis of accuracy-fairness trade-offs using the COMPAS dataset compiled by Angwin et al. (shrink)

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)

It is well-established that algorithms can be instruments of injustice. It is less frequently discussed, however, how current modes of AI deployment often make the very discovery of injustice difficult, if not impossible. In this article, we focus on the effects of algorithmic profiling on epistemic agency. We show how algorithmic profiling can give rise to epistemic injustice through the depletion of epistemic resources that are needed to interpret and evaluate certain experiences. By doing so, we not only demonstrate how (...) the philosophical conceptual framework of epistemic injustice can help pinpoint potential, systematic harms from algorithmic profiling, but we also identify a novel source of hermeneutical injustice that to date has received little attention in the relevant literature: epistemic fragmentation. Epistemic fragmentation is a structural characteristic of algorithmically-mediated environments that isolate individuals, making it more difficult to develop, uptake and apply new epistemic resources. This, in turn, can impede the identification and conceptualisation of emerging harms in these environments. We trace the occurrence of hermeneutical injustice back to the fragmentation of the epistemic experiences of individuals, who are left more vulnerable by the inability to share, compare, and learn from shared experiences. (shrink)

This paper examines Netflix's use of personalized thumbnails, framing the discussion within cybernetic and psychological contexts to explore the nature of intentionality in automated systems. While algorithms are designed to optimize user engagement, the ascription of "intentionality" to such systems requires careful delineation between human and machine agency. By analyzing the processes underlying automated personalization, this paper situates algorithmic behavior within a cybernetic framework, showing that while algorithmic ‘intentionality’ arises from emergent processes, it remains functionally distinct from moral agency. The (...) discussion contributes to the broader debate on the ethical implications of algorithmic design and communication strategies. (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)

I consider statistical criteria of algorithmic fairness from the perspective of the _ideals_ of fairness to which these criteria are committed. I distinguish and describe three theoretical roles such ideals might play. The usefulness of this program is illustrated by taking Base Rate Tracking and its ratio variant as a case study. I identify and compare the ideals of these two criteria, then consider them in each of the aforementioned three roles for ideals. This ideals program may present a way (...) forward in the normative evaluation of candidate statistical criteria of algorithmic fairness. (shrink)

Is the pool of fair predictive algorithms fair? It depends, naturally, on both the criteria of fairness and on how we pool. We catalog the relevant facts for some of the most prominent statistical criteria of algorithmic fairness and the dominant approaches to pooling forecasts: linear, geometric, and multiplicative. Only linear pooling, a format at the heart of ensemble methods, preserves any of the central criteria we consider. Drawing on work in the social sciences and social epistemology on the theoretical (...) foundations of the wisdom of crowds, we explain how our observations present an exception to the general trend of finding tradeoffs between the accuracy and fairness of forecasts. (shrink)

Progress in machine learning and artificial intelligence has spurred the widespread adoption of automated decision systems (ADS). An extensive literature explores what conditions must be met for these systems’ decisions to be fair. However, questions of legitimacy—why those in control of ADS are entitled to make such decisions—have received comparatively little attention. This paper shows that when such questions are raised theorists often incorrectly conflate legitimacy with either public acceptance or other substantive values such as fairness, accuracy, expertise or efficiency. (...) In search of better theories, we conduct a critical analysis of the philosophical literature on the legitimacy of the state, focusing on consent, public reason, and democratic authorisation. This analysis reveals that these theories require careful consideration before being applied to legitimate ADS. Specifically, we find that consent and public reason theories can legitimate some automated decision systems, but encounter serious difficulties which limit their applicability. Democratic authorisation, though not without its own challenges, emerges as a more promising and broadly applicable means of legitimation. (shrink)

People are increasingly subject to algorithmic decisions, and it is generally agreed that end-users should be provided an explanation or rationale for these decisions. There are different purposes that explanations can have, such as increasing user trust in the system or allowing users to contest the decision. One specific purpose that is gaining more traction is algorithmic recourse. We first pro- pose that recourse should be viewed as a recommendation problem, not an explanation problem. Then, we argue that the capability (...) approach provides plausible and fruitful ethical standards for re- course. We illustrate by considering the case of diversity constraints on algorithmic recourse. Finally, we discuss the significance and implications of adopting the capability approach for algorithmic recourse research. (shrink)

Discrimination is typically understood to be either direct or indirect. However, we argue that some cases that clearly are instances of discrimination are neither direct nor indirect. This is not just a logical taxonomical point. Highly salient, contemporary cases of algorithmic discrimination – a form of discrimination which was not around (or, at least, not conspicuously so) when the distinction between direct and indirect discrimination was originally articulated – are best construed as a third form of discrimination – non-direct discrimination, (...) we shall call it. If we are right, the dominant dichotomous distinction between direct and indirect discrimination should be replaced by our tripartite distinction between direct, indirect, and non-direct discrimination. We show how non-direct discrimination covers not only important types of algorithmic discrimination, but also allows us to make sense of some instances of implicit bias discrimination. (shrink)

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)

Artificial intelligent (AI) systems used in medicine are often very reliable and accurate, but at the price of their being increasingly opaque. This raises the question whether a system’s opacity undermines the ability of medical doctors to acquire knowledge on the basis of its outputs. We investigate this question by focusing on a case in which a patient’s risk of recurring breast cancer is predicted by an opaque AI system. We argue that, given the system’s opacity, as well as the (...) possibility of malfunctioning AI systems, practitioners’ inability to check the correctness of their outputs, and the high stakes of such cases, the knowledge of medical practitioners is indeed undermined. They are lucky to form true beliefs based on the AI systems’ outputs, and knowledge is incompatible with luck. We supplement this claim with a specific version of the safety condition on knowledge, Safety*. We argue that, relative to the perspective of the medical doctor in our example case, his relevant beliefs could easily be false, and this despite his evidence that the AI system functions reliably. Assuming that Safety* is necessary for knowledge, the practitioner therefore doesn’t know. We address three objections to our proposal before turning to practical suggestions for improving the epistemic situation of medical doctors. (shrink)

The increasing use of artificial intelligence (AI) systems in our daily life through various applications, services, and products explains the significance of trust/distrust in AI from a user perspective. AI-driven systems have significantly diffused into various fields of our lives, serving as beneficial tools used by human agents. These systems are also evolving to act as co-assistants or semi-agents in specific domains, potentially influencing human thought, decision-making, and agency. Trust/distrust in AI plays the role of a regulator and could significantly (...) control the level of this diffusion, as trust can increase, and distrust may reduce the rate of adoption of AI. Recently, varieties of studies have paid attention to the variant dimension of trust/distrust in AI, and its relevant considerations. In this systematic literature review, after conceptualization of trust in the current AI literature review, we will investigate trust in different types of human-Machine interaction, and its impact on technology acceptance in different domains. In addition to that, we propose a taxonomy of technical (i.e., safety, accuracy, robustness) and non-technical axiological (i.e., ethical, legal, and mixed) trustworthiness metrics, and some trustworthy measurements. Moreover, we examine some major trust-breakers in AI (e.g., autonomy and dignity threat), and trust makers; and propose some future directions and probable solutions for the transition to a trustworthy AI. (shrink)

The literature addressing bias and fairness in AI models (fair-AI) is growing at a fast pace, making it difficult for novel researchers and practitioners to have a bird’s-eye view picture of the field. In particular, many policy initiatives, standards, and best practices in fair-AI have been proposed for setting principles, procedures, and knowledge bases to guide and operationalize the management of bias and fairness. The first objective of this paper is to concisely survey the state-of-the-art of fair-AI methods and resources, (...) and the main policies on bias in AI, with the aim of providing such a bird’s-eye guidance for both researchers and practitioners. The second objective of the paper is to contribute to the policy advice and best practices state-of-the-art by leveraging from the results of the NoBIAS research project. We present and discuss a few relevant topics organized around the NoBIAS architecture, which is made up of a Legal Layer, focusing on the European Union context, and a Bias Management Layer, focusing on understanding, mitigating, and accounting for bias. (shrink)

There are calls from various sectors, including the popular press, industry, and academia, to incorporate artificial intelligence (AI)-based technologies in general, and large language models (LLMs) (such as ChatGPT and Gemini) in particular, into various spheres of the South African higher education sector. Nonetheless, the implementation of such technologies is not without ethical risks, notably those related to bias, unfairness, privacy violations, misinformation, lack of transparency, and threats to autonomy. This paper gives an overview of the more pertinent ethical concerns (...) that may result from the deployment of various current AI technologies in the South African higher education context. It provides a broad overview of the relevant AI ethics literature and distills a few general AI-ethics principles that can serve as guidelines for the ethical development, adoption, and use of AI systems. Suggestions are made as to how these might be applied to mitigate the ethical concerns in the South African higher education context. Overall, it is argued that AI technologies should only be adopted if they offer demonstrable benefits to stakeholders affected by them and that care should be taken to ensure that any potential harms are adequately addressed. (shrink)

Generative AI technologies have the potential to be socially and politically transformative. In this paper, we focus on exploring the potential impacts that Generative AI could have on the functioning of our democracies and the nature of citizenship. We do so by drawing on accounts of deliberative democracy and the deliberative virtues associated with it, as well as the reciprocal impacts that social media and Generative AI will have on each other and the broader information landscape. Drawing on this background (...) theory, we outline some of the key positive and negative impacts that Generative AI is likely to have on democratic citizenship. The political significance of these impacts suggests the need for further regulation. (shrink)

Educational disparities between traditional and non-traditional student groups in higher distance education can potentially be reduced by alleviating social identity threat and strengthening students’ sense of belonging in the academic context. We present a use case of how Learning Analytics and Machine Learning can be applied to develop and implement an algorithm to classify students as at-risk of experiencing social identity threat. These students would be presented with an intervention fostering a sense of belonging. We systematically analyze the intervention’s intended (...) positive consequences to reduce structural discrimination and increase educational equity, as well as potential risks based on privacy, data protection, and algorithmic fairness considerations. Finally, we provide recommendations for Higher Education Institutions to mitigate risk of bias and unintended consequences during algorithm development and implementation from an ethical perspective. (shrink)

This paper reviews the different meanings of vulnerability in the AI Act (AIA). We show that the AIA follows a rather established tradition of looking at vulnerability as a trait or a state of certain individuals and groups. It also includes a promising account of vulnerability as a relation but does not clarify if and how AI changes this relation. We spot the missing piece of the AIA: the lack of recognition that vulnerability is an inherent feature of all human-AI (...) interactions, varying in degree based on design choices and modes of interaction. Finally, we show how such a meaning of vulnerability may be incorporated into the AIA by interpreting the concept of “specific social situation” in Article 5 (b). (shrink)

Algorithmic risk assessment tools, such as COMPAS, are increasingly used in criminal justice systems to predict the risk of defendants to reoffend in the future. This paper argues that these tools may not only predict recidivism, but may themselves causally induce recidivism through self-fulfilling predictions. We argue that such “performative” effects can yield severe harms both to individuals and to society at large, which raise epistemic-ethical responsibilities on the part of developers and users of risk assessment tools. To meet these (...) responsibilities, we present a novel desideratum on algorithmic tools, called explainability-in-context, which requires clarifying how these tools causally interact with the social, technological, and institutional environments they are embedded in. Risk assessment practices are thus subject to high epistemic standards, which haven't been sufficiently appreciated to date. Explainability-in-context, we contend, is a crucial goal to pursue in addressing the ethical challenges surrounding risk assessment tools. (shrink)

An increasing number of regulations propose the notion of ‘AI audits’ as an enforcement mechanism for achieving transparency and accountability for artificial intelligence (AI) systems. Despite some converging norms around various forms of AI auditing, auditing for the purpose of compliance and assurance currently have little to no agreed upon practices, procedures, taxonomies, and standards. We propose the ‘criterion audit’ as an operationalizable compliance and assurance external audit framework. We model elements of this approach after financial auditing practices, and argue (...) that AI audits should similarly provide assurance to their stakeholders about AI organizations’ ability to govern their algorithms in ways that mitigate harms and uphold human values. We discuss the necessary conditions for the criterion audit, and provide a procedural blueprint for performing an audit engagement in practice. We illustrate how this framework can be adapted to current regulations by deriving the criteria on which ‘bias audits’ for hiring algorithms can be performed, as required by the recently effective New York City Local Law 144 of 2021. We conclude by offering critical discussion on the benefits, inherent limitations, and implementation challenges of applying practices of the more mature financial auditing industry to AI auditing where robust guardrails against quality assurance issues are only starting to emerge. Our discussion as informed by experiences in performing these audits in practice highlights the critical role that an audit ecosystem plays in ensuring the effectiveness of such methodology. (shrink)

Optimism about our ability to enhance societal decision‐making by leaning on Machine Learning (ML) for cheap, accurate predictions has palled in recent years, as these ‘cheap’ predictions have come at significant social cost, contributing to systematic harms suffered by already disadvantaged populations. But what precisely goes wrong when ML goes wrong? We argue that, as well as more obvious concerns about the downstream effects of ML‐based decision‐making, there can be moral grounds for the criticism of these predictions themselves. We introduce (...) and defend a theory of predictive justice, according to which differential model performance for systematically disadvantaged groups can be grounds for moral criticism of the model, independently of its downstream effects. As well as helping resolve some urgent disputes around algorithmic fairness, this theory points the way to a novel dimension of epistemic ethics, related to the recently discussed category of doxastic wrong. (shrink)

This article presents a fairness principle for evaluating decision-making based on predictions: a decision rule is unfair when the individuals directly impacted by the decisions who are equal with respect to the features that justify inequalities in outcomes do not have the same statistical prospects of being benefited or harmed by them, irrespective of their socially salient morally arbitrary traits. The principle can be used to evaluate prediction-based decision-making from the point of view of a wide range of antecedently specified (...) substantive views about justice in outcome distributions. (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)

From music recommendation to assessment of asylum applications, machine-learning algorithms play a fundamental role in our lives. Naturally, the rise of AI implementation strategies has brought to public attention the ethical risks involved. However, the dominant anti-discrimination discourse, too often preoccupied with identifying particular instances of harmful AIs, has yet to bring clearly into focus the more structural roots of AI-based injustice. This paper addresses the problem of AI-based injustice from a distinctively epistemic angle. More precisely, I argue that the (...) injustice generated by the implementation of AI machines in our societies is, in some paradigmatic cases, also a form of epistemic injustice. With a particular focus on AIs employed as gatekeepers of our epistemic resources, this paper shows how their epistemically conformist behaviour is responsible for the marginalisation and the ostracism of minoritarian perspectives. Because it clarifies key structural flaws and weaknesses of current AI design, this paper helps make headway in critical discussion of current AI technologies. And because it forges new theoretical tools to understand forms of epistemic oppression, this paper also contributes to the advancement of feminist theorisation. (shrink)

We introduce a fairness criterion that we call Spanning. Spanning i) is implied by Calibration, ii) retains interesting properties of Calibration that some other ways of relaxing that criterion do not, and iii) unlike Calibration and other prominent ways of weakening it, is consistent with Equalized Odds outside of trivial cases.

We reply to Eva's comment on our "New Possibilities for Fair Algorithms," comparing and contrasting our Spanning criterion with his suggested Spacing criterion.

The now well-known impossibility results of algorithmic fairness demonstrate that an error-prone predictive model cannot simultaneously satisfy two plausible conditions for group fairness apart from exceptional circumstances where groups exhibit equal base rates. The results sparked, and continue to shape, lively debates surrounding algorithmic fairness conditions and the very possibility of building fair predictive models. This article, first, highlights three underlying points of disagreement in these debates, which have led to diverging assessments of the feasibility of fairness in prediction-based decision-making. (...) Second, the article explores whether and in what sense fairness as defined by the conjunction of the implicated fairness conditions is (un)attainable. Drawing on philosophical literature on the concept of feasibility and the role of feasibility in normative theory, I outline a cautiously optimistic argument for the _diachronic feasibility_ of fairness. In line with recent works on the topic, I argue that fairness can be _made possible_ through collective efforts to eliminate inequalities that feed into local decision-making procedures. (shrink)

There is a theorem that shows that it is impossible for an algorithm to jointly satisfy the statistical fairness criteria of Calibration and Equalized Odds non-trivially. But what about the recently advocated alternative to Calibration, Base Rate Tracking? Here we show that Base Rate Tracking is strictly weaker than Calibration, and then take up the question of whether it is possible to jointly satisfy Base Rate Tracking and Equalized Odds in non-trivial scenarios. We show that it is not, thereby establishing (...) an even more general impossibility theorem. (shrink)