A distinction has been drawn in fair machine learning research between 'group' and 'individual' fairness measures. Many technical papers assume that both are important, but conflicting, propose ways to minimise the trade-offs these This paper argues this apparent conflict is based on a misconception. It draws discussions from within research, political legal philosophy, argue individual group not fundamentally conflict. First, it outlines accounts of egalitarian which encompass plausible...

Calls for heightened consideration of fairness and accountability in algorithmically-informed public decisions---like taxation, justice, child protection---are now commonplace. How might designers support such human values? We interviewed 27 sector machine learning practitioners across 5 OECD countries regarding challenges understanding imbuing values into their work. The results suggest a disconnect between organisational institutional realities, constraints needs, those addressed by...

Tracking is a highly privacy-invasive data collection practice that has been ubiquitous in mobile apps for many years due to its role supporting advertising-based revenue models. In response, Apple introduced two significant changes with iOS 14: App Transparency (ATT), mandatory opt-in system enabling tracking on iOS, and Privacy Nutrition Labels, which disclose what kinds of each app processes. So far, the impact these individual privacy control not well understood. This paper addresses...

Discriminatory bias in algorithmic systems is widely documented. How should the law respond? A broad consensus suggests approaching issue principally through lens of indirect discrimination, focusing on systems' impact. In this article, we set out to challenge analysis, arguing that while discrimination has an important role play, a narrow focus regime context machine learning algorithms both normatively undesirable and legally flawed. We illustrate how certain forms frequently deployed...

Abstract Arguments in favor of tempering algorithmic decision making with human judgment often appeal to concepts and criteria derived from legal philosophy about the nature law reasoning, arguing that systems cannot satisfy them (but humans can). Such arguments make implicit notion each case needs be assessed on its own merits, without comparison or generalization previous cases. This article argues this individual justice can only meaningfully served through judgment. It distinguishes...

Cite as: Veale, Michael and Binns, Reuben (2017) Fairer machine learning in the real world: Mitigating discrimination without collecting sensitive data. Big Data & Society 4(2). doi:10.1177/2053951717743530Decisions based on algorithmic, models can be unfair, reproducing biases historical data used to train them. While computational techniques are emerging address aspects of these concerns through communities such as discrimination-aware mining (DADM) fair, accountable transparent...

Since the emergence of generative AI, creative workers have spoken up about career-based harms they experienced arising from this new technology. A common theme in these accounts harm is that AI models are trained on workers' output without their consent and giving credit or compensation to original creators. This paper reports findings 20 interviews with three domains: visual art design, writing, programming. We investigate gaps between current governance strategies, what want out...

Independent algorithm audits hold the promise of bringing accountability to automated decision-making. However, third-party are often hindered by access restrictions, forcing auditors rely on limited, low-quality data. To study how these limitations impact research integrity, we conduct audit simulations two realistic case studies for recidivism and healthcare coverage prediction. We examine accuracy estimating group parity metrics across three levels access: (a) aggregated statistics, (b)...

As AI changes the way decisions are made in organizations and governments, it is ever more important to ensure that these systems work according values diverse users groups find important. Researchers have proposed numerous algorithmic techniques formalize statistical fairness notions, but emerging suggests must account for real-world contexts which they will be embedded order actually fairly. These findings call an expanded research focus beyond includes fundamental understandings of human...

BackgroundData protection law provides a set of rights and obligations in relation to the processing personal data.Amongst its substantive principles, such as lawfulness, fairness, transparency, various procedural elements risk-based measures that apply data processing, it also addresses use automated decision-making systems.Provisions around decisions are not new, having been part toolbox for several decades. 1 The Data Protection Directive 1995 regulated decision-making; 2 however practice...

What does it mean for an algorithmic decision-making system to be "fair" or "non-discriminatory" in terms that can operationalized? Providing a rigorous understanding of these has long been preoccupation moral and political philosophers. This article draws on such work elucidate emerging debates about fair algorithms.

Emerging scholarship suggests that the EU legal concept of direct discrimination - where a person is given different treatment on grounds protected characteristic may apply to various algorithmic decision-making contexts. This has important implications: unlike indirect discrimination, there generally no 'objective justification' stage in framework, which means deployment directly discriminatory algorithms will usually be unlawful per se. In this paper, we focus most likely candidate for...

The European Commission proposed a Directive on Platform Work at the end of 2021. While much attention has been placed its effort to address misclassification employed as self-employed, it also contains ambitious provisions for regulation algorithmic management prevalent these platforms. Overall, are well-drafted, yet they require extra scrutiny in light fierce lobbying and resistance will likely encounter legislative process, implementation enforcement. In this article, we place proposal...

Previous literature on 'fair' machine learning has appealed to legal frameworks of discrimination law motivate a variety and fairness metrics de-biasing measures. Such work typically applies the US doctrine disparate impact rather than alternative treatment, scholars EU have largely followed along similar lines, addressing algorithmic bias as form indirect direct discrimination. In recent work, we argued that such focus is unduly narrow in context European law: certain forms will constitute...

•Provisions in many data protection laws require a legal basis, or at the very least safeguards, for significant, solely automated decisions; Article 22 of GDPR is most notable. •Little attention has been paid to light decision-making processes with multiple stages, potentially both manual and automated, which together might impact upon decision subjects different ways. •Using stylised examples grounded real-world systems, we raise five distinct complications relating interpreting context...

There are various arguments in favour of tempering algorithmic decision-making with human judgement. One common family appeal to concepts and criteria derived from legal philosophy about the nature law reasoning, argue that systems cannot satisfy them (but humans can). This paper argues among latter arguments, there is often an implicit notion each case needs be assessed on its own merits, without comparison or generalisation previous cases. ‘individual justice’ has featured jurisprudential...

Critical examinations of AI systems often apply principles such as fairness, justice, accountability, and safety, which is reflected in regulations the EU Act. Are sufficient to promote design that support human flourishing? Even if a system some sense fair, just, or 'safe', it can nonetheless be exploitative, coercive, inconvenient, otherwise conflict with cultural, individual, social values. This paper proposes dimension interactional ethics thus far overlooked: ways should treat beings....

1. For anyone browsing the UK government online tendering portal ( ), steady march of algorithms in public sector is plain to see. Searches for “predictive analytics...

The web routinely spreads personal data from one jurisdiction to another, where levels of legal protection over such vary. This raises the potential for some jurisdictions become "data havens" specialising in either strong data, or allowing its unrestricted use [5],[3]. In order promote interoperability and harmonisation, with similar may approve each other's regimes, lifting restrictions on international transfers. article presents a quantitative analysis 16,000 transfer arrangements made...

A distinction has been drawn in fair machine learning research between `group' and `individual' fairness measures. Many technical papers assume that both are important, but conflicting, propose ways to minimise the trade-offs these This paper argues this apparent conflict is based on a misconception. It draws theoretical discussions from within research, political legal philosophy, argue individual group not fundamentally conflict. First, it outlines accounts of egalitarian which encompass...

A large portion of the content, recommendations and advertisements shown on web are targeted, based a profile an individual user. This paper explores two ways creating using such profiles. Behavioural profiling - commonly used technique which makes inferences individual's previous activity is compared to what I call Self-Authored Interest (SAI) profiling, information explicitly volunteered controlled by individual. present results experimental study comparing effectiveness systems in...

A recently published pre-print titled 'GDPR and the Lost Generation of Innovative Apps' by Jan{\ss}en et al. observes that a third apps on Google Play Store disappeared from this app store around introduction GDPR in May 2018. The authors deduce 'that is cause'. effects economy are an important field to study. Unfortunately, paper currently lacks control condition key variable. As result, exits reported likely overestimated, as we will discuss. We believe there other factors which may better...