Personalized recommender systems are rapidly expanding in applications across the web, encompassing social media, online shopping, search engine results, and more. Personalization provides significant value in navigating the vast array of items available, as it can help to disambiguate search terms and generally provide more accurate recommendations. However, alongside this growth, there has been increased recognition of the potential for algorithmic systems to exhibit and perpetuate biases. In particular, personalized recommender systems may risk unfairness in recommendations, leading to biases in access to information and epistemic utility. In this work, we differentiate between idealized and realized notions of personalization in the extreme to motivate our fairness definition: an equity-based approach to epistemic utility. Our approach bypasses many issues with comparative notions of fairness and instead aims to guarantee all users baseline access to information. We discuss key trade-offs and challenges for operationalizing our definition and provide policy recommendations towards achieving fairness in personalized systems.
All fairness algorithms unavoidably rely on normative assumptions about fair treatment, yet these assumptions often remain implicit. We argue that these assumptions should be formalized as explicit, auditable policies and introduce FAIR-SHEPHERD , a structural policy-based framework for transparent fairness in real-world settings with noisy labels and shifting distributions. FAIR-SHEPHERD uses gradients as attribution signals, encoded in a Structural Fairness Specification (SFS) that defines vertical coherence and orthogonality to vulnerable proxies. We introduce SFS metrics: Vertical Coherence Score (VCS) to measure directional coherence across adjacent normative slices, and Horizontal Leakage Score (HLS), augmented by a signed directional variant, to detect gradient alignment with policy-declared vulnerable or proxy directions. To enforce these policies, we develop S-agnostic tools that optimize policy-aligned gradient objectives using lattice-defined proxies. We demonstrate that outcome-based auditing is brittle to measurement error: under 10% label noise on Adult, Worst-Group AUC for ERM drops by 0.159. In contrast, our gradient-based structural metrics provide a label-agnostic audit of the decision logic, remaining stable even when evaluation labels are corrupted. Specialized fairness baselines including ARL, JTT, and GoG retain substantial structural leakage on COMPAS, with HLS values from about 0.47 to 0.71. Some also reduce EOD relative to ERM, which shows that outcome and structural criteria can diverge. Using our S-agnostic Gradient Penalty tools, we reduce policy-specified structural leakage by roughly 85% on COMPAS (0.478 → 0.073) and over 94% on Adult (0.138 → 0.008) while retaining competitive AUC and Worst-Group AUC. We have released our source code to facilitate further research here.
Persuasive technologies are increasingly embedded in areas such as health, education, and e-commerce, raising ethical concerns about their influence on user autonomy, privacy, consent, and trust. While prior theoretical work emphasizes the importance of ethical design for building trustworthy systems, there remains limited empirical evidence quantifying how specific ethical factors shape user trust in persuasive technologies. This study examines the relationship between users’ ethical perceptions and trust in persuasive technologies using survey data analyzed through Partial Least Squares Structural Equation Modeling. The results show that autonomy and security are significant predictors of trust, whereas privacy, consent, and transparency do not have a meaningful influence in this context. These findings suggest that not all ethical design dimensions contribute equally to trust formation, underscoring the need for more targeted and context-sensitive approaches to ethical system design and governance. The study concludes with design and policy recommendations, including autonomy-preserving interaction mechanisms, visible security assurances, and governance tools such as digital nutrition labels and ethical impact statements to support trustworthy, persuasive technologies.
Rapid advancements of artificial intelligence (AI) call for a robust governance approach that keeps pace with technological progression and addresses the needs of society, institutions, and innovators. Emerging AI methods have demonstrated various successes, as well as concerning error patterns ranging from discrimination to hallucinations. Nonetheless, there is substantial momentum in the direction of broad and rapid deployment in critical use cases, including strategic decision support tools for healthcare, raising significant ethical concerns for responsible adoption. We propose an adaptive governance framework that views governance as a dynamic, human-centred control policy that explicitly integrates theories of change about technology acceptance, adoption, and rejection as part of a feedback loop to ensure ethical, secure, and effective implementation of the health AI (HAI) environment. The goal is to inform how policies and regulations can be optimised to balance responsible deployment for the greater good of society without stifling technological innovation.
The aim of this special section is to curate an interdisciplinary conversation between computer and information scientists and philosophers. The contributions explore how Fairness and Machine Learning: Limitations and Opportunities can be used as a springboard for examining the role of empirical computer science in the political philosophy of computing, and vice versa. Our hope is that this symposium can spark a multidisciplinary dialogue around questions that are at root philosophical but touch on technical aspects, the socio-technical systems in which ML is embedded and ML's social impacts. The aim of this special section is to curate an interdisciplinary conversation between computer and information scientists and philosophers. The contributions explore how Fairness and Machine Learning: Limitations and Opportunities can be used as a springboard for examining the role of empirical computer science in the political philosophy of computing, and vice versa. Our hope is that this symposium can spark a multidisciplinary dialogue around questions that are at root philosophical but touch on technical aspects, the socio-technical systems in which ML is embedded and ML's social impacts.
