Abstract The manner in which the software platforms and tools that surround us are built and marketed is designed to lead us not into conversations, but silences. It has been 20 years since Cass Sunstein first warned us of the manner in which new media lead us to choose as consumers, rather than citizens, and yet we still find that the dominant paradigm for technology adoption primes us to consider our decision-making around ICT tools as independent and neutral, needing only to meet a measure of value for money. Because of the all-pervasive nature of code and its compiled products, however, this particular socio-technical imaginary brings with it an exceptional power to alienate us not just from each other, but, in a post-Marxian twist, from the very tools, we might use to manage this deterioration in interpersonal exchange. This is, in particular, the case with artificial intelligence, due to the opacity and complexity of the methods deployed, and the dark ethical underbelly of its effects on power, labour, and rights. This article will look at this and other imaginaries that hold us back from forging a cultural relationship to AI and other forms of coded culture that is not just human-centred, but community aware, embodying the oft-cited values of human-centredness and trustworthiness. To do this, it will propose strategies by which to recentre code as a site for the discursive activity of collective sensemaking. Starting from the roots of the human-readable in code, the argument coalesces around the details of two experimental cases that have sought to foster human conversations that include code as an active participant.

Abstract This paper develops a normative framework for the alignment of artificial intelligence (AI) systems with human agency. Moving beyond models that treat values as inferable data, it reconceptualizes alignment as the structural integration of AI within human normative domains. The argument unfolds in three steps. First, it introduces the concept of extended human agency , viewing AI as a teleological extension of human purposiveness rather than an autonomous moral subject. Second, it grounds this integration in practical autonomy —the human capacity to act for reasons and to assume responsibility within justificatory structures. Third, it proposes the design concept of a normative interface : a mediating architecture that connects machine behavior with human norms, ensuring teleological coherence, normative intelligibility, and accountability. Law serves as a paradigmatic case, demonstrating how institutionalized practices of justification can guide the ethical embedding of AI. According to this model, alignment is not achieved through internalized value learning but through participation in norm-governed action spaces, where human agents remain the ultimate bearers of responsibility. Finally, this paper argues that responsible alignment requires a normative and not merely technical architecture, linking artificial systems to the historical and institutional forms of human rationality.

Abstract Along with general-purpose generative AI (GenAI) tools like ChatGPT, educational GenAI tools are being increasingly launched to support teachers in enhancing their instructional practices and productivity. To establish a relationship of trust with teachers, tools need to be transparent and provide clear disclosures regarding their underlying mechanisms, training data and processes, origin of information, privacy policies, and limitations. This document analysis examined the transparency of 20 GenAI tools for teachers by analyzing publicly available information on the tools' websites. Findings indicate a significant lack of transparency across these tools. Most failed to provide explicit details about their AI mechanisms, and none disclosed their training procedures or data sources. While several tools assured users that their data would not be used for further training, discussions on limitations and potential risks were largely absent. In addition, necessary information was often difficult to locate and understand. These findings underscore the need for increased transparency in GenAI tools, so that teachers can critically assess the tools’ functionalities, responses, reliability, and suitability for educational use. The study recommends that developers of educational GenAI provide clearer, more accessible disclosures regarding their tools to build a relationship of trust with educators. Educators should actively seek important information before integrating GenAI tools into their teaching. Greater transparency will facilitate informed decision-making of educators, ultimately enhancing the responsible and effective use of GenAI in education.

Abstract The growing integration of artificial intelligence (AI)-based brain–computer interface (BCI) systems into healthcare intensifies a fundamental legal question: who should bear responsibility for damage caused by adaptive technologies that directly interact with the human brain? Whilst the European Union’s new Product Liability Directive (2024/2853) represents a significant step towards modernising liability for digital products, it remains largely oriented towards identifiable technical defects and relatively stable products. This article examines AI-driven BCI systems through the lens of algorithmic normativity and introduces the concept of a Reflexive Normative Cascade to analyse how responsibility evolves from pre-legal normative expectations to societal experiential feedback and, potentially, to reactive legal crystallisation. It argues that although societal experience clearly articulates concerns relating to autonomy, mental integrity, and transparency, these insights are not always successfully translated into stable liability rules under the current product liability framework. In particular, damages arising from normative design choices embedded in algorithms, continuous post-market adaptation, and distributed responsibility chains expose persistent accountability gaps for both injured parties and producers. By situating the Product Liability Directive within this reflexive process, the article contends that liability for AI-driven BCIs cannot be fully addressed through static doctrinal tools alone and calls for a dynamic normative framework integrating transparency, traceability, and complementary insurance mechanisms.

Abstract Despite being a well-studied problem, bias continues to affect modern Computer Vision (CV) datasets, models and systems, including Generative AI models that have been found to replicate and amplify harmful biases and stereotypes. In this work, we focus on framing bias : a type of bias that arises from the way images convey meaning. We introduce a semiotic perspective that treats image datasets as texts and argues that framing bias is not a statistical accident due to sampling noise or imbalance, but an inherent property of image datasets as meaning-making systems. We show that co-occurrence of visual elements is the primary mechanism through which image datasets frame concepts, and that not only do individual images contribute to dataset framing through co-occurrence, but dataset-level framing also shapes the interpretation of individual images. As a consequence, every image dataset embodies a point of view and cannot be fully unbiased . We illustrate these claims through a case study of the Visual Genome dataset, revealing a framing of human activities that over-represents leisure, systematically excludes everyday labour, and is characterised by a predominantly American/Western viewpoint. Finally, we reflect on the epistemological implications of analysing image datasets, highlighting the mediated and interpretive nature of knowledge produced through AI systems, and propose revisionism and source criticism as epistemological paradigms to address these issues.

Abstract This paper offers an exploratory examination of the emerging discourse and practices on decolonising AI ethics. Whilst the field of AI ethics has made substantial progress in proposing normative frameworks for responsible innovation, these frameworks remain predominantly grounded in Western epistemological traditions, foundations which have limited the capacity of AI ethics to engage with plural ontologies and non-Western moral traditions, particularly those emerging from the Global South. Drawing from decolonial and critical data studies, this paper interrogates what an alternative, decolonial orientation to AI ethics might entail in both conceptual and practical terms. We contend that mainstream AI ethics frameworks fail to recognise epistemic diversity and to establish dialogical relations with non-Western knowledge systems, thereby reproducing colonial hierarchies of thought and governance within technological design and deployment. An answer to the question of decolonising AI ethics is not simple or straightforward and will, therefore, not be single handedly delivered in this paper, whilst we acknowledge the situatedness and limitations of our study. Whilst we understand that there are many other realms to examine for a decolonial AI ethics (such as critical infrastructure, labour, and political economy), we decide to focus on the following reflections: (1) AI as a means to redress power asymmetries and structural exploitation; (2) AI grounded in local data sovereignty and equitable epistemic participation; and (3) AI conceived as relationally entangled with humans, more-than-humans, and the Earth. These reflections foreground the material and ecological dimensions of data—its extractive infrastructures, environmental externalities, and geo-political inequalities—as central to the ethical assessment of AI. Through a literature-based exploration of these domains, we argue that decolonising AI ethics requires moving beyond the universalist and procedural tendencies of Western ethical paradigms towards pluriversal, situated, and ecologically embedded forms of ethical reasoning. The paper concludes by outlining how these reflections can serve as a theoretical scaffold for rearticulating AI ethics in and from the Global South, resisting the uncritical replication of Northern frameworks and opening pathways for epistemic plurality in global AI governance.