6 results
Now showing1 - 6 of 6
- Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Enhancing AI interpretability through localised architecturesRecent advances in generative AI, especially powerful Large Language Models (LLMs), raise concerns over the interpretability, safety and sustainability of these large and opaque AI models. The power of such architectures is derived not only from the scalability of deep neural networks (DNNs), but also massively parallel hardware such as GPU clusters. The diffuse nature of DNNs gives them great function-approximation capability when provided with sufficient training data but imposes a cost in interpretability and computational efficiency. Observing that localised machine learning (ML) models tend to be more interpretable and computationally efficient than DNNs on small datasets, we reason by analogy that similar advantages may apply to specific localised hardware ML architectures. We argue that localised architectures with lower bandwidth but higher expressivity per node have the potential to be fundamentally more interpretable than DNNs running on GPU clusters while remaining competitive for smaller datasets. We then survey the suitability of various hardware ML paradigms for implementing such localised architectures and evaluate their per-node expressivity, energy efficiency and practical maturity of the technology required.Publicationtype: Conference Paper - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Inspirational responsibility, intellectual property, and AI trainingGenerative AI systems, including large language models and image or video generators, are typically trained on data scraped from the internet. Since the creators of the works used for AI training have often neither consented to nor been compensated for such use, this has led to allegations of “theft”. At the same time, since generative AI models do not store copies of their training data, there is a widespread view that they no not necessarily infringe copyright. Moreover, in human creative practice, drawing inspiration from and learning from the works of others is common, if not unavoidable. The central challenge, then, is to formulate an intellectual property right that allows creators to restrict AI training without limiting human learning and creativity. McIntyre ( 2026 ) proposes new strategies for addressing this challenge, most notably through an argument relying on proprietary rights. Critically engaging with this approach, I argue that it overlooks the power dynamics in the art world and leaves creators in a vulnerable and dependent position. Instead, I outline a different pathway to restricting AI training that centres on what I call “inspirational responsibility”. This concept captures the responsibility creators bear for how they use and engage with the work of others. Since human creators can be held inspirationally responsible, whereas AI models cannot, the notion of inspirational responsibility provides a normatively relevant basis for restricting AI training without imposing parallel constraints on human learning and creativity.Publicationtype: Journal Article20 16 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , Promises and risks in using LLMs in scientific review(IEEE, 2026-03); ; ; ; The current scientific peer review process faces increasing strain and is becoming difficult to sustain. While large language models (LLMs) can help reviewers by summarizing papers, assessing clarity, and generating timely feedback, the research community remains cautious about relying on LLM-based reviews. In this paper, we explore how LLMs could support and potentially improve parts of the peer review process. We examine both the benefits and limitations by analysing LLM-generated reviews of our own previously published papers, comparing them with human reviewer feedback, identifying points of agreement and disagreement, and considering requirements for the responsible and ethical use of LLMs in peer review.Publicationtype: Conference Paper26 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: without files , Think first, code later: enhancing student reasoning and learning in practical engineering education through guided generative AI(2026); ; ; 5 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , The ethics of analog AIAI ethics has matured as a field, yet its concepts and tools have been developed almost entirely with digital AI in mind. Emerging “analog” AI approaches, that is, systems that compute over continuous, rather than binary, signals, challenge this digital default. Analog AI promises substantial energy savings and alternative modes of information processing. However, it also departs in ethically significant ways from the widely discussed digital AI systems such that its differences from digital AI could mean that current ethical frameworks may not address analog AI’s unique challenges. This paper highlights how the distinctive characteristics of analog AI raise new ethical questions that require careful consideration. We examine four key areas of AI ethics, namely fairness, privacy, explainability, and safety, and propose a roadmap for the ethical examination of analog AI. Our aim is to lay the foundation for an important but hitherto unrecognized subfield within AI ethics: the ethics of analog AI.Publicationtype: Journal Article19 8 - Some of the metrics are blocked by yourconsent settings
Item type:Publication, Access status: with files , Rethinking trust in responsible AI(CEUR-WS, 2025); ; ; ; Trust is widely recognized as a core principle of Responsible AI, yet its interpretation varies significantly across disciplines. This paper examines how computer science, sociology, philosophy, and law conceptualize trust in AI systems, highlighting both tensions and complementarities. From a computer science perspective, trust is often approached as a set of system-level properties that should be formalized and evaluated with metrics. In contrast, the social sciences and humanities emphasize its relational, normative, and institutional dimensions. We argue that trust cannot be reduced to a single system property or technical measure, as it emerges from social-technical interactions involving users, developers, legal norms, and social expectations. To support interdisciplinary dialogue, we propose treating trust as a boundary concept that enables cooperation across epistemic communities acknowledging conceptual differences. Trust is widely recognized as a core principle of Responsible AI, yet its interpretation varies significantly across disciplines. This paper examines how computer science, sociology, philosophy, and law conceptualize trust in AI systems, highlighting both tensions and complementarities. From a computer science perspective, trust is often approached as a set of system-level properties that should be formalized and evaluated with metrics. In contrast, the social sciences and humanities emphasize its relational, normative, and institutional dimensions. We argue that trust cannot be reduced to a single system property or technical measure, as it emerges from social-technical interactions involving users, developers, legal norms, and social expectations. To support interdisciplinary dialogue, we propose treating trust as a boundary concept that enables cooperation across epistemic communities acknowledging conceptual differences.Publicationtype: Conference Paper22 4