Research Papers

Model editing offers a low-cost technique to inject or correct a particular behavior in a pre-trained model without extensive retraining, supporting applications such as factual correction and bias mitigation. Despite this common practice, it remains unknown whether edits persist after fine-tuning or whether they are inadvertently reversed. This question has fundamental practical implications. For example, if fine-tuning removes prior edits, it could serve as a defence mechanism against hidden malicious edits. Vice versa, the unintended removal of edits related to bias mitigation could pose serious safety concerns. We systematically investigate the interaction between model editing and fine-tuning in the context of T2I diffusion models, which are known to exhibit biases and generate inappropriate content. Our study spans two T2I model families (Stable Diffusion and FLUX), two sota editing techniques, and three fine-tuning methods (DreamBooth, LoRA, and DoRA). Through an extensive empirical analysis across diverse editing tasks and evaluation metrics, our findings reveal a trend: edits generally fail to persist through fine-tuning, even when fine-tuning is tangential or unrelated to the edits. Notably, we observe that DoRA exhibits the strongest edit reversal effect. At the same time, among editing methods, UCE demonstrates greater robustness, retaining significantly higher efficacy post-fine-tuning compared to ReFACT. These findings highlight a crucial limitation in current editing methodologies, emphasizing the need for more robust techniques to ensure reliable long-term control and alignment of deployed AI systems. These findings have dual implications for AI safety: they suggest that fine-tuning could serve as a remediation mechanism for malicious edits while simultaneously highlighting the need for re-editing after fine-tuning to maintain beneficial safety and alignment properties.

Patient-Centred Explainability in IVF Outcome Prediction

Adarsa Sivaprasad, Ehud Reiter, David McLernon +3

This paper evaluates the user interface of an in vitro fertility (IVF) outcome prediction tool, focussing on its understandability for patients or potential patients. We analyse four years of anonymous patient feedback, followed by a user survey and interviews to quantify trust and understandability. Results highlight a lay user's need for prediction model \emph{explainability} beyond the model feature space. We identify user concerns about data shifts and model exclusions that impact trust. The results call attention to the shortcomings of current practices in explainable AI research and design and the need for explainability beyond model feature space and epistemic assumptions, particularly in high-stakes healthcare contexts where users gather extensive information and develop complex mental models. To address these challenges, we propose a dialogue-based interface and explore user expectations for personalised explanations.

We investigate whether Large Language Models (LLMs) exhibit human-like cognitive patterns under four established frameworks from psychology: Thematic Apperception Test (TAT), Framing Bias, Moral Foundations Theory (MFT), and Cognitive Dissonance. We evaluated several proprietary and open-source models using structured prompts and automated scoring. Our findings reveal that these models often produce coherent narratives, show susceptibility to positive framing, exhibit moral judgments aligned with Liberty/Oppression concerns, and demonstrate self-contradictions tempered by extensive rationalization. Such behaviors mirror human cognitive tendencies yet are shaped by their training data and alignment methods. We discuss the implications for AI transparency, ethical deployment, and future work that bridges cognitive psychology and AI safety

This vision paper presents initial research on assessing the robustness and reliability of AI-enabled systems, and key factors in ensuring their safety and effectiveness in practical applications, including a focus on accountability. By exploring evolving definitions of these concepts and reviewing current literature, the study highlights major challenges and approaches in the field. A case study is used to illustrate real-world applications, emphasizing the need for innovative testing solutions. The incorporation of accountability is crucial for building trust and ensuring responsible AI development. The paper outlines potential future research directions and identifies existing gaps, positioning robustness, reliability, and accountability as vital areas for the development of trustworthy AI systems of the future.

In this workshop paper, we discuss the potential for measures of user-centric benefits (such as emotional well-being) that could be explored when evaluating explainable AI (XAI) systems within the arts. As a background to this, we draw from our recent review of creativity support tool (CST) evaluations, that found a paucity of studies evaluating CSTs for user-centric measures that benefit the user themselves. Specifically, we discuss measures of: (1) developing intrinsic abilities, (2) emotional well-being, (3) self-reflection, and (4) self-perception. By discussing these user-centric measures within the context of XAI and the arts, we wish to provoke discussion regarding the potential of such measures.

This paper examines the views of software providers in the German dairy industry with regard to dairy farmers' needs for explanation of digital decision support systems. The study is based on mastitis detection in dairy cows using a hypothetical herd management system. We designed four exemplary explanation formats for mastitis assessments with different types of presentation (textual, rule-based, herd comparison, and time series). In our previous study, 14 dairy farmers in Germany had rated these formats in terms of comprehensibility and the trust they would have in a system providing each format. In this study, we repeat the survey with 13 software providers active in the German dairy industry. We ask them how well they think the formats would be received by farmers. We hypothesized that there may be discrepancies between the views of both groups that are worth investigating, partly to find reasons for the reluctance to adopt digital systems. A comparison of the feedback from both groups supports the hypothesis and calls for further investigation. The results show that software providers tend to make assumptions about farmers' preferences that are not necessarily accurate. Our study, although not representative due to the small sample size, highlights the potential benefits of a thorough user requirements analysis (farmers' needs) to improve software adaptation and user acceptance.

Interpretable Classification of Levantine Ceramic Thin Sections via Neural Networks

Sara Capriotti, Alessio Devoto, Simone Scardapane +2

Classification of ceramic thin sections is fundamental for understanding ancient pottery production techniques, provenance, and trade networks. Although effective, traditional petrographic analysis is time-consuming. This study explores the application of deep learning models, specifically Convolutional Neural Networks (CNNs) and Vision Transformers (ViTs), as complementary tools to support the classification of Levantine ceramics based on their petrographic fabrics. A dataset of 1,424 thin section images from 178 ceramic samples belonging to several archaeological sites across the Levantine area, mostly from the Bronze Age, with few samples dating to the Iron Age, was used to train and evaluate these models. The results demonstrate that transfer learning significantly improves classification performance, with a ResNet18 model achieving 92.11% accuracy and a ViT reaching 88.34%. Explainability techniques, including Guided Grad-CAM and attention maps, were applied to interpret and visualize the models' decisions, revealing that both CNNs and ViTs successfully focus on key mineralogical features for the classification of the samples into their respective petrographic fabrics. These findings highlight the potential of explainable AI in archaeometric studies, providing a reproducible and efficient methodology for ceramic analysis while maintaining transparency in model decision-making.

Mental health is a growing global concern, prompting interest in AI-driven solutions to expand access to psychosocial support. Peer support, grounded in lived experience, offers a valuable complement to professional care. However, variability in training, effectiveness, and definitions raises concerns about quality, consistency, and safety. Large Language Models (LLMs) present new opportunities to enhance peer support interactions, particularly in real-time, text-based interactions. We present and evaluate an AI-supported system with an LLM-simulated distressed client (SimClient), context-sensitive LLM-generated suggestions (Suggestions), and real-time emotion visualisations. 2 mixed-methods studies with 12 peer supporters and 6 mental health professionals (i.e., experts) examined the system's effectiveness and implications for practice. Both groups recognised its potential to enhance training and improve interaction quality. However, we found a key tension emerged: while peer supporters engaged meaningfully, experts consistently flagged critical issues in peer supporter responses, such as missed distress cues and premature advice-giving. This misalignment highlights potential limitations in current peer support training, especially in emotionally charged contexts where safety and fidelity to best practices are essential. Our findings underscore the need for standardised, psychologically grounded training, especially as peer support scales globally. They also demonstrate how LLM-supported systems can scaffold this development--if designed with care and guided by expert oversight. This work contributes to emerging conversations on responsible AI integration in mental health and the evolving role of LLMs in augmenting peer-delivered care.

Ensuring responsible use of artificial intelligence (AI) has become imperative as autonomous systems increasingly influence critical societal domains. However, the concept of trustworthy AI remains broad and multi-faceted. This thesis advances knowledge in the safety, fairness, transparency, and accountability of AI systems. In safety, we extend classical deterministic shielding techniques to become resilient against delayed observations, enabling practical deployment in real-world conditions. We also implement both deterministic and probabilistic safety shields into simulated autonomous vehicles to prevent collisions with road users, validating the use of these techniques in realistic driving simulators. We introduce fairness shields, a novel post-processing approach to enforce group fairness in sequential decision-making settings over finite and periodic time horizons. By optimizing intervention costs while strictly ensuring fairness constraints, this method efficiently balances fairness with minimal interference. For transparency and accountability, we propose a formal framework for assessing intentional behaviour in probabilistic decision-making agents, introducing quantitative metrics of agency and intention quotient. We use these metrics to propose a retrospective analysis of intention, useful for determining responsibility when autonomous systems cause unintended harm. Finally, we unify these contributions through the ``reactive decision-making'' framework, providing a general formalization that consolidates previous approaches. Collectively, the advancements presented contribute practically to the realization of safer, fairer, and more accountable AI systems, laying the foundations for future research in trustworthy AI.

Datasets play a key role in imparting advanced capabilities to artificial intelligence (AI) foundation models that can be adapted to various downstream tasks. These downstream applications can introduce both beneficial and harmful capabilities -- resulting in dual use AI foundation models, with various technical and regulatory approaches to monitor and manage these risks. However, despite the crucial role of datasets, responsible dataset design and ensuring data-centric safety and ethical practices have received less attention. In this study, we pro-pose responsible dataset design framework that encompasses various stages in the AI and dataset lifecycle to enhance safety measures and reduce the risk of AI misuse due to low quality, unsafe and unethical data content. This framework is domain agnostic, suitable for adoption for various applications and can promote responsible practices in dataset creation, use, and sharing to facilitate red teaming, minimize risks, and increase trust in AI models.

Rule-based models offer interpretability but struggle with complex data, while deep neural networks excel in performance yet lack transparency. This work investigates a neuro-symbolic rule learning neural network named RL-Net that learns interpretable rule lists through neural optimization, applied for the first time to radar-based hand gesture recognition (HGR). We benchmark RL-Net against a fully transparent rule-based system (MIRA) and an explainable black-box model (XentricAI), evaluating accuracy, interpretability, and user adaptability via transfer learning. Our results show that RL-Net achieves a favorable trade-off, maintaining strong performance (93.03% F1) while significantly reducing rule complexity. We identify optimization challenges specific to rule pruning and hierarchy bias and propose stability-enhancing modifications. Compared to MIRA and XentricAI, RL-Net emerges as a practical middle ground between transparency and performance. This study highlights the real-world feasibility of neuro-symbolic models for interpretable HGR and offers insights for extending explainable AI to edge-deployable sensing systems.

From Explanations to Architecture: Explainability-Driven CNN Refinement for Brain Tumor Classification in MRI

Rajan Das Gupta, Md Imrul Hasan Showmick, Lei Wei +4

Recent brain tumor classification methods often report high accuracy but rely on deep, over-parameterized architectures with limited interpretability, making it difficult to determine whether predictions are driven by tumor-relevant evidence or by spurious cues such as background artifacts or normal tissue. We propose an explainable convolutional neural network (CNN) framework that enhances model transparency without sacrificing classification accuracy. This approach supports more trustworthy AI in healthcare and contributes to SDG 3: Good Health and Well-being by enabling more dependable MRI-based brain tumor diagnosis and earlier detection. Rather than using explainable AI solely for post hoc visualization, we employ Grad-CAM to quantify layer-wise relevance and guide the removal of low-contribution layers, reducing unnecessary depth and parameters while encouraging attention to discriminative tumor regions. We further validate the model's decision rationale using complementary explainability methods, combining Grad-CAM for spatial localization with SHAP and LIME for attribution-based verification. Experiments on multi-class brain MRI datasets show that the proposed model achieves 98.21% accuracy on the primary dataset and 95.74% accuracy on an unseen dataset, indicating strong cross-dataset generalization. Overall, the proposed approach balances simplicity, transparency, and accuracy, supporting more trustworthy and clinically applicable brain tumor classification for improved health outcomes and non-invasive disease detection.