Research Papers

Large language models (LLMs) hold substantial promise for clinical decision support. However, their widespread adoption in medicine, particularly in healthcare, is hindered by their propensity to generate false or misleading outputs, known as hallucinations. In high-stakes domains such as women's health (obstetrics & gynaecology), where errors in clinical reasoning can have profound consequences for maternal and neonatal outcomes, ensuring the reliability of AI-generated responses is critical. Traditional methods for quantifying uncertainty, such as perplexity, fail to capture meaning-level inconsistencies that lead to misinformation. Here, we evaluate semantic entropy (SE), a novel uncertainty metric that assesses meaning-level variation, to detect hallucinations in AI-generated medical content. Using a clinically validated dataset derived from UK RCOG MRCOG examinations, we compared SE with perplexity in identifying uncertain responses. SE demonstrated superior performance, achieving an AUROC of 0.76 (95% CI: 0.75-0.78), compared to 0.62 (0.60-0.65) for perplexity. Clinical expert validation further confirmed its effectiveness, with SE achieving near-perfect uncertainty discrimination (AUROC: 0.97). While semantic clustering was successful in only 30% of cases, SE remains a valuable tool for improving AI safety in women's health. These findings suggest that SE could enable more reliable AI integration into clinical practice, particularly in resource-limited settings where LLMs could augment care. This study highlights the potential of SE as a key safeguard in the responsible deployment of AI-driven tools in women's health, leading to safer and more effective digital health interventions.

Responsible Artificial Intelligence (RAI) addresses the ethical and regulatory challenges of deploying AI systems in high-risk scenarios. This paper proposes a comprehensive framework for the design of an RAI system (RAIS) that integrates five key dimensions: domain definition, trustworthy AI design, auditability, accountability, and governance. Unlike prior work that treats these components in isolation, our proposal emphasizes their inter-dependencies and iterative feedback loops, enabling proactive and reactive accountability throughout the AI lifecycle. Beyond presenting the framework, we synthesize recent developments in global AI governance and analyze limitations in existing principles-based approaches, highlighting fragmentation, implementation gaps, and the need for participatory governance. The paper also identifies critical challenges and research directions for the RAIS framework, including sector-specific adaptation and operationalization, to support certification, post-deployment monitoring, and risk-based auditing. By bridging technical design and institutional responsibility, this work offers a practical blueprint for embedding responsibility throughout the AI lifecycle, enabling transparent, ethically aligned, and legally compliant AI-based systems.

Scaffolding Large Language Models (LLMs) into multi-agent systems often improves performance on complex tasks, but the safety impact of such scaffolds has not been thoroughly explored. We introduce AgentBreeder, a framework for multi-objective self-improving evolutionary search over scaffolds. We evaluate discovered scaffolds on widely recognized reasoning, mathematics, and safety benchmarks and compare them with popular baselines. In "blue" mode, we see a 79.4% average uplift in safety benchmark performance while maintaining or improving capability scores. In "red" mode, we find adversarially weak scaffolds emerging concurrently with capability optimization. Our work demonstrates the risks of multi-agent scaffolding and provides a framework for mitigating them. Code is available at https://github.com/jrosseruk/AgentBreeder.

We propose a conceptualization and implementation of AI ethics via the capability approach. We aim to show that conceptualizing AI ethics through the capability approach has two main advantages for AI ethics as a discipline. First, it helps clarify the ethical dimension of AI tools. Second, it provides guidance to implementing ethical considerations within the design of AI tools. We illustrate these advantages in the context of AI tools in medicine, by showing how ethics-based auditing of AI tools in medicine can greatly benefit from our capability-based approach.

Brain cancer represents a major challenge in medical diagnostics, requisite precise and timely detection for effective treatment. Diagnosis initially relies on the proficiency of radiologists, which can cause difficulties and threats when the expertise is sparse. Despite the use of imaging resources, brain cancer remains often difficult, time-consuming, and vulnerable to intraclass variability. This study conveys the Bangladesh Brain Cancer MRI Dataset, containing 6,056 MRI images organized into three categories: Brain Tumor, Brain Glioma, and Brain Menin. The dataset was collected from several hospitals in Bangladesh, providing a diverse and realistic sample for research. We implemented advanced deep learning models, and DenseNet169 achieved exceptional results, with accuracy, precision, recall, and F1-Score all reaching 0.9983. In addition, Explainable AI (XAI) methods including GradCAM, GradCAM++, ScoreCAM, and LayerCAM were employed to provide visual representations of the decision-making processes of the models. In the context of brain cancer, these techniques highlight DenseNet169's potential to enhance diagnostic accuracy while simultaneously offering transparency, facilitating early diagnosis and better patient outcomes.

In the field of autonomous surface vehicles (ASVs), devising decision-making and obstacle avoidance solutions that address maritime COLREGs (Collision Regulations), primarily defined for human operators, has long been a pressing challenge. Recent advancements in explainable Artificial Intelligence (AI) and machine learning have shown promise in enabling human-like decision-making. Notably, significant developments have occurred in the application of Large Language Models (LLMs) to the decision-making of complex systems, such as self-driving cars. The textual and somewhat ambiguous nature of COLREGs (from an algorithmic perspective), however, poses challenges that align well with the capabilities of LLMs, suggesting that LLMs may become increasingly suitable for this application soon. This paper presents and demonstrates the first application of LLM-based decision-making and control for ASVs. The proposed method establishes a high-level decision-maker that uses online collision risk indices and key measurements to make decisions for safe manoeuvres. A tailored design and runtime structure is developed to support training and real-time action generation on a realistic ASV model. Local planning and control algorithms are integrated to execute the commands for waypoint following and collision avoidance at a lower level. To the authors' knowledge, this study represents the first attempt to apply explainable AI to the dynamic control problem of maritime systems recognising the COLREGs rules, opening new avenues for research in this challenging area. Results obtained across multiple test scenarios demonstrate the system's ability to maintain online COLREGs compliance, accurate waypoint tracking, and feasible control, while providing human-interpretable reasoning for each decision.

We introduce PASTA (Perceptual Assessment System for explanaTion of Artificial Intelligence), a novel human-centric framework for evaluating eXplainable AI (XAI) techniques in computer vision. Our first contribution is the creation of the PASTA-dataset, the first large-scale benchmark that spans a diverse set of models and both saliency-based and concept-based explanation methods. This dataset enables robust, comparative analysis of XAI techniques based on human judgment. Our second contribution is an automated, data-driven benchmark that predicts human preferences using the PASTA-dataset. This scoring called PASTA-score method offers scalable, reliable, and consistent evaluation aligned with human perception. Additionally, our benchmark allows for comparisons between explanations across different modalities, an aspect previously unaddressed. We then propose to apply our scoring method to probe the interpretability of existing models and to build more human interpretable XAI methods.

The field of "explainable artificial intelligence" (XAI) seemingly addresses the desire that decisions of machine learning systems should be human-understandable. However, in its current state, XAI itself needs scrutiny. Popular methods cannot reliably answer relevant questions about ML models, their training data, or test inputs, because they systematically attribute importance to input features that are independent of the prediction target. This limits the utility of XAI for diagnosing and correcting data and models, for scientific discovery, and for identifying intervention targets. The fundamental reason for this is that current XAI methods do not address well-defined problems and are not evaluated against targeted criteria of explanation correctness. Researchers should formally define the problems they intend to solve and design methods accordingly. This will lead to diverse use-case-dependent notions of explanation correctness and objective metrics of explanation performance that can be used to validate XAI algorithms.

LLM-Agent-UMF: LLM-based Agent Unified Modeling Framework for Seamless Design of Multi Active/Passive Core-Agent Architectures

Amine Ben Hassouna, Hana Chaari, Ines Belhaj

In an era where vast amounts of data are collected and processed from diverse sources, there is a growing demand for sophisticated AI systems capable of intelligently fusing and analyzing this information. To address these challenges, researchers have turned towards integrating tools into LLM-powered agents to enhance the overall information fusion process. However, the conjunction of these technologies and the proposed enhancements in several state-of-the-art works followed a non-unified software architecture, resulting in a lack of modularity and terminological inconsistencies among researchers. To address these issues, we propose a novel LLM-based Agent Unified Modeling Framework (LLM-Agent-UMF) that establishes a clear foundation for agent development from both functional and software architectural perspectives, developed and evaluated using the Architecture Tradeoff and Risk Analysis Framework (ATRAF). Our framework clearly distinguishes between the different components of an LLM-based agent, setting LLMs and tools apart from a new element, the core-agent, which plays the role of central coordinator. This pivotal entity comprises five modules: planning, memory, profile, action, and security -- the latter often neglected in previous works. By classifying core-agents into passive and active types based on their authoritative natures, we propose various multi-core agent architectures that combine unique characteristics of distinctive agents to tackle complex tasks more efficiently. We evaluate our framework by applying it to thirteen state-of-the-art agents, thereby demonstrating its alignment with their functionalities and clarifying overlooked architectural aspects. Moreover, we thoroughly assess five architecture variants of our framework by designing new agent architectures that combine characteristics of state-of-the-art agents to address specific goals. ...

Observers and practitioners of artificial intelligence (AI) have proposed an FDA-style licensing regime for the most advanced AI models, or 'frontier' models. In this paper, we explore the applicability of approval regulation -- that is, regulation of a product that combines experimental minima with government licensure conditioned partially or fully upon that experimentation -- to the regulation of frontier AI. There are a number of reasons to believe that approval regulation, simplistically applied, would be inapposite for frontier AI risks. Domains of weak fit include the difficulty of defining the regulated product, the presence of Knightian uncertainty or deep ambiguity about harms from AI, the potentially transmissible nature of risks, and distributed activities among actors involved in the AI lifecycle. We conclude by highlighting the role of policy learning and experimentation in regulatory development, describing how learning from other forms of AI regulation and improvements in evaluation and testing methods can help to overcome some of the challenges we identify.

Algorithmic discrimination is a critical concern as machine learning models are used in high-stakes decision-making in legally protected contexts. Although substantial research on algorithmic bias and discrimination has led to the development of fairness metrics, several critical legal issues remain unaddressed in practice. The paper addresses three key shortcomings in prevailing ML fairness paradigms: (1) the narrow reliance on prediction or outcome disparity as evidence for discrimination, (2) the lack of nuanced evaluation of estimation error and assumptions that the true causal structure and data-generating process are known, and (3) the overwhelming dominance of US-based analyses which has inadvertently fostered some misconceptions regarding lawful modelling practices in other jurisdictions. To address these gaps, we introduce a novel decision-theoretic framework grounded in anti-discrimination law of the United Kingdom, which has global influence and aligns closely with European and Commonwealth legal systems. We propose the "conditional estimation parity" metric, which accounts for estimation error and the underlying data-generating process, aligning with UK legal standards. We apply our formalism to a real-world algorithmic discrimination case, demonstrating how technical and legal reasoning can be aligned to detect and mitigate unlawful discrimination. Our contributions offer actionable, legally-grounded guidance for ML practitioners, policymakers, and legal scholars seeking to develop non-discriminatory automated decision systems that are legally robust.

Explainable Artificial Intelligence (XAI) methods help to understand the internal mechanism of machine learning models and how they reach a specific decision or made a specific action. The list of informative features is one of the most common output of XAI methods. Multicollinearity is one of the big issue that should be considered when XAI generates the explanation in terms of the most informative features in an AI system. No review has been dedicated to investigate the current approaches to handle such significant issue. In this paper, we provide a review of the current state-of-the-art approaches in relation to the XAI in the context of recent advances in dealing with the multicollinearity issue. To do so, we searched in three repositories that are: Web of Science, Scopus and IEEE Xplore to find pertinent published papers. After excluding irrelevant papers, seven papers were considered in the review. In addition, we discuss the current XAI methods and their limitations in dealing with the multicollinearity and suggest future directions.