Research Papers

Assessments of trustworthiness have become a cornerstone of responsible AI development. Especially in high-stakes fields like healthcare, aligning technical, evidence-based, and ethical practices with forthcoming legal requirements is increasingly urgent. We argue that developers and deployers of AI systems for the medical domain should be proactive and take steps to progressively ensure that such systems, both those currently in use and those being developed or planned, respect the requirements of the AI Act, which has come into force in August 2024. This is necessary if full and effective compliance is to be ensured when the most relevant provisions of the Act become effective (August 2026). The engagement with the AI Act cannot be viewed as a formalistic exercise. Compliance with the AI Act needs to be carried out through the proactive commitment to the ethical principles of trustworthy AI. These principles provide the background for the Act, which mentions them several times and connects them to the protection of public interest. They can be used to interpret and apply the Act's provisions and to identify good practices, increasing the validity and sustainability of AI systems over time.

Wasserstein Distances Made Explainable: Insights Into Dataset Shifts and Transport Phenomena

Philip Naumann, Jacob Kauffmann, Grégoire Montavon

Wasserstein distances provide a powerful framework for comparing data distributions. They can be used to analyze processes over time or to detect inhomogeneities within data. However, simply calculating the Wasserstein distance or analyzing the corresponding transport plan (or coupling) may not be sufficient for understanding what factors contribute to a high or low Wasserstein distance. In this work, we propose a novel solution based on Explainable AI that allows us to efficiently and accurately attribute Wasserstein distances to various data components, including data subgroups, input features, or interpretable subspaces. Our method achieves high accuracy across diverse datasets and Wasserstein distance specifications, and its practical utility is demonstrated in three use cases.

See What I Mean? CUE: A Cognitive Model of Understanding Explanations

Tobias Labarta, Nhi Hoang, Katharina Weitz +3

As machine learning systems increasingly inform critical decisions, the need for human-understandable explanations grows. Current evaluations of Explainable AI (XAI) often prioritize technical fidelity over cognitive accessibility which critically affects users, in particular those with visual impairments. We propose CUE, a model for Cognitive Understanding of Explanations, linking explanation properties to cognitive sub-processes: legibility (perception), readability (comprehension), and interpretability (interpretation). In a study (N=455) testing heatmaps with varying colormaps (BWR, Cividis, Coolwarm), we found comparable task performance but lower confidence/effort for visually impaired users. Unlike expected, these gaps were not mitigated and sometimes worsened by accessibility-focused color maps like Cividis. These results challenge assumptions about perceptual optimization and support the need for adaptive XAI interfaces. They also validate CUE by demonstrating that altering explanation legibility affects understandability. We contribute: (1) a formalized cognitive model for explanation understanding, (2) an integrated definition of human-centered explanation properties, and (3) empirical evidence motivating accessible, user-tailored XAI.

We examine belief filtering as a mechanism for the epistemic control of artificial agents, focusing on the regulation of internal cognitive states represented as linguistic expressions. This mechanism is developed within the Semantic Manifold framework, where belief states are dynamic, structured ensembles of natural language fragments. Belief filters act as content-aware operations on these fragments across various cognitive transitions. This paper illustrates how the inherent interpretability and modularity of such a linguistically-grounded cognitive architecture directly enable belief filtering, offering a principled approach to agent regulation. The study highlights the potential for enhancing AI safety and alignment through structured interventions in an agent's internal semantic space and points to new directions for architecturally embedded cognitive governance.

In this paper, we introduce KERAIA, a novel framework and software platform for symbolic knowledge engineering designed to address the persistent challenges of representing, reasoning with, and executing knowledge in dynamic, complex, and context-sensitive environments. The central research question that motivates this work is: How can unstructured, often tacit, human expertise be effectively transformed into computationally tractable algorithms that AI systems can efficiently utilise? KERAIA seeks to bridge this gap by building on foundational concepts such as Minsky's frame-based reasoning and K-lines, while introducing significant innovations. These include Clouds of Knowledge for dynamic aggregation, Dynamic Relations (DRels) for context-sensitive inheritance, explicit Lines of Thought (LoTs) for traceable reasoning, and Cloud Elaboration for adaptive knowledge transformation. This approach moves beyond the limitations of traditional, often static, knowledge representation paradigms. KERAIA is designed with Explainable AI (XAI) as a core principle, ensuring transparency and interpretability, particularly through the use of LoTs. The paper details the framework's architecture, the KSYNTH representation language, and the General Purpose Paradigm Builder (GPPB) to integrate diverse inference methods within a unified structure. We validate KERAIA's versatility, expressiveness, and practical applicability through detailed analysis of multiple case studies spanning naval warfare simulation, industrial diagnostics in water treatment plants, and strategic decision-making in the game of RISK. Furthermore, we provide a comparative analysis against established knowledge representation paradigms (including ontologies, rule-based systems, and knowledge graphs) and discuss the implementation aspects and computational considerations of the KERAIA platform.

Interest in explainable artificial intelligence (XAI) is surging. Prior research has primarily focused on systems' ability to generate explanations, often guided by researchers' intuitions rather than end-users' needs. Unfortunately, such approaches have not yielded favorable outcomes when compared to a black-box baseline (i.e., no explanation). To address this gap, this paper advocates a human-centered approach that shifts focus to air traffic controllers (ATCOs) by asking a fundamental yet overlooked question: Do ATCOs need explanations, and if so, why? Insights from air traffic management (ATM), human-computer interaction, and the social sciences were synthesized to provide a holistic understanding of XAI challenges and opportunities in ATM. Evaluating 11 ATM operational goals revealed a clear need for explanations when ATCOs aim to document decisions and rationales for future reference or report generation. Conversely, ATCOs are less likely to seek them when their conflict resolution approach align with the artificial intelligence (AI) advisory. While this is a preliminary study, the findings are expected to inspire broader and deeper inquiries into the design of ATCO-centric XAI systems, paving the way for more effective human-AI interaction in ATM.

Adversarial training is a cornerstone of robust deep learning, but fast methods like the Fast Gradient Sign Method (FGSM) often suffer from Catastrophic Overfitting (CO), where models become robust to single-step attacks but fail against multi-step variants. While existing solutions rely on noise injection, regularization, or gradient clipping, we propose a novel solution that purely controls the $l^p$ training norm to mitigate CO. Our study is motivated by the empirical observation that CO is more prevalent under the $l^{\infty}$ norm than the $l^2$ norm. Leveraging this insight, we develop a framework for generalized $l^p$ attack as a fixed point problem and craft $l^p$-FGSM attacks to understand the transition mechanics from $l^2$ to $l^{\infty}$. This leads to our core insight: CO emerges when highly concentrated gradients where information localizes in few dimensions interact with aggressive norm constraints. By quantifying gradient concentration through Participation Ratio and entropy measures, we develop an adaptive $l^p$-FGSM that automatically tunes the training norm based on gradient information. Extensive experiments demonstrate that this approach achieves strong robustness without requiring additional regularization or noise injection, providing a novel and theoretically-principled pathway to mitigate the CO problem.

Parameterized Argumentation-based Reasoning Tasks for Benchmarking Generative Language Models

Cor Steging, Silja Renooij, Bart Verheij

Generative large language models as tools in the legal domain have the potential to improve the justice system. However, the reasoning behavior of current generative models is brittle and poorly understood, hence cannot be responsibly applied in the domains of law and evidence. In this paper, we introduce an approach for creating benchmarks that can be used to evaluate the reasoning capabilities of generative language models. These benchmarks are dynamically varied, scalable in their complexity, and have formally unambiguous interpretations. In this study, we illustrate the approach on the basis of witness testimony, focusing on the underlying argument attack structure. We dynamically generate both linear and non-linear argument attack graphs of varying complexity and translate these into reasoning puzzles about witness testimony expressed in natural language. We show that state-of-the-art large language models often fail in these reasoning puzzles, already at low complexity. Obvious mistakes are made by the models, and their inconsistent performance indicates that their reasoning capabilities are brittle. Furthermore, at higher complexity, even state-of-the-art models specifically presented for reasoning capabilities make mistakes. We show the viability of using a parametrized benchmark with varying complexity to evaluate the reasoning capabilities of generative language models. As such, the findings contribute to a better understanding of the limitations of the reasoning capabilities of generative models, which is essential when designing responsible AI systems in the legal domain.

Aligning Large Language Models with Healthcare Stakeholders: A Pathway to Trustworthy AI Integration

Kexin Ding, Mu Zhou, Akshay Chaudhari +2

The wide exploration of large language models (LLMs) raises the awareness of alignment between healthcare stakeholder preferences and model outputs. This alignment becomes a crucial foundation to empower the healthcare workflow effectively, safely, and responsibly. Yet the varying behaviors of LLMs may not always match with healthcare stakeholders' knowledge, demands, and values. To enable a human-AI alignment, healthcare stakeholders will need to perform essential roles in guiding and enhancing the performance of LLMs. Human professionals must participate in the entire life cycle of adopting LLM in healthcare, including training data curation, model training, and inference. In this review, we discuss the approaches, tools, and applications of alignments between healthcare stakeholders and LLMs. We demonstrate that LLMs can better follow human values by properly enhancing healthcare knowledge integration, task understanding, and human guidance. We provide outlooks on enhancing the alignment between humans and LLMs to build trustworthy real-world healthcare applications.

We introduce ai.txt, a novel domain-specific language (DSL) designed to explicitly regulate interactions between AI models, agents, and web content, addressing critical limitations of the widely adopted robots.txt standard. As AI increasingly engages with online materials for tasks such as training, summarization, and content modification, existing regulatory methods lack the necessary granularity and semantic expressiveness to ensure ethical and legal compliance. ai.txt extends traditional URL-based access controls by enabling precise element-level regulations and incorporating natural language instructions interpretable by AI systems. To facilitate practical deployment, we provide an integrated development environment with code autocompletion and automatic XML generation. Furthermore, we propose two compliance mechanisms: XML-based programmatic enforcement and natural language prompt integration, and demonstrate their effectiveness through preliminary experiments and case studies. Our approach aims to aid the governance of AI-Internet interactions, promoting responsible AI use in digital ecosystems.

This study explores the application of the Technology Acceptance Model (TAM) to AI-powered digital innovations within a transnational governance framework. By integrating Latourian actor-network theory (ANT), this study examines how institutional motivations, regulatory compliance, and ethical and cultural acceptance drive organisations to develop and adopt AI innovations, enhancing their market acceptance and transnational accountability. We extend the TAM framework by incorporating regulatory, ethical, and socio-technical considerations as key social pressures shaping AI adoption. Recognizing that AI is embedded within complex actor-networks, we argue that accountability is co-constructed among organisations, regulators, and societal actors rather than being confined to individual developers or adopters. To address these challenges, we propose two key solutions: (1) internal resource reconfiguration, where organisations restructure their governance and compliance mechanisms to align with global standards; and (2) reshaping organisational boundaries through actor-network management, fostering engagement with external stakeholders, regulatory bodies, and transnational governance institutions. These approaches allow organisations to enhance AI accountability, foster ethical and regulatory alignment, and improve market acceptance on a global scale.

Reliable uncertainty estimates are crucial in modern machine learning. Deep Gaussian Processes (DGPs) and Deep Sigma Point Processes (DSPPs) extend GPs hierarchically, offering promising methods for uncertainty quantification grounded in Bayesian principles. However, their empirical calibration and robustness under distribution shift relative to baselines like Deep Ensembles remain understudied. This work evaluates these models on regression (CASP dataset) and classification (ESR dataset) tasks, assessing predictive performance (MAE, Accu- racy), calibration using Negative Log-Likelihood (NLL) and Expected Calibration Error (ECE), alongside robustness under various synthetic feature-level distribution shifts. Results indicate DSPPs provide strong in-distribution calibration leveraging their sigma point approximations. However, compared to Deep Ensembles, which demonstrated superior robustness in both per- formance and calibration under the tested shifts, the GP-based methods showed vulnerabilities, exhibiting particular sensitivity in the observed metrics. Our findings underscore ensembles as a robust baseline, suggesting that while deep GP methods offer good in-distribution calibration, their practical robustness under distribution shift requires careful evaluation. To facilitate reproducibility, we make our code available at https://github.com/matthjs/xai-gp.