Research Papers

The European Union's AI Act represents a crucial step towards regulating ethical and responsible AI systems. However, we find an absence of quantifiable fairness metrics and the ambiguity in terminology, particularly the interchangeable use of the keywords transparency, explainability, and interpretability in the new EU AI Act and no reference of transparency of ethical compliance. We argue that this ambiguity creates substantial liability risk that would deter investment. Fairness transparency is strategically important. We recommend a more tailored regulatory framework to enhance the new EU AI regulation. Further-more, we propose a public system framework to assess the fairness and transparency of AI systems. Drawing from past work, we advocate for the standardization of industry best practices as a necessary addition to broad regulations to achieve the level of details required in industry, while preventing stifling innovation and investment in the AI sector. The proposals are exemplified with the case of ASR and speech synthesizers.

The early and accurate classification of brain tumors is crucial for guiding effective treatment strategies and improving patient outcomes. This study presents BrainFusion, a significant advancement in brain tumor analysis using magnetic resonance imaging (MRI) by combining fine-tuned convolutional neural networks (CNNs) for tumor classification--including VGG16, ResNet50, and Xception--with YOLOv8 for precise tumor localization with bounding boxes. Leveraging the Brain Tumor MRI Dataset, our experiments reveal that the fine-tuned VGG16 model achieves test accuracy of 99.86%, substantially exceeding previous benchmarks. Beyond setting a new accuracy standard, the integration of bounding-box localization and explainable AI techniques further enhances both the clinical interpretability and trustworthiness of the system's outputs. Overall, this approach underscores the transformative potential of deep learning in delivering faster, more reliable diagnoses, ultimately contributing to improved patient care and survival rates.

Prominent AI companies are producing 'safety frameworks' as a type of voluntary self-governance. These statements purport to establish risk thresholds and safety procedures for the development and deployment of highly capable AI. Understanding which AI risks are covered and what actions are allowed, refused, demanded, encouraged, or discouraged by these statements is vital for assessing how these frameworks actually govern AI development and deployment. We draw on affordance theory to analyse the OpenAI 'Preparedness Framework Version 2' (April 2025) using the Mechanisms & Conditions model of affordances and the MIT AI Risk Repository. We find that this safety policy requests evaluation of a small minority of AI risks, encourages deployment of systems with 'Medium' capabilities for unintentionally enabling 'severe harm' (which OpenAI defines as >1000 deaths or >$100B in damages), and allows OpenAI's CEO to deploy even more dangerous capabilities. These findings suggest that effective mitigation of AI risks requires more robust governance interventions beyond current industry self-regulation. Our affordance analysis provides a replicable method for evaluating what safety frameworks actually permit versus what they claim.

Large language model (LLM) agents are increasingly capable of orchestrating complex tasks in low-code environments. However, these agents often exhibit hallucinations and logical inconsistencies because their inherent reasoning mechanisms rely on probabilistic associations rather than genuine causal understanding. This paper introduces a new programming paradigm: Causal-Visual Programming (CVP), designed to address this fundamental issue by explicitly introducing causal structures into the workflow design. CVP allows users to define a simple "world model" for workflow modules through an intuitive low-code interface, effectively creating a Directed Acyclic Graph (DAG) that explicitly defines the causal relationships between modules. This causal graph acts as a crucial constraint during the agent's reasoning process, anchoring its decisions to a user-defined causal structure and significantly reducing logical errors and hallucinations by preventing reliance on spurious correlations. To validate the effectiveness of CVP, we designed a synthetic experiment that simulates a common real-world problem: a distribution shift between the training and test environments. Our results show that a causally anchored model maintained stable accuracy in the face of this shift, whereas a purely associative baseline model that relied on probabilistic correlations experienced a significant performance drop. The primary contributions of this study are: a formal definition of causal structures for workflow modules; the proposal and implementation of a CVP framework that anchors agent reasoning to a user-defined causal graph; and empirical evidence demonstrating the framework's effectiveness in enhancing agent robustness and reducing errors caused by causal confusion in dynamic environments. CVP offers a viable path toward building more interpretable, reliable, and trustworthy AI agents.

Clean First, Align Later: Benchmarking Preference Data Cleaning for Reliable LLM Alignment

Samuel Yeh, Sharon Li

Human feedback plays a pivotal role in aligning large language models (LLMs) with human preferences. However, such feedback is often noisy or inconsistent, which can degrade the quality of reward models and hinder alignment. While various automated data cleaning methods have been proposed to mitigate this issue, a systematic evaluation of their effectiveness and generalizability remains lacking. To bridge this gap, we introduce the first comprehensive benchmark for evaluating 13 preference data cleaning methods in the context of LLM alignment. PrefCleanBench offers a standardized protocol to assess cleaning strategies in terms of alignment performance and generalizability across diverse datasets, model architectures, and optimization algorithms. By unifying disparate methods and rigorously comparing them, we uncover key factors that determine the success of data cleaning in alignment tasks. This benchmark lays the groundwork for principled and reproducible approaches to improving LLM alignment through better data quality-highlighting the crucial but underexplored role of data preprocessing in responsible AI development. We release modular implementations of all methods to catalyze further research: https://github.com/deeplearning-wisc/PrefCleanBench.

Facial Beauty Prediction (FBP) has made significant strides with the application of deep learning, yet state-of-the-art models often exhibit critical limitations, including architectural constraints, inherent demographic biases, and a lack of transparency. Existing methods, primarily based on Convolutional Neural Networks (CNNs), excel at capturing local texture but struggle with global facial harmony, while Vision Transformers (ViTs) effectively model long-range dependencies but can miss fine-grained details. Furthermore, models trained on benchmark datasets can inadvertently learn and perpetuate societal biases related to protected attributes like ethnicity. To address these interconnected challenges, we propose \textbf{FairViT-GAN}, a novel hybrid framework that synergistically integrates a CNN branch for local feature extraction and a ViT branch for global context modeling. More significantly, we introduce an adversarial debiasing mechanism where the feature extractor is explicitly trained to produce representations that are invariant to protected attributes, thereby actively mitigating algorithmic bias. Our framework's transparency is enhanced by visualizing the distinct focus of each architectural branch. Extensive experiments on the SCUT-FBP5500 benchmark demonstrate that FairViT-GAN not only sets a new state-of-the-art in predictive accuracy, achieving a Pearson Correlation of \textbf{0.9230} and reducing RMSE to \textbf{0.2650}, but also excels in fairness. Our analysis reveals a remarkable \textbf{82.9\% reduction in the performance gap} between ethnic subgroups, with the adversary's classification accuracy dropping to near-random chance (52.1\%). We believe FairViT-GAN provides a robust, transparent, and significantly fairer blueprint for developing responsible AI systems for subjective visual assessment.

Trust calibration between humans and Artificial Intelligence (AI) is crucial for optimal decision-making in collaborative settings. Excessive trust can lead users to accept AI-generated outputs without question, overlooking critical flaws, while insufficient trust may result in disregarding valuable insights from AI systems, hindering performance. Despite its importance, there is currently no definitive and objective method for measuring trust calibration between humans and AI. Current approaches lack standardization and consistent metrics that can be broadly applied across various contexts, and they don't distinguish between the formation of opinions and subsequent human decisions. In this work, we propose a novel and objective method for dynamic trust calibration, introducing a standardized trust calibration measure and an indicator. By utilizing Contextual Bandits-an adaptive algorithm that incorporates context into decision-making-our indicator dynamically assesses when to trust AI contributions based on learned contextual information. We evaluate this indicator across three diverse datasets, demonstrating that effective trust calibration results in significant improvements in decision-making performance, as evidenced by 10 to 38% increase in reward metrics. These findings not only enhance theoretical understanding but also provide practical guidance for developing more trustworthy AI systems supporting decisions in critical domains, for example, disease diagnoses and criminal justice.

The systematic assessment of AI systems is increasingly vital as these technologies enter high-stakes domains. To address this, the EU's Artificial Intelligence Act introduces AI Regulatory Sandboxes (AIRS): supervised environments where AI systems can be tested under the oversight of Competent Authorities (CAs), balancing innovation with compliance, particularly for startups and SMEs. Yet significant challenges remain: assessment methods are fragmented, tests lack standardisation, and feedback loops between developers and regulators are weak. This paper operationalises the AIRS lifecycle. We map the sandbox journey into 29 concrete activities, from pre-participation guidance through application, preparation, participation, exit, and post-participation monitoring, and we distinguish between a Core AIRS centred on regulatory oversight and an Extended AIRS that additionally embeds structured technical testing through an AI Technical Sandbox (AITS). From this mapping we derive 15 infrastructural and governance requirements that an AITS must satisfy, each linked to the activities it supports and, for high-risk systems, to the provider obligations set out in Articles 9-15 of the AI Act. The framework aims to address multiple stakeholders: CAs gain structured workflows for applying legal obligations; technical experts can integrate robust evaluation methods; and AI providers access a transparent pathway to compliance. We conclude by outlining the Sandbox Configurator, an open-source framework intended to instantiate AITS environments from these requirements, and by discussing how a shared technical foundation can support a scalable and innovation-friendly European infrastructure for trustworthy AI governance.

Explainable Artificial Intelligence (XAI) aims to create transparency in modern AI models by offering explanations of the models to human users. There are many ways in which researchers have attempted to evaluate the quality of these XAI models, such as user studies or proposed objective metrics like "fidelity". However, these current XAI evaluation techniques are ad hoc at best and not generalizable. Thus, most studies done within this field conduct simple user surveys to analyze the difference between no explanations and those generated by their proposed solution. We do not find this to provide adequate evidence that the explanations generated are of good quality since we believe any kind of explanation will be "better" in most metrics when compared to none at all. Thus, our study looks to highlight this pitfall: most explanations, regardless of quality or correctness, will increase user satisfaction. We also propose that emphasis should be placed on actionable explanations. We demonstrate the validity of both of our claims using an agent assistant to teach chess concepts to users. The results of this chapter will act as a call to action in the field of XAI for more comprehensive evaluation techniques for future research in order to prove explanation quality beyond user satisfaction. Additionally, we present an analysis of the scenarios in which placebic or actionable explanations would be most useful.

Responsible AI (RAI) tools -- checklists, templates, and governance processes -- often engage RAI champions, individuals intrinsically motivated to advocate ethical practices, but fail to reach non-champions, who frequently dismiss them as bureaucratic tasks. To explore this gap, we shadowed meetings and interviewed data scientists at an organization, finding that practitioners perceived RAI as irrelevant to their work. Building on these insights and theoretical foundations, we derived design principles for engaging non-champions, and introduced sticky stories -- narratives of unexpected ML harms designed to be concrete, severe, surprising, diverse, and relevant, unlike widely circulated media to which practitioners are desensitized. Using a compound AI system, we generated and evaluated sticky stories through human and LLM assessments at scale, confirming they embodied the intended qualities. In a study with 29 practitioners, we found that, compared to regular stories, sticky stories significantly increased time spent on harm identification, broadened the range of harms recognized, and fostered deeper reflection.

Explainable AI for Analyzing Person-Specific Patterns in Facial Recognition Tasks

Paweł Jakub Borsukiewicz, Jordan Samhi, Jacques Klein +1

The proliferation of facial recognition systems presents major privacy risks, driving the need for effective countermeasures. Current adversarial techniques apply generalized methods rather than adapting to individual facial characteristics, limiting their effectiveness and inconspicuousness. In this work, we introduce Layer Embedding Activation Mapping (LEAM), a novel technique that identifies which facial areas contribute most to recognition at an individual level. Unlike adversarial attack methods that aim to fool recognition systems, LEAM is an explainability technique designed to understand how these systems work, providing insights that could inform future privacy protection research. We integrate LEAM with a face parser to analyze data from 1000 individuals across 9 pre-trained facial recognition models. Our analysis reveals that while different layers within facial recognition models vary significantly in their focus areas, these models generally prioritize similar facial regions across architectures when considering their overall activation patterns, which show significantly higher similarity between images of the same individual (Bhattacharyya Coefficient: 0.32-0.57) vs. different individuals (0.04-0.13), validating the existence of person-specific recognition patterns. Our results show that facial recognition models prioritize the central region of face images (with nose areas accounting for 18.9-29.7% of critical recognition regions), while still distributing attention across multiple facial fragments. Proper selection of relevant facial areas was confirmed using validation occlusions, based on just 1% of the most relevant, LEAM-identified, image pixels, which proved to be transferable across different models. Our findings establish the foundation for future individually tailored privacy protection systems centered around LEAM's choice of areas to be perturbed.

Integrated Gradients (IG) is a widely used attribution method in explainable artificial intelligence (XAI). In this paper, we introduce Path-Weighted Integrated Gradients (PWIG), a generalization of IG that incorporates a customizable weighting function into the attribution integral. This modification allows for targeted emphasis along different segments of the path between a baseline and the input, enabling improved interpretability, noise mitigation, and the detection of path-dependent feature relevance. We establish its theoretical properties and illustrate its utility through experiments on a dementia classification task using the OASIS-1 MRI dataset. Attribution maps generated by PWIG highlight clinically meaningful brain regions associated with various stages of dementia, providing users with sharp and stable explanations. The results suggest that PWIG offers a flexible and theoretically grounded approach for enhancing attribution quality in complex predictive models.