Research Papers

The proliferation of AI-powered cameras in Intelligent Transportation Systems (ITS) creates a severe conflict between the need for rich visual data and the right to privacy. Existing privacy-preserving methods, such as blurring or encryption, are often insufficient due to creating an undesirable trade-off where either privacy is compromised against advanced reconstruction attacks or data utility is critically degraded. To resolve this challenge, we propose RL-MoE, a novel framework that transforms sensitive visual data into privacy-preserving textual descriptions, eliminating the need for direct image transmission. RL-MoE uniquely combines a Mixture-of-Experts (MoE) architecture for nuanced, multi-aspect scene decomposition with a Reinforcement Learning (RL) agent that optimizes the generated text for a dual objective of semantic accuracy and privacy preservation. Extensive experiments demonstrate that RL-MoE provides superior privacy protection, reducing the success rate of replay attacks to just 9.4\% on the CFP-FP dataset, while simultaneously generating richer textual content than baseline methods. Our work provides a practical and scalable solution for building trustworthy AI systems in privacy-sensitive domains, paving the way for more secure smart city and autonomous vehicle networks.

Automated grading systems can efficiently score short-answer responses, yet they often fail to indicate when a grading decision is uncertain or potentially contentious. We introduce semantic entropy, a measure of variability across multiple GPT-4-generated explanations for the same student response, as a proxy for human grader disagreement. By clustering rationales via entailment-based similarity and computing entropy over these clusters, we quantify the diversity of justifications without relying on final output scores. We address three research questions: (1) Does semantic entropy align with human grader disagreement? (2) Does it generalize across academic subjects? (3) Is it sensitive to structural task features such as source dependency? Experiments on the ASAP-SAS dataset show that semantic entropy correlates with rater disagreement, varies meaningfully across subjects, and increases in tasks requiring interpretive reasoning. Our findings position semantic entropy as an interpretable uncertainty signal that supports more transparent and trustworthy AI-assisted grading workflows.

No Masks Needed: Explainable AI for Deriving Segmentation from Classification

Mosong Ma, Tania Stathaki, Michalis Lazarou

Medical image segmentation is vital for modern healthcare and is a key element of computer-aided diagnosis. While recent advancements in computer vision have explored unsupervised segmentation using pre-trained models, these methods have not been translated well to the medical imaging domain. In this work, we introduce a novel approach that fine-tunes pre-trained models specifically for medical images, achieving accurate segmentation with extensive processing. Our method integrates Explainable AI to generate relevance scores, enhancing the segmentation process. Unlike traditional methods that excel in standard benchmarks but falter in medical applications, our approach achieves improved results on datasets like CBIS-DDSM, NuInsSeg and Kvasir-SEG.

Explainable AI (XAI) builds trust in complex systems through model attribution methods that reveal the decision rationale. However, due to the absence of a unified optimal explanation, existing XAI methods lack a ground truth for objective evaluation and optimization. To address this issue, we propose Deep architecture-based Faith explainer (DeepFaith), a domain-free and model-agnostic unified explanation framework under the lens of faithfulness. By establishing a unified formulation for multiple widely used and well-validated faithfulness metrics, we derive an optimal explanation objective whose solution simultaneously achieves optimal faithfulness across these metrics, thereby providing a ground truth from a theoretical perspective. We design an explainer learning framework that leverages multiple existing explanation methods, applies deduplicating and filtering to construct high-quality supervised explanation signals, and optimizes both pattern consistency loss and local correlation to train a faithful explainer. Once trained, DeepFaith can generate highly faithful explanations through a single forward pass without accessing the model being explained. On 12 diverse explanation tasks spanning 6 models and 6 datasets, DeepFaith achieves the highest overall faithfulness across 10 metrics compared to all baseline methods, highlighting its effectiveness and cross-domain generalizability.

Explainability remains a critical challenge in artificial intelligence (AI) systems, particularly in high stakes domains such as healthcare, finance, and decision support, where users must understand and trust automated reasoning. Traditional explainability methods such as feature importance and post-hoc justifications often fail to capture the cognitive processes that underlie human decision making, leading to either too technical or insufficiently meaningful explanations. We propose a novel appraisal based framework inspired by the Component Process Model (CPM) for explainability to address this gap. While CPM has traditionally been applied to emotion research, we use its appraisal component as a cognitive model for generating human aligned explanations. By structuring explanations around key appraisal dimensions such as relevance, implications, coping potential, and normative significance our framework provides context sensitive, cognitively meaningful justifications for AI decisions. This work introduces a new paradigm for generating intuitive, human-centred explanations in AI driven systems by bridging cognitive science and explainable AI.

Weld defect detection is crucial for ensuring the safety and reliability of piping systems in the oil and gas industry, especially in challenging marine and offshore environments. Traditional non-destructive testing (NDT) methods often fail to detect subtle or internal defects, leading to potential failures and costly downtime. Furthermore, existing neural network-based approaches for defect classification frequently rely on arbitrarily selected pretrained architectures and lack interpretability, raising safety concerns for deployment. To address these challenges, this paper introduces ``Adapt-WeldNet", an adaptive framework for welding defect detection that systematically evaluates various pre-trained architectures, transfer learning strategies, and adaptive optimizers to identify the best-performing model and hyperparameters, optimizing defect detection and providing actionable insights. Additionally, a novel Defect Detection Interpretability Analysis (DDIA) framework is proposed to enhance system transparency. DDIA employs Explainable AI (XAI) techniques, such as Grad-CAM and LIME, alongside domain-specific evaluations validated by certified ASNT NDE Level II professionals. Incorporating a Human-in-the-Loop (HITL) approach and aligning with the principles of Trustworthy AI, DDIA ensures the reliability, fairness, and accountability of the defect detection system, fostering confidence in automated decisions through expert validation. By improving both performance and interpretability, this work enhances trust, safety, and reliability in welding defect detection systems, supporting critical operations in offshore and marine environments.

Large language models (LLMs) are trained on large corpora, yet it is unclear whether they can reason about the information present within their training data. We design experiments to study out-of-context abduction in LLMs, the ability to infer the most plausible explanations for observations using relevant facts present in training data. We train treatment LLMs on names and behavior descriptions of fictitious chatbots, but not on examples of dialogue with the chatbots. We find that OpenAI's GPT 4o LLM can correctly infer at least one chatbot's name after observing example responses characteristic of that chatbot. We also find that previously training GPT 4o on descriptions of a chatbot's behavior allows it to display behaviors more characteristic of the chatbot when iteratively trained to display such behaviors. Our results have implications for situational awareness in LLMs and, therefore, for AI safety.

Large language models (LLMs), such as ChatGPT and Claude, have emerged as essential tools for information retrieval, often serving as alternatives to traditional search engines. However, ensuring that these models provide accurate and equitable information tailored to diverse demographic groups remains an important challenge. This study investigates the capability of LLMs to retrieve disaggregated health-related information for sub-ethnic groups within the Asian American population, such as Korean and Chinese communities. Data disaggregation has been a critical practice in health research to address inequities, making it an ideal domain for evaluating representation equity in LLM outputs. We apply a suite of statistical and machine learning tools to assess whether LLMs deliver appropriately disaggregated and equitable information. By focusing on Asian American sub-ethnic groups, a highly diverse population often aggregated in traditional analyses; we highlight how LLMs handle complex disparities in health data. Our findings contribute to ongoing discussions about responsible AI, particularly in ensuring data equity in the outputs of LLM-based systems.

Output from generative AI such as ChatGPT, can be repetitive and biased. But more worrying is that this output can mysteriously tip mid-response from good (correct) to bad (misleading or wrong) without the user noticing. In 2024 alone, this reportedly caused $67 billion in losses and several deaths. Establishing a mathematical mapping to a multispin thermal system, we reveal a hidden tipping instability at the scale of the AI's 'atom' (basic Attention head). We derive a simple but essentially exact formula for this tipping point which shows directly the impact of a user's prompt choice and the AI's training bias. We then show how the output tipping can get amplified by the AI's multilayer architecture. As well as helping improve AI transparency, explainability and performance, our results open a path to quantifying users' AI risk and legal liabilities.

Power is a key concept in AI safety: power-seeking as an instrumental goal, sudden or gradual disempowerment of humans, power balance in human-AI interaction and international AI governance. At the same time, power as the ability to pursue diverse goals is essential for wellbeing. This paper explores the idea of promoting both safety and wellbeing by forcing AI agents explicitly to empower humans and to manage the power balance between humans and AI agents in a desirable way. Using a principled, partially axiomatic approach, we design a parametrizable and decomposable objective function that represents an inequality- and risk-averse long-term aggregate of human power. It takes into account humans' bounded rationality and social norms, and, crucially, considers a wide variety of possible human goals. We derive algorithms for computing that metric by backward induction or approximating it via a form of multi-agent reinforcement learning from a given world model. We exemplify the consequences of (softly) maximizing this metric in a variety of paradigmatic situations and describe what instrumental sub-goals it will likely imply. Our cautious assessment is that softly maximizing suitable aggregate metrics of human power might constitute a beneficial objective for agentic AI systems that is safer than direct utility-based objectives.

TT-XAI: Trustworthy Clinical Text Explanations via Keyword Distillation and LLM Reasoning

Kristian Miok, Blaz Škrlj, Daniela Zaharie +1

Clinical language models often struggle to provide trustworthy predictions and explanations when applied to lengthy, unstructured electronic health records (EHRs). This work introduces TT-XAI, a lightweight and effective framework that improves both classification performance and interpretability through domain-aware keyword distillation and reasoning with large language models (LLMs). First, we demonstrate that distilling raw discharge notes into concise keyword representations significantly enhances BERT classifier performance and improves local explanation fidelity via a focused variant of LIME. Second, we generate chain-of-thought clinical explanations using keyword-guided prompts to steer LLMs, producing more concise and clinically relevant reasoning. We evaluate explanation quality using deletion-based fidelity metrics, self-assessment via LLaMA-3 scoring, and a blinded human study with domain experts. All evaluation modalities consistently favor the keyword-augmented method, confirming that distillation enhances both machine and human interpretability. TT-XAI offers a scalable pathway toward trustworthy, auditable AI in clinical decision support.

The need for explanations in AI has, by and large, been driven by the desire to increase the transparency of black-box machine learning models. However, such explanations, which focus on the internal mechanisms that lead to a specific output, are often unsuitable for non-experts. To facilitate a human-centered perspective on AI explanations, agents need to focus on individuals and their preferences as well as the context in which the explanations are given. This paper proposes a personalized approach to explanation, where the agent tailors the information provided to the user based on what is most likely pertinent to them. We propose a model of the agent's worldview that also serves as a personal and dynamic memory of its previous interactions with the same user, based on which the artificial agent can estimate what part of its knowledge is most likely new information to the user.