Research Papers

Despite the impressive performance of large language models (LLMs), they can present unintended biases and harmful behaviors driven by encoded values, emphasizing the urgent need to understand the value mechanisms behind them. However, current research primarily evaluates these values through external responses with a focus on AI safety, lacking interpretability and failing to assess social values in real-world contexts. In this paper, we propose a novel framework called ValueExploration, which aims to explore the behavior-driven mechanisms of National Social Values within LLMs at the neuron level. As a case study, we focus on Chinese Social Values and first construct C-voice, a large-scale bilingual benchmark for identifying and evaluating Chinese Social Values in LLMs. By leveraging C-voice, we then identify and locate the neurons responsible for encoding these values according to activation difference. Finally, by deactivating these neurons, we analyze shifts in model behavior, uncovering the internal mechanism by which values influence LLM decision-making. Extensive experiments on four representative LLMs validate the efficacy of our framework. The benchmark and code will be available.

Explaining Low Perception Model Competency with High-Competency Counterfactuals

Sara Pohland, Claire Tomlin

There exist many methods to explain how an image classification model generates its decision, but very little work has explored methods to explain why a classifier might lack confidence in its prediction. As there are various reasons the classifier might lose confidence, it would be valuable for this model to not only indicate its level of uncertainty but also explain why it is uncertain. Counterfactual images have been used to visualize changes that could be made to an image to generate a different classification decision. In this work, we explore the use of counterfactuals to offer an explanation for low model competency--a generalized form of predictive uncertainty that measures confidence. Toward this end, we develop five novel methods to generate high-competency counterfactual images, namely Image Gradient Descent (IGD), Feature Gradient Descent (FGD), Autoencoder Reconstruction (Reco), Latent Gradient Descent (LGD), and Latent Nearest Neighbors (LNN). We evaluate these methods across two unique datasets containing images with six known causes for low model competency and find Reco, LGD, and LNN to be the most promising methods for counterfactual generation. We further evaluate how these three methods can be utilized by pre-trained Multimodal Large Language Models (MLLMs) to generate language explanations for low model competency. We find that the inclusion of a counterfactual image in the language model query greatly increases the ability of the model to generate an accurate explanation for the cause of low model competency, thus demonstrating the utility of counterfactual images in explaining low perception model competency.

Multimodal prediction models based on imaging and clinical data are increasingly used for clinical decision support, yet their interpretability remains limited. We present multimodal Deep Transformation Models (DTMs) combining statistical approaches and neural networks to achieve strong predictive performance while preserving interpretability for tabular data. A key contribution of this work is the adaption of the xAI methods Grad-CAM and Occlusion to DTMs relying on 3D CNNs, enabling interpretation of the image branch through the generation of explanation maps. We developed DTMs to predict functional independence three months after stroke using diffusion-weighted imaging and clinical data from 407 patients. In a ten-fold cross-validation, the models achieved state-of-the-art predictive performance (AUC 0.81 [0.75, 0.87]) while maintaining interpretability for tabular features, with functional independence before stroke and stroke severity on admission emerging as the strongest predictors. Explanation maps from both xAI methods highlighted consistent regions, including frontal lobe areas which are known to be associated with age, a strong predictor of functional outcome. Notably, these regions disappeared once age was included as an explicit tabular predictor. Similarity analyses of explanation maps revealed distinct spatial patterns, providing meaningful insights into stroke pathophysiology, systematic error analysis and hypothesis generation.

This position paper for the CHI 2025 workshop "Everyday AR through AI-in-the-Loop" reflects on whether as a field HCI needs to define Responsible XR as a parallel to, and in conjunction with, Responsible AI, addressing the unique vulnerabilities posed by mass adoption of wearable AI-enabled AR glasses and XR devices that could enact AI-driven human perceptual augmentation.

The study of neural networks from the perspective of Fourier features has garnered significant attention. While existing analytical research suggests that neural networks tend to learn low-frequency features, a clear attribution method for identifying the specific learned Fourier features has remained elusive. To bridge this gap, we propose a novel Fourier feature attribution method grounded in signal decomposition theory. Additionally, we analyze the differences between game-theoretic attribution metrics for Fourier and spatial domain features, demonstrating that game-theoretic evaluation metrics are better suited for Fourier-based feature attribution. Our experiments show that Fourier feature attribution exhibits superior feature selection capabilities compared to spatial domain attribution methods. For instance, in the case of Vision Transformers (ViTs) on the ImageNet dataset, only $8\%$ of the Fourier features are required to maintain the original predictions for $80\%$ of the samples. Furthermore, we compare the specificity of features identified by our method against traditional spatial domain attribution methods. Results reveal that Fourier features exhibit greater intra-class concentration and inter-class distinctiveness, indicating their potential for more efficient classification and explainable AI algorithms.

To improve the trustworthiness of an AI model, finding consistent, understandable representations of its inference process is essential. This understanding is particularly important in high-stakes operations such as weather forecasting, where the identification of underlying meteorological mechanisms is as critical as the accuracy of the predictions. Despite the growing literature that addresses this issue through explainable AI, the applicability of their solutions is often limited due to their AI-centric development. To fill this gap, we follow a user-centric process to develop an example-based concept analysis framework, which identifies cases that follow a similar inference process as the target instance in a target model and presents them in a user-comprehensible format. Our framework provides the users with visually and conceptually analogous examples, including the probability of concept assignment to resolve ambiguities in weather mechanisms. To bridge the gap between vector representations identified from models and human-understandable explanations, we compile a human-annotated concept dataset and implement a user interface to assist domain experts involved in the the framework development.

CFIRE: A General Method for Combining Local Explanations

Sebastian Müller, Vanessa Toborek, Tamás Horváth +1

We propose a novel eXplainable AI algorithm to compute faithful, easy-to-understand, and complete global decision rules from local explanations for tabular data by combining XAI methods with closed frequent itemset mining. Our method can be used with any local explainer that indicates which dimensions are important for a given sample for a given black-box decision. This property allows our algorithm to choose among different local explainers, addressing the disagreement problem, \ie the observation that no single explanation method consistently outperforms others across models and datasets. Unlike usual experimental methodology, our evaluation also accounts for the Rashomon effect in model explainability. To this end, we demonstrate the robustness of our approach in finding suitable rules for nearly all of the 700 black-box models we considered across 14 benchmark datasets. The results also show that our method exhibits improved runtime, high precision and F1-score while generating compact and complete rules.

This paper introduces a novel formal SAT-based explanation model for deep learning in video understanding. The proposed method integrates SAT solving techniques with the principles of formal explainable AI to address the limitations of existing XAI techniques in this domain. By encoding deep learning models and video data into a logical framework and formulating explanation queries as satisfiability problems, the method aims to generate logic-based explanations with formal guarantees. The paper details the conceptual framework, the process of encoding deep learning models and video data, the formulation of "Why?" and "Why not?" questions, and a novel architecture integrating a SAT solver with a deep learning video understanding model. While challenges related to computational complexity and the representational power of propositional logic remain, the proposed approach offers a promising direction for enhancing the explainability of deep learning in the complex and critical domain of video understanding.

This paper proposes the e-person architecture for constructing a unified and incremental development of AI ethics. The e-person architecture takes the reduction of uncertainty through collaborative cognition and action with others as a unified basis for ethics. By classifying and defining uncertainty along two axes - (1) first, second, and third person perspectives, and (2) the difficulty of inference based on the depth of information - we support the development of unified and incremental development of AI ethics. In addition, we propose the e-person framework based on the free energy principle, which considers the reduction of uncertainty as a unifying principle of brain function, with the aim of implementing the e-person architecture, and we show our previous works and future challenges based on the proposed framework.

Developing competency in artificial intelligence is becoming increasingly crucial for computer science (CS) students at all levels of the CS curriculum. However, most previous research focuses on advanced CS courses, as traditional introductory courses provide limited opportunities to develop AI skills and knowledge. This paper introduces an introductory CS course where students learn computational thinking through computer vision, a sub-field of AI, as an application context. The course aims to achieve computational thinking outcomes alongside critical thinking outcomes that expose students to AI approaches and their societal implications. Through experiential activities such as individual projects and reading discussions, our course seeks to balance technical learning and critical thinking goals. Our evaluation, based on pre-and post-course surveys, shows an improved sense of belonging, self-efficacy, and AI ethics awareness among students. The results suggest that an AI-focused context can enhance participation and employability, student-selected projects support self-efficacy, and ethically grounded AI instruction can be effective for interdisciplinary audiences. Students' discussions on reading assignments demonstrated deep engagement with the complex challenges in today's AI landscape. Finally, we share insights on scaling such courses for larger cohorts and improving the learning experience for introductory CS students.

Large Language Models (LLMs) have demonstrated impressive capabilities on a broad array of NLP tasks, but their tendency to produce hallucinations$\unicode{x2013}$plausible-sounding but factually incorrect content$\unicode{x2013}$poses severe challenges in high-stakes domains. Existing hallucination detection methods either bear the computational cost of multiple inference passes or sacrifice accuracy for efficiency with single-pass approaches, neither of which is ideal in resource-constrained environments such as edge devices. We propose the Shannon Entropy Distribution Hallucination Detector (ShED-HD), a novel hallucination detection framework that bridges this gap by classifying sequence-level entropy patterns using a lightweight BiLSTM architecture with single-headed attention. In contrast to prior approaches, ShED-HD efficiently detects distinctive uncertainty patterns across entire output sequences, preserving contextual awareness. Through in-depth evaluation on three datasets (BioASQ, TriviaQA, and Jeopardy Questions), we show that ShED-HD significantly outperforms other computationally efficient approaches in the out-of-distribution setting, while achieving comparable performance in the in-distribution setting. ShED-HD facilitates hallucination detection that is low-cost, accurate, and generalizable, improving the credibility of content generated by LLMs in resource-constrained environments where trustworthy AI functionality is crucial.

The trajectory of intelligence evolution is often framed around the emergence of artificial general intelligence (AGI) and its alignment with human values. This paper challenges that framing by introducing the concept of intelligence sequencing: the idea that the order in which AGI and decentralized collective intelligence (DCI) emerge determines the long-term attractor basin of intelligence. Using insights from dynamical systems, evolutionary game theory, and network models, it argues that intelligence follows a path-dependent, irreversible trajectory. Once development enters a centralized (AGI-first) or decentralized (DCI-first) regime, transitions become structurally infeasible due to feedback loops and resource lock-in. Intelligence attractors are modeled in functional state space as the co-navigation of conceptual and adaptive fitness spaces. Early-phase structuring constrains later dynamics, much like renormalization in physics. This has major implications for AI safety: traditional alignment assumes AGI will emerge and must be controlled after the fact, but this paper argues that intelligence sequencing is more foundational. If AGI-first architectures dominate before DCI reaches critical mass, hierarchical monopolization and existential risk become locked in. If DCI-first emerges, intelligence stabilizes around decentralized cooperative equilibrium. The paper further explores whether intelligence structurally biases itself toward an attractor based on its self-modeling method -- externally imposed axioms (favoring AGI) vs. recursive internal visualization (favoring DCI). Finally, it proposes methods to test this theory via simulations, historical lock-in case studies, and intelligence network analysis. The findings suggest that intelligence sequencing is a civilizational tipping point: determining whether the future is shaped by unbounded competition or unbounded cooperation.