Research Papers

Large pre-trained language models have become a crucial backbone for many downstream tasks in natural language processing (NLP), and while they are trained on a plethora of data containing a variety of biases, such as gender biases, it has been shown that they can also inherit such biases in their weights, potentially affecting their prediction behavior. However, it is unclear to what extent these biases also affect feature attributions generated by applying "explainable artificial intelligence" (XAI) techniques, possibly in unfavorable ways. To systematically study this question, we create a gender-controlled text dataset, GECO, in which the alteration of grammatical gender forms induces class-specific words and provides ground truth feature attributions for gender classification tasks. This enables an objective evaluation of the correctness of XAI methods. We apply this dataset to the pre-trained BERT model, which we fine-tune to different degrees, to quantitatively measure how pre-training induces undesirable bias in feature attributions and to what extent fine-tuning can mitigate such explanation bias. To this extent, we provide GECOBench, a rigorous quantitative evaluation framework for benchmarking popular XAI methods. We show a clear dependency between explanation performance and the number of fine-tuned layers, where XAI methods are observed to benefit particularly from fine-tuning or complete retraining of embedding layers.

The scientific method is the cornerstone of human progress across all branches of the natural and applied sciences, from understanding the human body to explaining how the universe works. The scientific method is based on identifying systematic rules or principles that describe the phenomenon of interest in a reproducible way that can be validated through experimental evidence. In the era of generative artificial intelligence, there are discussions on how AI systems may discover new knowledge. We argue that human complex reasoning for scientific discovery remains of vital importance, at least before the advent of artificial general intelligence. Yet, AI can be leveraged for scientific discovery via explainable AI. More specifically, knowing the `principles' the AI systems used to make decisions can be a point of contact with domain experts and scientists, that can lead to divergent or convergent views on a given scientific problem. Divergent views may spark further scientific investigations leading to interpretability-guided explanations (IGEs), and possibly to new scientific knowledge. We define this field as Explainable AI for Science, where domain experts -- potentially assisted by generative AI -- formulate scientific hypotheses and explanations based on the interpretability of a predictive AI system.

CoXQL: A Dataset for Parsing Explanation Requests in Conversational XAI Systems

Qianli Wang, Tatiana Anikina, Nils Feldhus +2

Conversational explainable artificial intelligence (ConvXAI) systems based on large language models (LLMs) have garnered significant interest from the research community in natural language processing (NLP) and human-computer interaction (HCI). Such systems can provide answers to user questions about explanations in dialogues, have the potential to enhance users' comprehension and offer more information about the decision-making and generation processes of LLMs. Currently available ConvXAI systems are based on intent recognition rather than free chat, as this has been found to be more precise and reliable in identifying users' intentions. However, the recognition of intents still presents a challenge in the case of ConvXAI, since little training data exist and the domain is highly specific, as there is a broad range of XAI methods to map requests onto. In order to bridge this gap, we present CoXQL, the first dataset in the NLP domain for user intent recognition in ConvXAI, covering 31 intents, seven of which require filling multiple slots. Subsequently, we enhance an existing parsing approach by incorporating template validations, and conduct an evaluation of several LLMs on CoXQL using different parsing strategies. We conclude that the improved parsing approach (MP+) surpasses the performance of previous approaches. We also discover that intents with multiple slots remain highly challenging for LLMs.

Faithful and Accurate Self-Attention Attribution for Message Passing Neural Networks via the Computation Tree Viewpoint

Yong-Min Shin, Siqing Li, Xin Cao +1

The self-attention mechanism has been adopted in various popular message passing neural networks (MPNNs), enabling the model to adaptively control the amount of information that flows along the edges of the underlying graph. Such attention-based MPNNs (Att-GNNs) have also been used as a baseline for multiple studies on explainable AI (XAI) since attention has steadily been seen as natural model interpretations, while being a viewpoint that has already been popularized in other domains (e.g., natural language processing and computer vision). However, existing studies often use naive calculations to derive attribution scores from attention, undermining the potential of attention as interpretations for Att-GNNs. In our study, we aim to fill the gap between the widespread usage of Att-GNNs and their potential explainability via attention. To this end, we propose GATT, edge attribution calculation method for self-attention MPNNs based on the computation tree, a rooted tree that reflects the computation process of the underlying model. Despite its simplicity, we empirically demonstrate the effectiveness of GATT in three aspects of model explanation: faithfulness, explanation accuracy, and case studies by using both synthetic and real-world benchmark datasets. In all cases, the results demonstrate that GATT greatly improves edge attribution scores, especially compared to the previous naive approach. Our code is available at https://github.com/jordan7186/GAtt.

It is imperative that breast cancer is detected precisely and timely to improve patient outcomes. Diagnostic methodologies have traditionally relied on unimodal approaches; however, medical data analytics is integrating diverse data sources beyond conventional imaging. Using multi-modal techniques, integrating both image and non-image data, marks a transformative advancement in breast cancer diagnosis. The purpose of this review is to explore the burgeoning field of multimodal techniques, particularly the fusion of histopathology images with non-image data. Further, Explainable AI (XAI) will be used to elucidate the decision-making processes of complex algorithms, emphasizing the necessity of explainability in diagnostic processes. This review utilizes multi-modal data and emphasizes explainability to enhance diagnostic accuracy, clinician confidence, and patient engagement, ultimately fostering more personalized treatment strategies for breast cancer, while also identifying research gaps in multi-modality and explainability, guiding future studies, and contributing to the strategic direction of the field.

Knowledge Transfer, Knowledge Gaps, and Knowledge Silos in Citation Networks

Eoghan Cunningham, Derek Greene

The advancement of science relies on the exchange of ideas across disciplines and the integration of diverse knowledge domains. However, tracking knowledge flows and interdisciplinary integration in rapidly evolving, multidisciplinary fields remains a significant challenge. This work introduces a novel network analysis framework to study the dynamics of knowledge transfer directly from citation data. By applying dynamic community detection to cumulative, time-evolving citation networks, we can identify research areas as groups of papers sharing knowledge sources and outputs. Our analysis characterises the life-cycles and knowledge transfer patterns of these dynamic communities over time. We demonstrate our approach through a case study of eXplainable Artificial Intelligence (XAI) research, an emerging interdisciplinary field at the intersection of machine learning, statistics, and psychology. Key findings include: (i) knowledge transfer between these important foundational topics and the contemporary topics in XAI research is limited, and the extent of knowledge transfer varies across different contemporary research topics; (ii) certain application domains exist as isolated "knowledge silos"; (iii) significant "knowledge gaps" are identified between related XAI research areas, suggesting opportunities for cross-pollination and improved knowledge integration. By mapping interdisciplinary integration and bridging knowledge gaps, this work can inform strategies to synthesise ideas from disparate sources and drive innovation. More broadly, our proposed framework enables new insights into the evolution of knowledge ecosystems directly from citation data, with applications spanning literature review, research planning, and science policy.

The rapid proliferation of open-source language models significantly increases the risks of downstream backdoor attacks. These backdoors can introduce dangerous behaviours during model deployment and can evade detection by conventional cybersecurity monitoring systems. In this paper, we introduce a novel class of backdoors in transformer models, that, in contrast to prior art, are unelicitable in nature. Unelicitability prevents the defender from triggering the backdoor, making it impossible to properly evaluate ahead of deployment even if given full white-box access and using automated techniques, such as red-teaming or certain formal verification methods. We show that our novel construction is not only unelicitable thanks to using cryptographic techniques, but also has favourable robustness properties. We confirm these properties in empirical investigations, and provide evidence that our backdoors can withstand state-of-the-art mitigation strategies. Additionally, we expand on previous work by showing that our universal backdoors, while not completely undetectable in white-box settings, can be harder to detect than some existing designs. By demonstrating the feasibility of seamlessly integrating backdoors into transformer models, this paper fundamentally questions the efficacy of pre-deployment detection strategies. This offers new insights into the offence-defence balance in AI safety and security.

AlphaGo plays chess and Go in a creative and novel way. It is natural for us to attribute contents to it, such as that it doesn't view being several pawns behind, if it has more board space, as bad. The framework introduced in Cappelen and Dever (2021) provides a way of thinking about the semantics and the metasemantics of AI content: does AlphaGo entertain contents like this, and if so, in virtue of what does a given state of the program mean that particular content? One salient question Cappelen and Dever didn't consider was the possibility of alien content. Alien content is content that is not or cannot be expressed by human beings. It's highly plausible that AlphaGo, or any other sophisticated AI system, expresses alien contents. That this is so, moreover, is plausibly a metasemantic fact: a fact that has to do with how AI comes to entertain content in the first place, one that will heed the vastly different etiology of AI and human content. This chapter explores the question of alien content in AI from a semantic and metasemantic perspective. It lays out the logical space of possible responses to the semantic and metasemantic questions alien content poses, considers whether and how we humans could communicate with entities who express alien content, and points out that getting clear about such questions might be important for more 'applied' issues in the philosophy of AI, such as existential risk and XAI.

With advanced AI/ML, there has been growing research on explainable AI (XAI) and studies on how humans interact with AI and XAI for effective human-AI collaborative decision-making. However, we still have a lack of understanding of how AI systems and XAI should be first presented to users without technical backgrounds. In this paper, we present the findings of semi-structured interviews with health professionals (n=12) and students (n=4) majoring in medicine and health to study how to improve onboarding with AI and XAI. For the interviews, we built upon human-AI interaction guidelines to create onboarding materials of an AI system for stroke rehabilitation assessment and AI explanations and introduce them to the participants. Our findings reveal that beyond presenting traditional performance metrics on AI, participants desired benchmark information, the practical benefits of AI, and interaction trials to better contextualize AI performance, and refine the objectives and performance of AI. Based on these findings, we highlight directions for improving onboarding with AI and XAI and human-AI collaborative decision-making.

Mitigating Text Toxicity with Counterfactual Generation

Milan Bhan, Jean-Noel Vittaut, Nina Achache +5

Toxicity mitigation consists in rephrasing text in order to remove offensive or harmful meaning. Neural natural language processing (NLP) models have been widely used to target and mitigate textual toxicity. However, existing methods fail to detoxify text while preserving the initial non-toxic meaning at the same time. In this work, we propose to apply counterfactual generation methods from the eXplainable AI (XAI) field to target and mitigate textual toxicity. In particular, we perform text detoxification by applying local feature importance and counterfactual generation methods to a toxicity classifier distinguishing between toxic and non-toxic texts. We carry out text detoxification through counterfactual generation on three datasets and compare our approach to three competitors. Automatic and human evaluations show that recently developed NLP counterfactual generators can mitigate toxicity accurately while better preserving the meaning of the initial text as compared to classical detoxification methods. Finally, we take a step back from using automated detoxification tools, and discuss how to manage the polysemous nature of toxicity and the risk of malicious use of detoxification tools. This work is the first to bridge the gap between counterfactual generation and text detoxification and paves the way towards more practical application of XAI methods.

The accelerated progress of artificial intelligence (AI) has popularized deep learning models across various domains, yet their inherent opacity poses challenges, particularly in critical fields like healthcare, medicine, and the geosciences. Explainable AI (XAI) has emerged to shed light on these 'black box' models, aiding in deciphering their decision-making processes. However, different XAI methods often produce significantly different explanations, leading to high inter-method variability that increases uncertainty and undermines trust in deep networks' predictions. In this study, we address this challenge by introducing a novel framework designed to enhance the explainability of deep networks through a dual focus on maximizing both accuracy and comprehensibility in the explanations. Our framework integrates outputs from multiple established XAI methods and leverages a non-linear neural network model, termed the 'explanation optimizer,' to construct a unified, optimal explanation. The optimizer evaluates explanations using two key metrics: faithfulness (accuracy in reflecting the network's decisions) and complexity (comprehensibility). By balancing these, it provides accurate and accessible explanations, addressing a key XAI limitation. Experiments on multi-class and binary classification in 2D object and 3D neuroscience imaging confirm its efficacy. Our optimizer achieved faithfulness scores 155% and 63% higher than the best XAI methods in 3D and 2D tasks, respectively, while also reducing complexity for better understanding. These results demonstrate that optimal explanations based on specific quality criteria are achievable, offering a solution to the issue of inter-method variability in the current XAI literature and supporting more trustworthy deep network predictions

Design Opportunities for Explainable AI Paraphrasing Tools: A User Study with Non-native English Speakers

Yewon Kim, Thanh-Long V. Le, Donghwi Kim +2

We investigate how non-native English speakers (NNESs) interact with diverse information aids to assess and select AI-generated paraphrases. We develop ParaScope, an AI paraphrasing assistant that integrates diverse information aids, such as back-translation, explanations, and usage examples, and logs user interaction data. Our in-lab study with 22 NNESs reveals that user preferences for information aids vary by language proficiency, with workflows progressing from global to more detailed information. While back-translation was the most frequently used aid, it was not a decisive factor in suggestion acceptance; users combined multiple information aids to make informed decisions. Our findings demonstrate the potential of explainable AI paraphrasing tools to enhance NNESs' confidence, autonomy, and writing efficiency, while also emphasizing the importance of thoughtful design to prevent information overload. Based on these findings, we offer design implications for explainable AI paraphrasing tools that support NNESs in making informed decisions when using AI writing systems.