Research Papers

The rapid adoption of AI systems presents enterprises with a dual challenge: accelerating innovation while ensuring responsible governance. Current AI governance approaches suffer from fragmentation, with risk management frameworks that focus on isolated domains, regulations that vary across jurisdictions despite conceptual alignment, and high-level standards lacking concrete implementation guidance. This fragmentation increases governance costs and creates a false dichotomy between innovation and responsibility. We propose the Unified Control Framework (UCF): a comprehensive governance approach that integrates risk management and regulatory compliance through a unified set of controls. The UCF consists of three key components: (1) a comprehensive risk taxonomy synthesizing organizational and societal risks, (2) structured policy requirements derived from regulations, and (3) a parsimonious set of 42 controls that simultaneously address multiple risk scenarios and compliance requirements. We validate the UCF by mapping it to the Colorado AI Act, demonstrating how our approach enables efficient, adaptable governance that scales across regulations while providing concrete implementation guidance. The UCF reduces duplication of effort, ensures comprehensive coverage, and provides a foundation for automation, enabling organizations to achieve responsible AI governance without sacrificing innovation speed.

A Unified Framework with Novel Metrics for Evaluating the Effectiveness of XAI Techniques in LLMs

Melkamu Abay Mersha, Mesay Gemeda Yigezu, Hassan Shakil +3

The increasing complexity of LLMs presents significant challenges to their transparency and interpretability, necessitating the use of eXplainable AI (XAI) techniques to enhance trustworthiness and usability. This study introduces a comprehensive evaluation framework with four novel metrics for assessing the effectiveness of five XAI techniques across five LLMs and two downstream tasks. We apply this framework to evaluate several XAI techniques LIME, SHAP, Integrated Gradients, Layer-wise Relevance Propagation (LRP), and Attention Mechanism Visualization (AMV) using the IMDB Movie Reviews and Tweet Sentiment Extraction datasets. The evaluation focuses on four key metrics: Human-reasoning Agreement (HA), Robustness, Consistency, and Contrastivity. Our results show that LIME consistently achieves high scores across multiple LLMs and evaluation metrics, while AMV demonstrates superior Robustness and near-perfect Consistency. LRP excels in Contrastivity, particularly with more complex models. Our findings provide valuable insights into the strengths and limitations of different XAI methods, offering guidance for developing and selecting appropriate XAI techniques for LLMs.

Generating Counterfactual Explanations Under Temporal Constraints

Andrei Buliga, Chiara Di Francescomarino, Chiara Ghidini +2

Counterfactual explanations are one of the prominent eXplainable Artificial Intelligence (XAI) techniques, and suggest changes to input data that could alter predictions, leading to more favourable outcomes. Existing counterfactual methods do not readily apply to temporal domains, such as that of process mining, where data take the form of traces of activities that must obey to temporal background knowledge expressing which dynamics are possible and which not. Specifically, counterfactuals generated off-the-shelf may violate the background knowledge, leading to inconsistent explanations. This work tackles this challenge by introducing a novel approach for generating temporally constrained counterfactuals, guaranteed to comply by design with background knowledge expressed in Linear Temporal Logic on process traces (LTLp). We do so by infusing automata-theoretic techniques for LTLp inside a genetic algorithm for counterfactual generation. The empirical evaluation shows that the generated counterfactuals are temporally meaningful and more interpretable for applications involving temporal dependencies.

In this study, we examine the reliability of AI-based Voting Advice Applications (VAAs) and large language models (LLMs) in providing objective political information. Our analysis is based upon a comparison with party responses to 38 statements of the Wahl-O-Mat, a well-established German online tool that helps inform voters by comparing their views with political party positions. For the LLMs, we identify significant biases. They exhibit a strong alignment (over 75% on average) with left-wing parties and a substantially lower alignment with center-right (smaller 50%) and right-wing parties (around 30%). Furthermore, for the VAAs, intended to objectively inform voters, we found substantial deviations from the parties' stated positions in Wahl-O-Mat: While one VAA deviated in 25% of cases, another VAA showed deviations in more than 50% of cases. For the latter, we even observed that simple prompt injections led to severe hallucinations, including false claims such as non-existent connections between political parties and right-wing extremist ties.

A Close Look at Decomposition-based XAI-Methods for Transformer Language Models

Leila Arras, Bruno Puri, Patrick Kahardipraja +2

Various XAI attribution methods have been recently proposed for the transformer architecture, allowing for insights into the decision-making process of large language models by assigning importance scores to input tokens and intermediate representations. One class of methods that seems very promising in this direction includes decomposition-based approaches, i.e., XAI-methods that redistribute the model's prediction logit through the network, as this value is directly related to the prediction. In the previous literature we note though that two prominent methods of this category, namely ALTI-Logit and LRP, have not yet been analyzed in juxtaposition and hence we propose to close this gap by conducting a careful quantitative evaluation w.r.t. ground truth annotations on a subject-verb agreement task, as well as various qualitative inspections, using BERT, GPT-2 and LLaMA-3 as a testbed. Along the way we compare and extend the ALTI-Logit and LRP methods, including the recently proposed AttnLRP variant, from an algorithmic and implementation perspective. We further incorporate in our benchmark two widely-used gradient-based attribution techniques. Finally, we make our carefullly constructed benchmark dataset for evaluating attributions on language models, as well as our code, publicly available in order to foster evaluation of XAI-methods on a well-defined common ground.

Machine learning algorithms have shown promise in reducing bias in wind gust predictions, while still underpredicting high gusts. Uncertainty quantification (UQ) supports this issue by identifying when predictions are reliable or need cautious interpretation. Using data from 61 extratropical storms in the Northeastern USA, we introduce evidential neural network (ENN) as a novel approach for UQ in gust predictions, leveraging atmospheric variables from the Weather Research and Forecasting (WRF) model. Explainable AI techniques suggested that key predictive features contributed to higher uncertainty, which correlated strongly with storm intensity and spatial gust gradients. Compared to WRF, ENN demonstrated a 47% reduction in RMSE and allowed the construction of gust prediction intervals without an ensemble, successfully capturing at least 95% of observed gusts at 179 out of 266 stations. From an operational perspective, providing gust forecasts with quantified uncertainty enhances stakeholders' confidence in risk assessment and response planning for extreme gust events.

Any exploit taking advantage of zero-day is called a zero-day attack. Previous research and social media trends show a massive demand for research in zero-day attack detection. This paper analyzes Machine Learning (ML) and Deep Learning (DL) based approaches to create Intrusion Detection Systems (IDS) and scrutinizing them using Explainable AI (XAI) by training an explainer based on randomly sampled data from the testing set. The focus is on using the KDD99 dataset, which has the most research done among all the datasets for detecting zero-day attacks. The paper aims to synthesize the dataset to have fewer classes for multi-class classification, test ML and DL approaches on pattern recognition, establish the robustness and dependability of the model, and establish the interpretability and scalability of the model. We evaluated the performance of four multilayer perceptron (MLP) trained on the KDD99 dataset, including baseline ML models, weighted ML models, truncated ML models, and weighted truncated ML models. Our results demonstrate that the truncated ML model achieves the highest accuracy (99.62%), precision, and recall, while weighted truncated ML model shows lower accuracy (97.26%) but better class representation (less bias) among all the classes with improved unweighted recall score. We also used Shapely Additive exPlanations (SHAP) to train explainer for our truncated models to check for feature importance among the two weighted and unweighted models.

Improving the safety and reliability of large language models (LLMs) is a crucial aspect of realizing trustworthy AI systems. Although alignment methods aim to suppress harmful content generation, LLMs are often still vulnerable to jailbreaking attacks that employ adversarial inputs that subvert alignment and induce harmful outputs. We propose the Randomized Embedding Smoothing and Token Aggregation (RESTA) defense, which adds random noise to the embedding vectors and performs aggregation during the generation of each output token, with the aim of better preserving semantic information. Our experiments demonstrate that our approach achieves superior robustness versus utility tradeoffs compared to the baseline defenses.

Adding explanations to audio deepfake detection (ADD) models will boost their real-world application by providing insight on the decision making process. In this paper, we propose a relevancy-based explainable AI (XAI) method to analyze the predictions of transformer-based ADD models. We compare against standard Grad-CAM and SHAP-based methods, using quantitative faithfulness metrics as well as a partial spoof test, to comprehensively analyze the relative importance of different temporal regions in an audio. We consider large datasets, unlike previous works where only limited utterances are studied, and find that the XAI methods differ in their explanations. The proposed relevancy-based XAI method performs the best overall on a variety of metrics. Further investigation on the relative importance of speech/non-speech, phonetic content, and voice onsets/offsets suggest that the XAI results obtained from analyzing limited utterances don't necessarily hold when evaluated on large datasets.

The use of AI in healthcare has the potential to improve patient care, optimize clinical workflows, and enhance decision-making. However, bias, data incompleteness, and inaccuracies in training datasets can lead to unfair outcomes and amplify existing disparities. This research investigates the current state of dataset documentation practices, focusing on their ability to address these challenges and support ethical AI development. We identify shortcomings in existing documentation methods, which limit the recognition and mitigation of bias, incompleteness, and other issues in datasets. We propose the 'Healthcare AI Datasheet' to address these gaps, a dataset documentation framework that promotes transparency and ensures alignment with regulatory requirements. Additionally, we demonstrate how it can be expressed in a machine-readable format, facilitating its integration with datasets and enabling automated risk assessments. The findings emphasise the importance of dataset documentation in fostering responsible AI development.

A Comprehensive Framework to Operationalize Social Stereotypes for Responsible AI Evaluations

Aida Davani, Sunipa Dev, Héctor Pérez-Urbina +1

Societal stereotypes are at the center of a myriad of responsible AI interventions targeted at reducing the generation and propagation of potentially harmful outcomes. While these efforts are much needed, they tend to be fragmented and often address different parts of the issue without adopting a unified or holistic approach to social stereotypes and how they impact various parts of the machine learning pipeline. As a result, current interventions fail to capitalize on the underlying mechanisms that are common across different types of stereotypes, and to anchor on particular aspects that are relevant in certain cases. In this paper, we draw on social psychological research and build on NLP data and methods, to propose a unified framework to operationalize stereotypes in generative AI evaluations. Our framework identifies key components of stereotypes that are crucial in AI evaluation, including the target group, associated attribute, relationship characteristics, perceiving group, and context. We also provide considerations and recommendations for its responsible use.

The increasing use of AI technologies has led to increasing AI incidents, posing risks and causing harm to individuals, organizations, and society. This study recognizes and addresses the lack of standardized protocols for reliably and comprehensively gathering such incident data crucial for preventing future incidents and developing mitigating strategies. Specifically, this study analyses existing open-access AI-incident databases through a systematic methodology and identifies nine gaps in current AI incident reporting practices. Further, it proposes nine actionable recommendations to enhance standardization efforts to address these gaps. Ensuring the trustworthiness of enabling technologies such as AI is necessary for sustainable digital transformation. Our research promotes the development of standards to prevent future AI incidents and promote trustworthy AI, thus facilitating achieving the UN sustainable development goals. Through international cooperation, stakeholders can unlock the transformative potential of AI, enabling a sustainable and inclusive future for all.