Research Papers

Multimodal learning has witnessed remarkable advancements in recent years, particularly with the integration of attention-based models, leading to significant performance gains across a variety of tasks. Parallel to this progress, the demand for explainable artificial intelligence (XAI) has spurred a growing body of research aimed at interpreting the complex decision-making processes of these models. This systematic literature review analyzes research published between January 2020 and early 2024 that focuses on the explainability of multimodal models. Framed within the broader goals of XAI, we examine the literature across multiple dimensions, including model architecture, modalities involved, explanation algorithms and evaluation methodologies. Our analysis reveals that most studies are concentrated on vision-language and language-only models, with attention-based techniques being the most commonly employed for explanation. However, these methods often fall short in capturing the full spectrum of interactions between modalities, a challenge further compounded by the architectural heterogeneity across domains. Importantly, we find that evaluation methods for XAI in multimodal settings are largely non-systematic, lacking consistency, robustness, and consideration for modality-specific cognitive and contextual factors. To address these gaps, we not only synthesize findings from the surveyed works but also incorporate a complementary analysis that integrates recent and emerging advances driving multimodal explainability. Based on these insights, we provide a comprehensive set of recommendations aimed at promoting rigorous, transparent, and standardized evaluation and reporting practices in multimodal XAI research. Our goal is to support future research in more interpretable, accountable, and responsible multimodal AI systems, with explainability at their core.

Accurate pain assessment in patients with limited ability to communicate, such as older adults with severe dementia, represents a critical healthcare challenge. Robust automated systems of pain behavior detection may facilitate such assessments. Existing pain detection datasets, however, suffer from limited ethnic/racial diversity, privacy constraints, and underrepresentation of older adults who are the primary target population for clinical deployment. We present SynPAIN, a large-scale synthetic dataset containing 10,710 facial expression images across five ethnicities/races, representing two age groups, and two genders. Using commercial generative AI tools, we created demographically balanced synthetic identities with clinically meaningful pain expressions. Our validation demonstrates that synthetic pain expressions exhibit expected pain patterns, scoring significantly higher than neutral and non-pain expressions using clinically validated pain assessment tools based on facial action unit analysis. We experimentally demonstrate SynPAIN's utility in identifying algorithmic bias in existing pain detection models. Through comprehensive bias evaluation, we reveal substantial performance disparities across demographics characteristics. These performance disparities were previously undetectable with smaller, less diverse datasets. Furthermore, we demonstrate that age-matched synthetic data augmentation improves pain detection performance on real clinical data, achieving a 2.4 percentage point improvement in average precision. SynPAIN addresses critical gaps in pain assessment research by providing the first publicly available, demographically diverse synthetic dataset specifically designed for older adult pain detection, while establishing a framework for measuring and mitigating algorithmic bias. The dataset, code, and trained models is available at https://mmzml.github.io/SynPAIN

Effective human-AI teaming heavily depends on swift trust, particularly in high-stakes scenarios such as emergency response, where timely and accurate decision-making is critical. In these time-sensitive and cognitively demanding settings, adaptive explainability is essential for fostering trust between human operators and AI systems. However, existing explainable AI (XAI) approaches typically offer uniform explanations and rely heavily on explicit feedback mechanisms, which are often impractical in such high-pressure scenarios. To address this gap, we propose a conceptual framework for adaptive XAI that operates non-intrusively by responding to users' real-time cognitive and emotional states through implicit feedback, thereby enhancing swift trust in high-stakes environments. The proposed adaptive explainability trust framework (AXTF) leverages physiological and behavioral signals, such as EEG, ECG, and eye tracking, to infer user states and support explanation adaptation. At its core is a multi-objective, personalized trust estimation model that maps workload, stress, and emotion to dynamic trust estimates. These estimates guide the modulation of explanation features enabling responsive and personalized support that promotes swift trust in human-AI collaboration. This conceptual framework establishes a foundation for developing adaptive, non-intrusive XAI systems tailored to the rigorous demands of high-pressure, time-sensitive environments.

Frontier AI systems are rapidly advancing in their capabilities to persuade, deceive, and influence human behaviour, with current models already demonstrating human-level persuasion and strategic deception in specific contexts. Humans are often the weakest link in cybersecurity systems, and a misaligned AI system deployed internally within a frontier company may seek to undermine human oversight by manipulating employees. Despite this growing threat, manipulation attacks have received little attention, and no systematic framework exists for assessing and mitigating these risks. To address this, we provide a detailed explanation of why manipulation attacks are a significant threat and could lead to catastrophic outcomes. Additionally, we present a safety case framework for manipulation risk, structured around three core lines of argument: inability, control, and trustworthiness. For each argument, we specify evidence requirements, evaluation methodologies, and implementation considerations for direct application by AI companies. This paper provides the first systematic methodology for integrating manipulation risk into AI safety governance, offering AI companies a concrete foundation to assess and mitigate these threats before deployment.

Understanding the sources of variability in annotations is crucial for developing fair NLP systems, especially for tasks like sexism detection where demographic bias is a concern. This study investigates the extent to which annotator demographic features influence labeling decisions compared to text content. Using a Generalized Linear Mixed Model, we quantify this inf luence, finding that while statistically present, demographic factors account for a minor fraction ( 8%) of the observed variance, with tweet content being the dominant factor. We then assess the reliability of Generative AI (GenAI) models as annotators, specifically evaluating if guiding them with demographic personas improves alignment with human judgments. Our results indicate that simplistic persona prompting often fails to enhance, and sometimes degrades, performance compared to baseline models. Furthermore, explainable AI (XAI) techniques reveal that model predictions rely heavily on content-specific tokens related to sexism, rather than correlates of demographic characteristics. We argue that focusing on content-driven explanations and robust annotation protocols offers a more reliable path towards fairness than potentially persona simulation.

Most safety testing efforts for large language models (LLMs) today focus on evaluating foundation models. However, there is a growing need to evaluate safety at the application level, as components such as system prompts, retrieval pipelines, and guardrails introduce additional factors that significantly influence the overall safety of LLM applications. In this paper, we introduce a practical framework for evaluating application-level safety in LLM systems, validated through real-world deployment across multiple use cases within our organization. The framework consists of two parts: (1) principles for developing customized safety risk taxonomies, and (2) practices for evaluating safety risks in LLM applications. We illustrate how the proposed framework was applied in our internal pilot, providing a reference point for organizations seeking to scale their safety testing efforts. This work aims to bridge the gap between theoretical concepts in AI safety and the operational realities of safeguarding LLM applications in practice, offering actionable guidance for safe and scalable deployment.

Artificial Intelligence (AI) tools for automating design artifact generation are increasingly used in Requirements Engineering (RE) to transform textual requirements into structured diagrams and models. While these AI tools, particularly those based on Natural Language Processing (NLP), promise to improve efficiency, their adoption remains limited in regulated industries where transparency and traceability are essential. In this paper, we investigate the explainability gap in AI-driven design artifact generation through semi-structured interviews with ten practitioners from safety-critical industries. We examine how current AI-based tools are integrated into workflows and the challenges arising from their lack of explainability. We also explore mitigation strategies, their impact on project outcomes, and features needed to improve usability. Our findings reveal that non-explainable AI outputs necessitate extensive manual validation, reduce stakeholder trust, struggle to handle domain-specific terminology, disrupt team collaboration, and introduce regulatory compliance risks, often negating the anticipated efficiency benefits. To address these issues, we identify key improvements, including source tracing, providing clear justifications for tool-generated decisions, supporting domain-specific adaptation, and enabling compliance validation. This study outlines a practical roadmap for improving the transparency, reliability, and applicability of AI tools in requirements engineering workflows, particularly in regulated and safety-critical environments where explainability is crucial for adoption and certification.

Artificial intelligence (AI) systems increasingly inform medical decision-making, yet concerns about algorithmic bias and inequitable outcomes persist, particularly for historically marginalized populations. This paper introduces the concept of Predictive Representativity (PR), a framework of fairness auditing that shifts the focus from the composition of the data set to outcomes-level equity. Through a case study in dermatology, we evaluated AI-based skin cancer classifiers trained on the widely used HAM10000 dataset and on an independent clinical dataset (BOSQUE Test set) from Colombia. Our analysis reveals substantial performance disparities by skin phototype, with classifiers consistently underperforming for individuals with darker skin, despite proportional sampling in the source data. We argue that representativity must be understood not as a static feature of datasets but as a dynamic, context-sensitive property of model predictions. PR operationalizes this shift by quantifying how reliably models generalize fairness across subpopulations and deployment contexts. We further propose an External Transportability Criterion that formalizes the thresholds for fairness generalization. Our findings highlight the ethical imperative for post-hoc fairness auditing, transparency in dataset documentation, and inclusive model validation pipelines. This work offers a scalable tool for diagnosing structural inequities in AI systems, contributing to discussions on equity, interpretability, and data justice and fostering a critical re-evaluation of fairness in data-driven healthcare.

Foundation Model Self-Play: Open-Ended Strategy Innovation via Foundation Models

Aaron Dharna, Cong Lu, Jeff Clune

Multi-agent interactions have long fueled innovation, from natural predator-prey dynamics to the space race. Self-play (SP) algorithms try to harness these dynamics by pitting agents against ever-improving opponents, thereby creating an implicit curriculum toward learning high-quality solutions. However, SP often fails to produce diverse solutions and can get stuck in locally optimal behaviors. We introduce Foundation-Model Self-Play (FMSP), a new direction that leverages the code-generation capabilities and vast knowledge of foundation models (FMs) to overcome these challenges by leaping across local optima in policy space. We propose a family of approaches: (1) \textbf{Vanilla Foundation-Model Self-Play (vFMSP)} continually refines agent policies via competitive self-play; (2) \textbf{Novelty-Search Self-Play (NSSP)} builds a diverse population of strategies, ignoring performance; and (3) the most promising variant, \textbf{Quality-Diveristy Self-Play (QDSP)}, creates a diverse set of high-quality policies by combining the diversity of NSSP and refinement of vFMSP. We evaluate FMSPs in Car Tag, a continuous-control pursuer-evader setting, and in Gandalf, a simple AI safety simulation in which an attacker tries to jailbreak an LLM's defenses. In Car Tag, FMSPs explore a wide variety of reinforcement learning, tree search, and heuristic-based methods, to name just a few. In terms of discovered policy quality, \ouralgo and vFMSP surpass strong human-designed strategies. In Gandalf, FMSPs can successfully automatically red-team an LLM, breaking through and jailbreaking six different, progressively stronger levels of defense. Furthermore, FMSPs can automatically proceed to patch the discovered vulnerabilities. Overall, FMSPs represent a promising new research frontier of improving self-play with foundation models, opening fresh paths toward more creative and open-ended strategy discovery

Evaluating the Critical Risks of Amazon's Nova Premier under the Frontier Model Safety Framework

Satyapriya Krishna, Ninareh Mehrabi, Abhinav Mohanty +4

Nova Premier is Amazon's most capable multimodal foundation model and teacher for model distillation. It processes text, images, and video with a one-million-token context window, enabling analysis of large codebases, 400-page documents, and 90-minute videos in a single prompt. We present the first comprehensive evaluation of Nova Premier's critical risk profile under the Frontier Model Safety Framework. Evaluations target three high-risk domains -- Chemical, Biological, Radiological & Nuclear (CBRN), Offensive Cyber Operations, and Automated AI R&D -- and combine automated benchmarks, expert red-teaming, and uplift studies to determine whether the model exceeds release thresholds. We summarize our methodology and report core findings. Based on this evaluation, we find that Nova Premier is safe for public release as per our commitments made at the 2025 Paris AI Safety Summit. We will continue to enhance our safety evaluation and mitigation pipelines as new risks and capabilities associated with frontier models are identified.

Compressing Deep Neural Networks Using Explainable AI

Kimia Soroush, Mohsen Raji, Behnam Ghavami

Deep neural networks (DNNs) have demonstrated remarkable performance in many tasks but it often comes at a high computational cost and memory usage. Compression techniques, such as pruning and quantization, are applied to reduce the memory footprint of DNNs and make it possible to accommodate them on resource-constrained edge devices. Recently, explainable artificial intelligence (XAI) methods have been introduced with the purpose of understanding and explaining AI methods. XAI can be utilized to get to know the inner functioning of DNNs, such as the importance of different neurons and features in the overall performance of DNNs. In this paper, a novel DNN compression approach using XAI is proposed to efficiently reduce the DNN model size with negligible accuracy loss. In the proposed approach, the importance score of DNN parameters (i.e. weights) are computed using a gradient-based XAI technique called Layer-wise Relevance Propagation (LRP). Then, the scores are used to compress the DNN as follows: 1) the parameters with the negative or zero importance scores are pruned and removed from the model, 2) mixed-precision quantization is applied to quantize the weights with higher/lower score with higher/lower number of bits. The experimental results show that, the proposed compression approach reduces the model size by 64% while the accuracy is improved by 42% compared to the state-of-the-art XAI-based compression method.

In the realm of Natural Language Processing (NLP), common approaches for handling human disagreement consist of aggregating annotators' viewpoints to establish a single ground truth. However, prior studies show that disregarding individual opinions can lead can lead to the side effect of underrepresenting minority perspectives, especially in subjective tasks, where annotators may systematically disagree because of their preferences. Recognizing that labels reflect the diverse backgrounds, life experiences, and values of individuals, this study proposes a new multi-perspective approach using soft labels to encourage the development of the next generation of perspective aware models, more inclusive and pluralistic. We conduct an extensive analysis across diverse subjective text classification tasks, including hate speech, irony, abusive language, and stance detection, to highlight the importance of capturing human disagreements, often overlooked by traditional aggregation methods. Results show that the multi-perspective approach not only better approximates human label distributions, as measured by Jensen-Shannon Divergence (JSD), but also achieves superior classification performance (higher F1 scores), outperforming traditional approaches. However, our approach exhibits lower confidence in tasks like irony and stance detection, likely due to the inherent subjectivity present in the texts. Lastly, leveraging Explainable AI (XAI), we explore model uncertainty and uncover meaningful insights into model predictions.