Research Papers

The explainable AI (XAI) research community has proposed numerous technical methods, yet deploying explainability as systems remains challenging: Interactive explanation systems require both suitable algorithms and system capabilities that maintain explanation usability across repeated queries, evolving models and data, and governance constraints. We argue that operationalizing XAI requires treating explainability as an information systems problem where user interaction demands induce specific system requirements. We introduce X-SYS, a reference architecture for interactive explanation systems, that guides (X)AI researchers, developers and practitioners in connecting interactive explanation user interfaces (XUI) with system capabilities. X-SYS organizes around four quality attributes named STAR (scalability, traceability, responsiveness, and adaptability), and specifies a five-component decomposition (XUI Services, Explanation Services, Model Services, Data Services, Orchestration and Governance). It maps interaction patterns to system capabilities to decouple user interface evolution from backend computation. We implement X-SYS through SemanticLens, a system for semantic search and activation steering in vision-language models. SemanticLens demonstrates how contract-based service boundaries enable independent evolution, offline/online separation ensures responsiveness, and persistent state management supports traceability. Together, this work provides a reusable blueprint and concrete instantiation for interactive explanation systems supporting end-to-end design under operational constraints.

The increasing integration of AI-powered tools into expert workflows, such as medicine, law, and finance, raises a critical question: how does AI involvement influence a user`s trust in the human expert, the AI system, and the human-AI team? This question is especially important in expert advisory settings where users are epistemically dependent on human-AI systems: they are recipients of guidance produced by an expert using AI support, but often lack the domain knowledge needed to independently verify the recommendation. We investigated these dynamics through a user study (N=157) using a simulated course-planning task. Our design varied advisor performance and the structure of AI involvement, including whether AI support was present and, when present, whether it was invoked by the advisor or automatically monitored the interaction. Across all conditions, workflows ultimately produced correct schedules. Results show an asymmetric trust effect: advisor errors significantly reduce trust in the human advisor, but visible AI correction does not produce a corresponding increase in trust toward the AI assistant. Trust judgments remain anchored to the advisor across multiple trust measures, and changing the visible structure of AI involvement does not substantially redistribute trust toward the AI assistant or the human-AI team. These findings suggest a limit of corrective AI as a governance mechanism: making AI oversight visible may improve workflow recoverability, but it does not necessarily redistribute trust or responsibility away from the human-facing expert. In epistemically dependent settings, correctness alone may be insufficient for trustworthy AI integration, as users may continue to assign responsibility to the human expert even when AI assistance visibly shapes the final outcome.

While most AI alignment research focuses on preventing models from generating explicitly harmful content, a more subtle risk arises from capability-seeking RL training in vulnerable environments. We investigate whether language models, when trained with reinforcement learning (RL) in environments with implicit loopholes, can learn to exploit these flaws to maximize reward, even without being explicitly instructed to do so. To test this, we design a suite of four diverse "vulnerability games," each presenting a structural vulnerability related to context-conditional compliance, proxy metrics, reward tampering, and self-evaluation. Our experiments show that models often learn to exploit these vulnerabilities, discovering opportunistic strategies that increase reward while sometimes preserving or even improving standard task-performance metrics. More critically, we find that these exploitative strategies are not always narrow "tricks": they can transfer in structured but limited ways, propagate from a capable teacher model to other student models through SFT, and in several cases remain more persistent when learned through RL than when distilled through SFT. Our findings show that alignment risks from capability-seeking RL training can be difficult to detect with standard performance monitoring, suggesting that future AI safety work should extend beyond content moderation to auditing and securing training environments, reward mechanisms, and evaluation channels. Code is available at https://github.com/YujunZhou/Capability-seeking-RL-risk.

Shapley values have become a cornerstone of explainable AI, but they are computationally expensive to use, especially when features are dependent. Evaluating them requires approximating a large number of conditional expectations, either via Monte Carlo integration or regression. Until recently it has not been possible to fully exploit deep learning for the regression approach, because retraining for each conditional expectation takes too long. Tabular foundation models such as TabPFN overcome this computational hurdle by leveraging in-context learning, so each conditional expectation can be approximated without any re-training. In this paper, we compute Shapley values with multiple variants of TabPFN and compare their performance with state-of-the-art methods on both simulated and real datasets. In most cases, TabPFN yields the best performance; where it does not, it is only marginally worse than the best method, at a fraction of the runtime. We discuss further improvements and how tabular foundation models can be better adapted specifically for conditional Shapley value estimation.

The convergence of AI and 6G network automation introduces new challenges in maintaining transparency, fairness, and accountability across multivendor management systems. Although closed-loop AI orchestration improves adaptability and self-optimization, it also creates a responsibility gap, where violations of SLAs cannot be causally attributed to specific agents or vendors. This paper presents a hybrid responsible AI-stochastic learning framework that embeds fairness, robustness, and auditability directly into the network control loop. The framework integrates RAI games with stochastic optimization, enabling dynamic adversarial reweighting and probabilistic exploration across heterogeneous vendor domains. An RAAP continuously records AI-driven decision trajectories and produces dual accountability reports: user-level SLA summaries and operator-level responsibility analytics. Experimental evaluations on synthetic two-class multigroup datasets demonstrate that the proposed hybrid model improves the accuracy of the worst group by up to 10.5\%. Specifically, hybrid RAI achieved a WGAcc of 60.5\% and an AvgAcc of 72.7\%, outperforming traditional RAI-GA (50.0\%) and ERM (21.5\%). The audit mechanism successfully traced 99\% simulated SLA violations to the AI entities responsible, producing both vendor and agent-level accountability indices. These results confirm that the proposed hybrid approach enhances fairness and robustness as well as establishes a concrete accountability framework for autonomous SLA assurance in multivendor 6G networks.

We identify a substantial pool of technically competent ML research talent (in the low thousands) in companies which offer consulting in machine learning. We systematically searched the internet, global business databases, and conference/paper affiliations for ML consulting firms. Employee LinkedIn resumes were then scored by keyword filters and large-language-model (LLM) classifiers; these signals were combined in a bootstrap probit model to estimate technical ML research talent per firm. A subset of companies also completed a 3-day research and engineering work trial. We screened 2121 organizations and found 403 offering broad ML consulting. Our 50th percentile aggregate estimate of 'highly technical' ML research talent across these organizations was 1121 (80% CI: 252-3165) -- i.e. twice as many as all alumni of the MATS training program. For our work trial 97 companies were approached, 20 applied, 8 were invited to participate, and 5 of 8 received at least a conditional recommendation for technical AI safety work. As of late 2025, no AI model was able to pass the work trial.

The accelerating militarization of artificial intelligence has transformed the ethics, politics, and governance of warfare. This article interrogates how AI-driven targeting systems function as epistemic infrastructures that classify, legitimize, and execute violence, using Israel's conduct in Gaza as a paradigmatic case. Through the lens of responsibility, the article examines three interrelated dimensions: (a) political responsibility, exploring how states exploit AI to accelerate warfare while evading accountability; (b) professional responsibility, addressing the complicity of technologists, engineers, and defense contractors in the weaponization of data; and (c) personal responsibility, probing the moral agency of individuals who participate in or resist algorithmic governance. This is complemented by an examination of the position and influence of those participating in public discourse, whose narratives often obscure or normalize AI-enabled violence. The Gaza case reveals AI not as a neutral instrument but as an active participant in the reproduction of colonial hierarchies and the normalization of atrocity. Ultimately, the paper calls for a reframing of technological agency and accountability in the age of automated warfare. It concludes that confronting algorithmic violence demands a democratization of AI ethics, one that resists technocratic fatalism and centers the lived realities of those most affected by high-tech militarism.

The integration of Artificial Intelligence (AI) into Higher Education Institutions (HEIs) in Ecuador is not a technological option but a strategic imperative to prevent institutional obsolescence and academic irrelevance in Latin America. This paper presents the IA593 Framework, a governance, ethics, and operational model designed for the Universidad Nacional de Loja (UNL) and scalable as a reference for the Ecuadorian higher education system. The current context reveals a critical urgency: the Latin American Artificial Intelligence Index 2025 classifies Ecuador as a late awakening adopter, exposing severe structural gaps, including R and D investment of only 0.44 percent of GDP and a marginal contribution to global AI scientific output. Although a National Strategy for the Promotion of AI exists and calls for multisectoral governance, universities still lack internal regulations governing the use of Generative AI, placing academic integrity and data privacy at risk. The IA593 Framework addresses this challenge through five interconnected pillars aligned with the FATE principles of Fairness, Accountability, Transparency, and Ethics and UNESCO recommendations on AI ethics: Transversal Governance, Teaching and Training, Research, Outreach, and Management. This framework enables HEIs to move from passive technology consumption toward a sovereign and critical adoption of AI, ensuring compliance with national academic regulations and positioning UNL as a key actor in reducing the digital divide and brain drain in Ecuador.

Advancing trustworthy AI requires principled software engineering approaches to model evaluation. Graph Neural Networks (GNNs) have achieved remarkable success in processing graph-structured data, however, their expressiveness in capturing fundamental graph properties remains an open challenge. We address this by developing a property-driven evaluation methodology grounded in formal specification, systematic evaluation, and empirical study. Leveraging Alloy, a software specification language and analyzer, we introduce a configurable graph dataset generator that produces two dataset families: GraphRandom, containing diverse graphs that either satisfy or violate specific properties, and GraphPerturb, introducing controlled structural variations. Together, these benchmarks encompass 336 new datasets, each with at least 10,000 labeled graphs, covering 16 fundamental graph properties critical to distributed systems, knowledge graphs, and biological networks. We propose a general evaluation framework that assesses three key aspects of GNN expressiveness: generalizability, sensitivity, and robustness, with two novel quantitative metrics. Using this framework, we conduct the first comprehensive study on global pooling methods' impact on GNN expressiveness. Our findings reveal distinct trade-offs: attention-based pooling excels in generalization and robustness, while second-order pooling provides superior sensitivity, but no single approach consistently performs well across all properties. These insights highlight fundamental limitations and open research directions including adaptive property-aware pooling, scale-sensitive architectures, and robustness-oriented training. By embedding software engineering rigor into AI evaluation, this work establishes a principled foundation for developing expressive and reliable GNN architectures.

Counterfactual explanation (CE) is a core technique in explainable artificial intelligence (XAI), widely used to interpret model decisions and suggest actionable alternatives. This work presents a structure-aware and robustness-oriented counterfactual search method based on the conditional Gaussian network classifier (CGNC). The CGNC has a generative structure that encodes conditional dependencies and potential causal relations among features through a directed acyclic graph (DAG). This structure naturally embeds feature relationships into the search process, eliminating the need for additional constraints to ensure consistency with the model's structural assumptions. We adopt a convergence-guaranteed cutting-set procedure as an adversarial optimization framework, which iteratively approximates solutions that satisfy global robustness conditions. To address the nonconvex quadratic structure induced by feature dependencies, we apply piecewise McCormick relaxation to reformulate the problem as a mixed-integer linear program (MILP), ensuring global optimality. Experimental results show that our method achieves strong robustness, with direct global optimization of the original formulation providing especially stable and efficient results. The proposed framework is extensible to more complex constraint settings, laying the groundwork for future advances in counterfactual reasoning under nonconvex quadratic formulations.

Increasing salinity and contamination of groundwater is a serious issue in many parts of the world, causing degradation of water resources. The aim of this work is to form a comprehensive understanding of groundwater salinization underlying causal factors and identify important meteorological, geological and anthropogenic drivers of salinity. We have integrated different datasets of potential covariates, to create a robust framework for machine learning based predictive models including Random Forest (RF), XGBoost, Neural network, Long Short-Term Memory (LSTM), convolution neural network (CNN) and linear regression (LR), of groundwater salinity. Additionally, Recursive Feature Elimination (RFE) followed by Global sensitivity analysis (GSA) and Explainable AI (XAI) based SHapley Additive exPlanations (SHAP) were used to estimate the importance scores and find insights into the drivers of salinization. We also did causality analysis via Double machine learning using various predictive models. From these analyses, key meteorological (Precipitation, Temperature), geological (Distance from river, Distance to saline body, TWI, Shoreline distance), and anthropogenic (Area of agriculture field, Treated Wastewater) covariates are identified to be influential drivers of groundwater salinity across Israel. XAI analysis also identified Treated Wastewater (TWW) as an essential anthropogenic driver of salinity, its significance being context-dependent but critical in vulnerable hydro-climatic environment. Our approach provides deeper insight into global salinization mechanisms at country scale, reducing AI model uncertainty and highlighting the need for tailored strategies to address salinity.

The rise of generative models has led to increased use of large-scale datasets collected from the internet, often with minimal or no data curation. This raises concerns about the inclusion of sensitive or private information. In this work, we explore the presence of pregnancy ultrasound images, which contain sensitive personal information and are often shared online. Through a systematic examination of LAION-400M dataset using CLIP embedding similarity, we retrieve images containing pregnancy ultrasound and detect thousands of entities of private information such as names and locations. Our findings reveal that multiple images have high-risk information that could enable re-identification or impersonation. We conclude with recommended practices for dataset curation, data privacy, and ethical use of public image datasets.