Abstract
Artificial intelligence (AI) is a digital technology that will be of major importance for the development of humanity in the near future. AI has raised fundamental questions about what we should do with such systems, what the systems themselves should do, what risks they involve and how we can control these. - After the background to the field (1), this article introduces the main debates (2), first on ethical issues that arise with AI systems as objects, i.e. tools made and used by humans; here, the main sections are privacy (2.1), manipulation (2.2), opacity (2.3), bias (2.4), autonomy & responsibility (2.6) and the singularity (2.7). Then we look at AI systems as subjects, i.e. when ethics is for the AI systems themselves in machine ethics (2.8.) and artificial moral agency (2.9). Finally we look at future developments and the concept of AI (3). For each section within these themes, we provide a general explanation of the ethical issues, we outline existing positions and arguments, then we analyse how this plays out with current technologies and finally what policy consequences may be drawn.

Abstract
This study demonstrates the extent to which prominent debates about the future of AI are best understood as subjective, philosophical disagreements over the history and future of technological change rather than as objective, material disagreements over the technologies themselves. It focuses on the deep disagreements over whether artificial general intelligence (AGI) will prove transformative for human society; a question that is analytically prior to that of whether this transformative effect will help or harm humanity. The study begins by distinguishing two fundamental camps in this debate. The first of these can be identified as "transformationalists," who argue that continued AI development will inevitably have a profound effect on society. Opposed to them are "skeptics," a more eclectic group united by their disbelief that AI can or will live up to such high expectations. Each camp admits further "strong" and "weak" variants depending on their tolerance for epistemic risk. These stylized contrasts help to identify a set of fundamental questions that shape the camps' respective interpretations of the future of AI. Three questions in particular are focused on: the possibility of non-biological intelligence, the appropriate time frame of technological predictions, and the assumed trajectory of technological development. In highlighting these specific points of non-technical disagreement, this study demonstrates the wide range of different arguments used to justify either the transformationalist or skeptical position. At the same time, it highlights the strong argumentative burden of the transformationalist position, the way that belief in this position creates competitive pressures to achieve first-mover advantage, and the need to widen the concept of "expertise" in debates surrounding the future development of AI.

Abstract
Large Vision-Language Models (LVLMs) augmented with Retrieval-Augmented Generation (RAG) are increasingly employed in medical AI to enhance factual grounding through external clinical image-text retrieval. However, this reliance creates a significant attack surface. We propose MedThreatRAG, a novel multimodal poisoning framework that systematically probes vulnerabilities in medical RAG systems by injecting adversarial image-text pairs. A key innovation of our approach is the construction of a simulated semi-open attack environment, mimicking real-world medical systems that permit periodic knowledge base updates via user or pipeline contributions. Within this setting, we introduce and emphasize Cross-Modal Conflict Injection (CMCI), which embeds subtle semantic contradictions between medical images and their paired reports. These mismatches degrade retrieval and generation by disrupting cross-modal alignment while remaining sufficiently plausible to evade conventional filters. While basic textual and visual attacks are included for completeness, CMCI demonstrates the most severe degradation. Evaluations on IU-Xray and MIMIC-CXR QA tasks show that MedThreatRAG reduces answer F1 scores by up to 27.66% and lowers LLaVA-Med-1.5 F1 rates to as low as 51.36%. Our findings expose fundamental security gaps in clinical RAG systems and highlight the urgent need for threat-aware design and robust multimodal consistency checks. Finally, we conclude with a concise set of guidelines to inform the safe development of future multimodal medical RAG systems.

Abstract
In recent years, breakthroughs in artificial intelligence (AI) technology have triggered global industrial transformations, with applications permeating various fields such as finance, healthcare, education, and manufacturing. However, this rapid iteration is accompanied by irrational development, where enterprises blindly invest due to technology hype, often overlooking systematic value assessments. This paper develops a multi-dimensional evaluation model that integrates information theory's entropy reduction principle, economics' bounded rationality framework, and psychology's irrational decision theories to quantify AI product value. Key factors include positive dimensions (e.g., uncertainty elimination, efficiency gains, cost savings, decision quality improvement) and negative risks (e.g., error probability, impact, and correction costs). A non-linear formula captures factor couplings, and validation through 10 commercial cases demonstrates the model's effectiveness in distinguishing successful and failed products, supporting hypotheses on synergistic positive effects, non-linear negative impacts, and interactive regulations. Results reveal value generation logic, offering enterprises tools to avoid blind investments and promote rational AI industry development. Future directions include adaptive weights, dynamic mechanisms, and extensions to emerging AI technologies like generative models.

Abstract
As AI agents become more widely deployed, we are likely to see an increasing number of incidents: events involving AI agent use that directly or indirectly cause harm. For example, agents could be prompt-injected to exfiltrate private information or make unauthorized purchases. Structured information about such incidents (e.g., user prompts) can help us understand their causes and prevent future occurrences. However, existing incident reporting processes are not sufficient for understanding agent incidents. In particular, such processes are largely based on publicly available data, which excludes useful, but potentially sensitive, information such as an agent's chain of thought or browser history. To inform the development of new, emerging incident reporting processes, we propose an incident analysis framework for agents. Drawing on systems safety approaches, our framework proposes three types of factors that can cause incidents: system-related (e.g., CBRN training data), contextual (e.g., prompt injections), and cognitive (e.g., misunderstanding a user request). We also identify specific information that could help clarify which factors are relevant to a given incident: activity logs, system documentation and access, and information about the tools an agent uses. We provide recommendations for 1) what information incident reports should include and 2) what information developers and deployers should retain and make available to incident investigators upon request. As we transition to a world with more agents, understanding agent incidents will become increasingly crucial for managing risks.

Abstract
As Artificial Intelligence (AI) becomes increasingly integrated into daily life, there is a growing need to equip the next generation with the ability to apply, interact with, evaluate, and collaborate with AI systems responsibly. Prior research highlights the urgent demand from K-12 educators to teach students the ethical and effective use of AI for learning. To address this need, we designed an Large-Language Model (LLM)-based module to teach prompting literacy. This includes scenario-based deliberate practice activities with direct interaction with intelligent LLM agents, aiming to foster secondary school students' responsible engagement with AI chatbots. We conducted two iterations of classroom deployment in 11 authentic secondary education classrooms, and evaluated 1) AI-based auto-grader's capability; 2) students' prompting performance and confidence changes towards using AI for learning; and 3) the quality of learning and assessment materials. Results indicated that the AI-based auto-grader could grade student-written prompts with satisfactory quality. In addition, the instructional materials supported students in improving their prompting skills through practice and led to positive shifts in their perceptions of using AI for learning. Furthermore, data from Study 1 informed assessment revisions in Study 2. Analyses of item difficulty and discrimination in Study 2 showed that True/False and open-ended questions could measure prompting literacy more effectively than multiple-choice questions for our target learners. These promising outcomes highlight the potential for broader deployment and highlight the need for broader studies to assess learning effectiveness and assessment design.

Abstract
K-12 educators are increasingly using Large Language Models (LLMs) to create instructional materials. These systems excel at producing fluent, coherent content, but often lack support for high-quality teaching. The reason is twofold: first, commercial LLMs, such as ChatGPT and Gemini which are among the most widely accessible to teachers, do not come preloaded with the depth of pedagogical theory needed to design truly effective activities; second, although sophisticated prompt engineering can bridge this gap, most teachers lack the time or expertise and find it difficult to encode such pedagogical nuance into their requests. This study shifts pedagogical expertise from the user's prompt to the LLM's internal architecture. We embed the well-established Knowledge-Learning-Instruction (KLI) framework into a Multi-Agent System (MAS) to act as a sophisticated instructional designer. We tested three systems for generating secondary Math and Science learning activities: a Single-Agent baseline simulating typical teacher prompts; a role-based MAS where agents work sequentially; and a collaborative MAS-CMD where agents co-construct activities through conquer and merge discussion. The generated materials were evaluated by 20 practicing teachers and a complementary LLM-as-a-judge system using the Quality Matters (QM) K-12 standards. While the rubric scores showed only small, often statistically insignificant differences between the systems, the qualitative feedback from educators painted a clear and compelling picture. Teachers strongly preferred the activities from the collaborative MAS-CMD, describing them as significantly more creative, contextually relevant, and classroom-ready. Our findings show that embedding pedagogical principles into LLM systems offers a scalable path for creating high-quality educational content.

Abstract
The concepts of ``human-centered AI'' and ``value-based decision'' have gained significant attention in both research and industry. However, many critical aspects remain underexplored and require further investigation. In particular, there is a need to understand how systems incorporate human values, how humans can identify these values within systems, and how to minimize the risks of harm or unintended consequences. In this paper, we highlight the need to rethink how we frame value alignment and assert that value alignment should move beyond static and singular conceptions of values. We argue that AI systems should implement long-term reasoning and remain adaptable to evolving values. Furthermore, value alignment requires more theories to address the full spectrum of human values. Since values often vary among individuals or groups, multi-agent systems provide the right framework for navigating pluralism, conflict, and inter-agent reasoning about values. We identify the challenges associated with value alignment and indicate directions for advancing value alignment research. In addition, we broadly discuss diverse perspectives of value alignment, from design methodologies to practical applications.

Abstract
The trustworthiness of AI is considered essential to the adoption and application of AI systems. However, the meaning of trust varies across industry, research and policy spaces. Studies suggest that professionals who develop and use AI regard an AI system as trustworthy based on their personal experiences and social relations at work. Studies about trust in AI and the constructs that aim to operationalise trust in AI (e.g., consistency, reliability, explainability and accountability). However, the majority of existing studies about trust in AI are situated in Western, Educated, Industrialised, Rich and Democratic (WEIRD) societies. The few studies about trust and AI in Africa do not include the views of people who develop, study or use AI in their work. In this study, we surveyed 157 people with professional and/or educational interests in AI from 25 African countries, to explore how they conceptualised trust in AI. Most respondents had links with workshops about trust and AI in Africa in Namibia and Ghana. Respondents' educational background, transnational mobility, and country of origin influenced their concerns about AI systems. These factors also affected their levels of distrust in certain AI applications and their emphasis on specific principles designed to foster trust. Respondents often expressed that their values are guided by the communities in which they grew up and emphasised communal relations over individual freedoms. They described trust in many ways, including applying nuances of Afro-relationalism to constructs in international discourse, such as reliability and reliance. Thus, our exploratory study motivates more empirical research about the ways trust is practically enacted and experienced in African social realities of AI design, use and governance.

Abstract
Artificial intelligence (AI) is expected to serve as a foundational capability across the entire lifecycle of 6G networks, spanning design, deployment, and operation. This article proposes a native AI-driven air interface architecture built around two core characteristics: compression and adaptation. On one hand, compression enables the system to understand and extract essential semantic information from the source data, focusing on task relevance rather than symbol-level accuracy. On the other hand, adaptation allows the air interface to dynamically transmit semantic information across diverse tasks, data types, and channel conditions, ensuring scalability and robustness. This article first introduces the native AI-driven air interface architecture, then discusses representative enabling methodologies, followed by a case study on semantic communication in 6G non-terrestrial networks. Finally, it presents a forward-looking discussion on the future of native AI in 6G, outlining key challenges and research opportunities.

Abstract
The increasing adoption of low-cost environmental sensors and AI-enabled applications has accelerated the demand for scalable and resilient data infrastructures, particularly in data-scarce and resource-constrained regions. This paper presents the design, implementation, and evaluation of the AirQo data pipeline: a modular, cloud-native Extract-Transform-Load (ETL) system engineered to support both real-time and batch processing of heterogeneous air quality data across urban deployments in Africa. It is Built using open-source technologies such as Apache Airflow, Apache Kafka, and Google BigQuery. The pipeline integrates diverse data streams from low-cost sensors, third-party weather APIs, and reference-grade monitors to enable automated calibration, forecasting, and accessible analytics. We demonstrate the pipeline's ability to ingest, transform, and distribute millions of air quality measurements monthly from over 400 monitoring devices while achieving low latency, high throughput, and robust data availability, even under constrained power and connectivity conditions. The paper details key architectural features, including workflow orchestration, decoupled ingestion layers, machine learning-driven sensor calibration, and observability frameworks. Performance is evaluated across operational metrics such as resource utilization, ingestion throughput, calibration accuracy, and data availability, offering practical insights into building sustainable environmental data platforms. By open-sourcing the platform and documenting deployment experiences, this work contributes a reusable blueprint for similar initiatives seeking to advance environmental intelligence through data engineering in low-resource settings.

Abstract
The transformative power of AI is undeniable - but as user adoption accelerates, so does the need to understand and mitigate the environmental impact of AI serving. However, no studies have measured AI serving environmental metrics in a production environment. This paper addresses this gap by proposing and executing a comprehensive methodology for measuring the energy usage, carbon emissions, and water consumption of AI inference workloads in a large-scale, AI production environment. Our approach accounts for the full stack of AI serving infrastructure - including active AI accelerator power, host system energy, idle machine capacity, and data center energy overhead. Through detailed instrumentation of Google's AI infrastructure for serving the Gemini AI assistant, we find the median Gemini Apps text prompt consumes 0.24 Wh of energy - a figure substantially lower than many public estimates. We also show that Google's software efficiency efforts and clean energy procurement have driven a 33x reduction in energy consumption and a 44x reduction in carbon footprint for the median Gemini Apps text prompt over one year. We identify that the median Gemini Apps text prompt uses less energy than watching nine seconds of television (0.24 Wh) and consumes the equivalent of five drops of water (0.26 mL). While these impacts are low compared to other daily activities, reducing the environmental impact of AI serving continues to warrant important attention. Towards this objective, we propose that a comprehensive measurement of AI serving environmental metrics is critical for accurately comparing models, and to properly incentivize efficiency gains across the full AI serving stack.

Abstract
Fairness in machine learning is more important than ever as ethical concerns continue to grow. Individual fairness demands that individuals differing only in sensitive attributes receive the same outcomes. However, commonly used machine learning algorithms often fail to achieve such fairness. To improve individual fairness, various training methods have been developed, such as incorporating fairness constraints as optimisation objectives. While these methods have demonstrated empirical effectiveness, they lack formal guarantees of fairness. Existing approaches that aim to provide fairness guarantees primarily rely on verification techniques, which can sometimes fail to produce definitive results. Moreover, verification alone does not actively enhance individual fairness during training. To address this limitation, we propose a novel framework that formally guarantees individual fairness throughout training. Our approach consists of two parts, i.e., (1) provably fair initialisation that ensures the model starts in a fair state, and (2) a fairness-preserving training algorithm that maintains fairness as the model learns. A key element of our method is the use of randomised response mechanisms, which protect sensitive attributes while maintaining fairness guarantees. We formally prove that this mechanism sustains individual fairness throughout the training process. Experimental evaluations confirm that our approach is effective, i.e., producing models that are empirically fair and accurate. Furthermore, our approach is much more efficient than the alternative approach based on certified training (which requires neural network verification during training).

Abstract
Machine Learning software systems are frequently used in our day-to-day lives. Some of these systems are used in various sensitive environments to make life-changing decisions. Therefore, it is crucial to ensure that these AI/ML systems do not make any discriminatory decisions for any specific groups or populations. In that vein, different bias detection and mitigation open-source software libraries (aka API libraries) are being developed and used. In this paper, we conduct a qualitative study to understand in what scenarios these open-source fairness APIs are used in the wild, how they are used, and what challenges the developers of these APIs face while developing and adopting these libraries. We have analyzed 204 GitHub repositories (from a list of 1885 candidate repositories) which used 13 APIs that are developed to address bias in ML software. We found that these APIs are used for two primary purposes (i.e., learning and solving real-world problems), targeting 17 unique use-cases. Our study suggests that developers are not well-versed in bias detection and mitigation; they face lots of troubleshooting issues, and frequently ask for opinions and resources. Our findings can be instrumental for future bias-related software engineering research, and for guiding educators in developing more state-of-the-art curricula.

Abstract
Urban transportation systems encounter diverse challenges across multiple tasks, such as traffic forecasting, electric vehicle (EV) charging demand prediction, and taxi dispatch. Existing approaches suffer from two key limitations: small-scale deep learning models are task-specific and data-hungry, limiting their generalizability across diverse scenarios, while large language models (LLMs), despite offering flexibility through natural language interfaces, struggle with structured spatiotemporal data and numerical reasoning in transportation domains. To address these limitations, we propose TransLLM, a unified foundation framework that integrates spatiotemporal modeling with large language models through learnable prompt composition. Our approach features a lightweight spatiotemporal encoder that captures complex dependencies via dilated temporal convolutions and dual-adjacency graph attention networks, seamlessly interfacing with LLMs through structured embeddings. A novel instance-level prompt routing mechanism, trained via reinforcement learning, dynamically personalizes prompts based on input characteristics, moving beyond fixed task-specific templates. The framework operates by encoding spatiotemporal patterns into contextual representations, dynamically composing personalized prompts to guide LLM reasoning, and projecting the resulting representations through specialized output layers to generate task-specific predictions. Experiments across seven datasets and three tasks demonstrate the exceptional effectiveness of TransLLM in both supervised and zero-shot settings. Compared to ten baseline models, it delivers competitive performance on both regression and planning problems, showing strong generalization and cross-task adaptability. Our code is available at https://github.com/BiYunying/TransLLM.

Abstract
The ninth AI City Challenge continues to advance real-world applications of computer vision and AI in transportation, industrial automation, and public safety. The 2025 edition featured four tracks and saw a 17% increase in participation, with 245 teams from 15 countries registered on the evaluation server. Public release of challenge datasets led to over 30,000 downloads to date. Track 1 focused on multi-class 3D multi-camera tracking, involving people, humanoids, autonomous mobile robots, and forklifts, using detailed calibration and 3D bounding box annotations. Track 2 tackled video question answering in traffic safety, with multi-camera incident understanding enriched by 3D gaze labels. Track 3 addressed fine-grained spatial reasoning in dynamic warehouse environments, requiring AI systems to interpret RGB-D inputs and answer spatial questions that combine perception, geometry, and language. Both Track 1 and Track 3 datasets were generated in NVIDIA Omniverse. Track 4 emphasized efficient road object detection from fisheye cameras, supporting lightweight, real-time deployment on edge devices. The evaluation framework enforced submission limits and used a partially held-out test set to ensure fair benchmarking. Final rankings were revealed after the competition concluded, fostering reproducibility and mitigating overfitting. Several teams achieved top-tier results, setting new benchmarks in multiple tasks.

Abstract
Machine learning interatomic potentials (MLIPs) have achieved remarkable accuracy on standard benchmarks, yet their ability to reproduce molecular kinetics -- critical for reaction rate calculations -- remains largely unexplored. We introduce Landscape17, a dataset of complete kinetic transition networks (KTNs) for the molecules of the MD17 dataset, computed using hybrid-level density functional theory. Each KTN contains minima, transition states, and approximate steepest-descent paths, along with energies, forces, and Hessian eigenspectra at stationary points. We develop a comprehensive test suite to evaluate the MLIP ability to reproduce these reference landscapes and apply it to a number of state-of-the-art architectures. Our results reveal limitations in current MLIPs: all the models considered miss over half of the DFT transition states and generate stable unphysical structures throughout the potential energy surface. Data augmentation with pathway configurations improves reproduction of DFT potential energy surfaces, resulting in significant improvement in the global kinetics. However, these models still produce many spurious stable structures, indicating that current MLIP architectures face underlying challenges in capturing the topology of molecular potential energy surfaces. The Landscape17 benchmark provides a straightforward but demanding test of MLIPs for kinetic applications, requiring only up to a few hours of compute time. We propose this test for validation of next-generation MLIPs targeting reaction discovery and rate prediction.

Abstract
The utilization of Machine Learning (ML) in contemporary software systems is extensive and continually expanding. However, its usage is energy-intensive, contributing to increased carbon emissions and demanding significant resources. While numerous studies examine the performance and accuracy of ML, only a limited few focus on its environmental aspects, particularly energy consumption. In addition, despite emerging efforts to compare energy consumption across various programming languages for specific algorithms and tasks, there remains a gap specifically in comparing these languages for ML-based tasks. This paper aims to raise awareness of the energy costs associated with employing different programming languages for ML model training and inference. Through this empirical study, we measure and compare the energy consumption along with run-time performance of five regression and five classification tasks implemented in Python and R, the two most popular programming languages in this context. Our study results reveal a statistically significant difference in costs between the two languages in 95% of the cases examined. Furthermore, our analysis demonstrates that the choice of programming language can influence energy efficiency significantly, up to 99.16% during model training and up to 99.8% during inferences, for a given ML task.

Abstract
Large Artificial Intelligence (AI) training workloads spanning several tens of thousands of GPUs present unique power management challenges. These arise due to the high variability in power consumption during the training. Given the synchronous nature of these jobs, during every iteration there is a computation-heavy phase, where each GPU works on the local data, and a communication-heavy phase where all the GPUs synchronize on the data. Because compute-heavy phases require much more power than communication phases, large power swings occur. The amplitude of these power swings is ever increasing with the increase in the size of training jobs. An even bigger challenge arises from the frequency spectrum of these power swings which, if harmonized with critical frequencies of utilities, can cause physical damage to the power grid infrastructure. Therefore, to continue scaling AI training workloads safely, we need to stabilize the power of such workloads. This paper introduces the challenge with production data and explores innovative solutions across the stack: software, GPU hardware, and datacenter infrastructure. We present the pros and cons of each of these approaches and finally present a multi-pronged approach to solving the challenge. The proposed solutions are rigorously tested using a combination of real hardware and Microsoft's in-house cloud power simulator, providing critical insights into the efficacy of these interventions under real-world conditions.

Abstract
This paper offers an overview of the prospects and ethics of using AI to achieve human enhancement, and more broadly what we call intellectual augmentation (IA). After explaining the central notions of human enhancement, IA, and AI, we discuss the state of the art in terms of the main technologies for IA, with or without brain-computer interfaces. Given this picture, we discuss potential ethical problems, namely inadequate performance, safety, coercion and manipulation, privacy, cognitive liberty, authenticity, and fairness in more detail. We conclude that while there are very significant technical hurdles to real human enhancement through AI, and significant ethical problems, there are also significant benefits that may realistically be achieved in ways that are consonant with a rights-based ethics as well. We also highlight the specific concerns that apply particularly to applications of AI for "sheer" IA (more realistic in the near term), and to enhancement applications, respectively.

Abstract
Sea ice is governed by highly complex, scale-invariant, and anisotropic processes that are challenging to represent in Earth system models. While advanced numerical models have improved our understanding of the sea-ice dynamics, their computational costs often limit their application in ensemble forecasting and climate simulations. Here, we introduce GenSIM, the first generative AI-based pan-Arctic model that predicts the evolution of all relevant key properties, including concentration, thickness, and drift, in a 12-hour window with improved accuracy over deterministic predictions and high computational efficiency, while remaining physically consistent. Trained on a long simulation from a state-of-the-art sea-ice--ocean system, GenSIM robustly reproduces statistics as observed in numerical models and observations, exhibiting brittle-like short-term dynamics while also depicting the long-term sea-ice decline. Driven solely by atmospheric forcings, we attribute GenSIM's emergent extrapolation capabilities to patterns that reflect the long-term impact of the ocean: it seemingly has learned an internal ocean emulator. This ability to infer slowly evolving climate-relevant dynamics from short-term predictions underlines the large potential of generative models to generalise for unseen climates and to encode hidden physics.