Abstract Agentic AI systems are increasingly capable of autonomous decision‐making and collaboration across distributed environments. However, existing frameworks largely assume trust among agents, without providing mechanisms for verifiable identity, behavioral accountability, or interaction transparency. This work proposes a blockchain‐based backbone infrastructure that enables a verifiable trust layer for autonomous multi‐agent systems. Agent identities are bound to decentralized identifiers (DIDs) registered on‐chain, while an on‐chain Agent Registry enables reputation‐aware capability discovery. By integrating smart contracts, DIDs, and verifiable credentials (VCs), the framework ensures that trust is not assumed but earned through interactions: agents issue VC endorsements after each task, and dedicated smart contracts automatically update reputation scores, applying incentives and penalties accordingly. Blockchain provides immutability, transparency, and programmability, transforming trust from an implicit assumption into an explicit, auditable, and enforceable property of the multi‐agent ecosystem.
Abstract Through advances in computing power, sensor systems, networking, and AI programming, new automated systems are being developed with capabilities for acting in the world that heretofore only people could do—driving automobiles, controlling aircraft systems, delivering packages. As we give computer programs control of complicated vehicles and devices, we need new tools to create and analyze designs systemically, to anticipate and understand how people and machines will behave and interact in safety‐critical situations—when action may be urgent and lives at stake. To secure the trust of consumers and certification by regulators, engineers need to adopt scientific design methods to verify that the behavior of instruments, devices, and programs fits how people perceive, reason, and act in challenging situations.
Abstract This article examines the structural reconfiguration of contemporary music‐making within AI‐mediated production environments. From a historical perspective of music technology, it compares algorithmic composition, electronic instruments, and Digital audio workstations (DAWs) to argue that the present transition represents not a rupture but a compression of musical practice. As generation, transformation, evaluation, and distribution converge within an AI‐native stack, the intermediate layer of musical work is increasingly absorbed into system‐level processes. Judgments that once appeared as externally visible artifacts, such as sketches, edit histories, chord charts, and revision processes, are now executed within integrated computational environments. The so‐called “missing middle” therefore signals not disappearance but a relocation of musical decision making. Prompt‐based creation is reframed as a shift from writing to direction, and as iteration becomes inexpensive, revision moves from stepwise rewriting toward regeneration‐centered cycles. In response, this article proposes the reconstruction of a “constructive middle” suited to AI‐native production, redefining controllability and revisability as core design parameters of AI‐mediated creative systems. In doing so, it articulates the design and governance conditions under which controllable and accountable generative music systems can sustain creative depth within accelerated production infrastructures. For AI leaders and decision‐makers, this framework identifies four priorities beyond output quality and speed: inspectable intermediate states, constrained local editing, portable provenance, and workforce capabilities for directing and auditing generative systems.
Reproducibility is a foundational tenet of science. As artificial intelligence (AI) becomes increasingly embedded across science, the need to accurately document the provenance, structure, and behavior of training data, models, and workflows grows correspondingly. Metadata, understood as explicit and structured knowledge about data and related entities, is a critical yet often underexamined component of AI systems that helps address this need. High-quality metadata describing datasets, models, and workflows supports the FAIR (Findable, Accessible, Interoperable, Reusable) principles, strengthens reproducibility, and enables evaluation of AI-readiness by making data and models interpretable, traceable, and structurally consistent. Despite its central importance, sustained discussion of metadata as a core component of AI infrastructure remains limited. This article examines the role of metadata in advancing AI-enabled research, focusing on the design, implementation, and operationalization of metadata standards within an evolving metadata ecosystem. We first present a conceptual view of the metadata ecosystem, framed by data structure, data value, data encoding, and syntax standards, as a foundation for understanding how metadata enables AI and how AI contributes to metadata generation and refinement. We then introduce four case studies that illustrate how metadata can be generated, refined, and leveraged within AI workflows. The discussion synthesizes the cases, highlights limitations, including metadata quality challenges and the role of structured constraints in addressing AI errors, and relates each case to the metadata ecosystem dimensions it engages. Taken together, these cases illustrate that metadata is not a peripheral add-on but an essential component of AI-ready, FAIR-aligned, transparent, and reproducible research systems.
Crafting cartoon captions requires an understanding of humor, context, and the relationship between image and text. Traditionally, illustrators and writers collaborate to strengthen visual storytelling and comedic timing. With advances in natural language generation, large language models (LLMs) can assist in this process. This study examines AI's role in caption generation by testing GPT-4o via the ChatGPT interface on IEEE Computer magazine cartoons. By removing captions and prompting AI to generate replacements, we assess its ability to produce jokes that match the depicted situation and narrative intent. Our findings show that while AI-generated captions are often humorous and contextually relevant, they sometimes diverge from the cartoon's intended meaning, for example, by missing irony, cultural references, or contextual constraints. However, AI can also produce alternatives that broaden creative exploration and occasionally improve upon the original humor. We argue that current AI systems are best used as an assistant rather than a replacement for human creativity. By integrating AI-generated suggestions, cartoonists can explore diverse humor styles, streamline ideation, and refine final captions while retaining creative control. This study highlights AI's potential as a practical tool for caption ideation within a hybrid human-AI workflow.
The exponential growth of scientific literature-over 2.5 million papers and 3.5 million patents annually-poses critical challenges for knowledge discovery. To address these, we propose SCIMKG (Scientific Knowledge Graph), a framework for constructing scientific knowledge graphs via multi-source heterogeneous multi-source. To address these challenges, we propose SCIMKG (Scientific Knowledge Graph), a framework for constructing scientific knowledge graphs through multi-source heterogeneous multi-source. SCIMKG integrates diverse academic resources including research papers, patents, experimental datasets, videos, charts, and tables to automate the construction of domain-specific knowledge graphs. Our framework introduces three key innovations: (1) a multi-source entity fusion mechanism employing Graph Attention Networks (GATs) with composite similarity metrics for cross-source alignment; (2) a dynamic hybrid verification pipeline integrating semantic, statistical, and rule-based methods; and (3) a Domain-Adaptive Relation Extraction model (DA-RE) utilizing adversarial training and prototype networks. Evaluations on SciBench, a multi-domain benchmark comprising 8.2 million academic resources from computer science, biomedical engineering, and materials science, show that SCIMKG achieves 82.3% entity recognition F1-score and 86.2% relation extraction accuracy, outperforming existing methods by 3.4 and 12.8 percentage points respectively. The constructed knowledge graph contains 14.3 million high-confidence triples and has been deployed on a national science platform, supporting intelligent search and trend analysis. A user satisfaction survey (N = 1,247 respondents, 95% CI [90.3%, 93.7%]) conducted across 12 institutions showed 92% satisfaction with system usability and search relevance.
Modern AI systems have exceeded human performance on many benchmarks meant to evaluate general cognitive capacities. However, it is often the case that benchmark performance does a poor job of predicting general capacities in real-world settings. In this article I describe several issues related to evaluation that can cause this mismatch, and propose six principles, inspired by developmental and comparative psychology, that need to be adopted to enable rigorous evaluation for AI systems. These principles are illustrated by case studies from the psychology and AI literature.
Students frequently make mistakes while solving mathematical problems, and traditional error correction methods are both time-consuming and labor-intensive. This paper introduces an innovative Virtual AI Teacher system (VATE) designed to autonomously analyze and correct student Errors. Leveraging advanced large language models (LLMs), the system utilizes student drafts as a primary source for error analysis, thereby enhancing the understanding of the student's learning process. It incorporates sophisticated prompt engineering and maintains an error pool to reduce computational overhead. The AI-driven system also features a real-time dialogue component for efficient student interaction. Our approach demonstrates significant advantages over traditional and machine learning-based error correction methods, including reduced educational costs, high scalability, and superior generalizability. The system has been deployed on the Squirrel AI learning platform for elementary mathematics education, where it achieves 78.3 accuracy in error analysis and shows a marked improvement in student learning efficiency. Satisfaction surveys indicate a strong positive reception, highlighting the system's potential to transform educational practices.
AI systems are rapidly transitioning from laboratory demonstrations to decision-making technologies deployed in high-stakes domains. Yet reliability remains a primary obstacle to responsible adoption: discriminative models can be confidently wrong under out-of-distribution (OOD) inputs, and foundation models (FMs) such as large language models (LLMs) can generate fluent but untruthful, harmful, or misaligned outputs. My research develops the foundations of reliable machine learning with minimal human supervision, unifying algorithms, and theory that make reliability a first-class objective alongside accuracy. I advance unknown-aware learning through automated outlier generation, introducing feature- and input-space synthesis frameworks that regularize decision boundaries and improve interpretability. I further establish principled methods for learning "in the wild" by leveraging unlabeled deployment data under mixture and contamination models, with theoretical guarantees and state-of-the-art performance for OOD detection and generalization under diverse shifts. Finally, I design reliability frameworks for FMs by exploiting unlabeled signals to detect hallucinations, defend against malicious prompts in vision-language models, and denoise noisy preference data for more dependable alignment. Collectively, these contributions provide a cohesive toolkit for deploying AI systems that remain accurate, calibrated, and trustworthy in open-world environments.
This study explores the evolving representation of Artificial Intelligence (AI) characters in media and its intersection with contemporary technological issues, focusing on the paradoxical human desire for emotional and creative replacement. By analyzing films such as the 2023 production M3GAN and the 2025 production Cassandra, the article examines how fictional AI has transitioned to substituted humans that occupy intimate familial and emotional roles. Furthermore, the research investigates AI arts through AI-generated films like 2018 production Zone Out and the recent AI-generated images trend on social media. It argues that while AI offers unprecedented efficiency and aesthetic refinement, it poses significant ethical challenges regarding digital sovereignty, copyright, and the commodification of human identity. Hence, the research emphasizes that the discourse on AI should shift from the technical prowess of mimicry to a normative re-evaluation of human agency, advocating for a creative paradigm that prioritizes the preservation of irreplaceable human values against the tide of unconscious substitution.
This article examines how data readiness for AI principles apply to large scientific datasets used to train foundation models. We analyze archetypal workflows across four representative domains-climate, nuclear fusion, life sciences, and materials-to identify common preprocessing patterns and domain-specific constraints. We introduce a two-dimensional readiness model that combines canonical preprocessing patterns with a five-level operational readiness scale, both tailored to high-performance computing (HPC) environments. This construct helps outline key challenges in transforming large-scale scientific data into formats suitable for scalable AI training. Together, these dimensions form a conceptual maturity matrix that characterizes scientific data readiness and guides infrastructure development toward standardized, cross-domain support for scalable and reproducible AI for science. Finally, we evaluate this maturity matrix in the context of case studies including ClimaX (climate), AFLOW (materials), OpenFold (proteomics), and DIII-D fusion disruption-prediction workflows, from which we distill lessons learned and provide recommendations to guide practitioners in developing robust AI-readiness pipelines. Finally, we discuss remaining cross-cutting challenges that persist across scientific domains.
This article investigates the impact of distribution shifts in trustworthy machine learning. To this end, we start by summarizing fine-grained types of distribution shifts commonly studied in machine learning communities. To tackle distribution shifts across domains, we present our research across various learning scenarios by enforcing knowledge transferability and trustworthiness. Specifically, we focus on two learning paradigms to improve knowledge transferability: distribution-informed representation learning and distribution-guided information propagation. Besides, we also explore how trustworthiness properties of a learning algorithm are affected by distribution shifts across domains. Finally, we discuss the open questions and future directions for handling distribution shifts in the era of large language models.
Data is the foundation of AI. Poor-quality data drive up costs and can lead to hidden problems for AI models, especially in complex fields such as healthcare and manufacturing. Meanwhile, biased data negatively affect the performance of AI models, and untested evaluation datasets can result in false positives or overestimates of model accuracy. For data publishers to realize their true potential in supporting the AI ecosystem and its impacts, they should take measures to ensure that their datasets support AI practitioners' needs; in other words, their data should be made AI-ready. In this article, we present a framework for data publishers to follow to make their datasets AI-ready. The framework provides specific, actionable guidance based on previous work and experience at the Open Data Institute and augmented with insights from literature and discussions with a range of experts. We first define AI-ready data before briefly discussing a selection of frameworks in the literature and where they are insufficient. We then provide a visual snapshot of our framework for AI-ready data, and a subsequent in-depth discussion of its criteria. Finally, we demonstrate the usage of our framework with a number of example datasets. We conclude by discussing the further steps that should be taken for the entire open data ecosystem to be made AI-ready in order to realize its true potential in supporting an innovative future.
The rapid proliferation of smart-city ecosystems has significantly amplified the demand for Li-ion batteries, which now serve as the primary energy source for sustainable transportation systems such as e-bikes. Ensuring battery safety and optimal performance is crucial, yet challenging due to complex intrinsic dynamics and extrinsic operating conditions. This paper presents LiBrain, an innovative LLM-powered, time-series-aware retrieval-augmented framework designed to simultaneously address both safety and performance challenges through three synergistic components: (1) a distributed IoT-enabled edge network for continuous real-time battery monitoring and data acquisition, (2) a pretrained deep multi-task diagnostic engine capable of comprehensive battery performance forecasting, and (3) a knowledge-base augmentation module that transforms technical diagnostics into clear, actionable guidance tailored for e-bike users. Functioning as an intelligent battery management assistant, LiBrain effectively bridges the gap between expert-level real-time analytics and practical, user-friendly instructions. Extensive validation across a real-world operational e-bike battery-swap network demonstrates LiBrain's exceptional capabilities, achieving a 95% adoption rate in hazardous alarm detection and 92% in battery-status prediction. In real application, Li-Brain has processed over 500 million battery events, managed almost 10 million inquiries and 1 million alarms annually, and identified 10% of on-site batteries daily for proactive replacement, thereby maintaining operational safety and reliability.
Electric bicycles (e-bikes) have become the dominant mode of transportation in China’s urban instant delivery industry. However, many riders lack the experience to navigate complex traffic networks and diverse road conditions, leading to reduced delivery efficiency. To address this issue, we present Talking Trails, an e-bike delivery route planning system built upon an LLM-enhanced spatiotemporal trajectory model. Trained on millions of real-world delivery trajectories, fused with spatiotemporal and semantic data information, the model achieves a top-5 rider displacement prediction accuracy of 95% and a route optimization rate of 82.1%. In practice, we augment the core planner with an LLM-driven semantic layer that translates high-level user intent into executable tasks, then pair it with a battery-swap module that continuously validates route feasibility so the vehicle never runs out of charge mid-mission. Currently serving tens of thousands of riders, the system is projected to reduce average delivery mileage by 17% and lower annual carbon emissions by 3978 tons. Overall, Talking Trails significantly improves delivery efficiency, offering a scalable and sustainable solution for instant delivery operations.
As a field, web development is roughly 30 years old, and during this period, it has been transformed several times already as it has moved from static websites to dynamic web applications. Now, with the introduction of Artificial Intelligence (AI), the field is again at the cusp of a transformation as the latest AI tools might change how to develop for the web yet again. The objective of this study is to look into this phenomenon and understand how AI is changing web development. To achieve this task, we chose to use the sequential qualitative-quantitative design method that combines interviews with a survey to validate and expand our findings from the interviews. We found that AI is used by web developers to increase their development efficiency, as even the current tools are easy to use and access, although they come with several minor downsides, including AI not being able to understand complex logic, the need for validation of AI output, and suggested code that could potentially lead to security issues. While there are clear benefits to using AI tools for web development and AI proficiency is a vital skill for web developers, there are still open questions related to the quality of code produced by AI tools.
This study explores the role of artificial intelligence (AI) in perception management as an emerging tool of political soft power. Drawing on the theoretical frameworks of social psychology, strategic communication, and political communication, the research investigates how AI-assisted strategies influence public perception, image, and trust in the context of modern statecraft. The study adopts a qualitative design based on semi-structured interviews with 16 experts-eight from psychology and eight from communication fields-selected through snowball sampling. Data were analyzed using qualitative content analysis to identify recurring patterns and thematic structures. The findings reveal four central themes: (1) AI enhances efficiency and precision in perception campaigns, (2) trust and credibility remain critical yet vulnerable dimensions, (3) ethical and governance dilemmas emerge in AI-mediated communication, and (4) human oversight continues to be essential for maintaining legitimacy. The results suggest that while AI strengthens states' capacity for strategic influence, overreliance without transparency may undermine the very trust it seeks to build. The study contributes to soft power and communication scholarship by providing expert-based evidence on the psychological and strategic mechanisms of AI-driven perception management. Policy recommendations are offered to promote transparency, accountability, and ethical oversight in AI-enabled diplomatic practices.
Generalist robot models promise broad applicability across domains but currently require extensive expert demonstrations for task specialization, which is a costly and impractical barrier for real-world deployment. In this article, which summarizes the author's presentation in the New Faculty Highlights Track of the 39(th) annual AAAI Conference on Artificial Intelligence, we present algorithms that enable non-expert users to adapt and continually improve robot policies through natural and lightweight feedback modalities, such as preference comparisons, rankings, ratings, natural language, and users' own demonstrations, combining them with active learning strategies to maximize data-efficiency. We further introduce methods for leveraging real-time human interventions as rich training signals, modeling both their timing and absence to refine policies continually. Our approaches achieve substantial gains in sample-efficiency, adaptability, and user-friendliness, demonstrated across simulated and real-world robotic tasks. By aligning robot learning with how humans naturally teach, we hope to move toward autonomous systems that are more personalized, capable, and deployable in everyday environments.
Machine learning (ML) is becoming prevalent in embedded AI sensing systems. These "ML sensors" enable context-sensitive, real-time data collection and decision-making across diverse applications ranging from anomaly detection in industrial settings to wildlife tracking for conservation efforts. As such, there is a need to provide transparency in the operation of such ML-enabled sensing systems through comprehensive documentation. This is needed to enable their reproducibility, to address new compliance and auditing regimes mandated in regulation and industry-specific policy, and to verify and validate the responsible nature of their operation. To address this gap, we introduce the datasheet for ML sensors framework. We provide a comprehensive template, collaboratively developed in academia-industry partnerships, that captures the distinct attributes of ML sensors, including hardware specifications, ML model and dataset characteristics, end-to-end performance metrics, and environmental impacts. Our framework addresses the continuous streaming nature of sensor data, real-time processing requirements, and embeds benchmarking methodologies that reflect real-world deployment conditions, ensuring practical viability. Aligned with the FAIR principles (Findability, Accessibility, Interoperability, and Reusability), our approach enhances the transparency and reusability of ML sensor documentation across academic, industrial, and regulatory domains. To show the application of our approach, we present two datasheets: the first for an open-source ML sensor designed in-house and the second for a commercial ML sensor developed by industry collaborators, both performing computer vision-based person detection.
Many research fields are currently reckoning with issues of poor levels of reproducibility. Some label it a “crisis,” and research employing or building machine learning (ML) models is no exception. Issues including lack of transparency, data or code, poor adherence to standards, and the sensitivity of ML training conditions mean that many papers are not even reproducible in principle. Where they are, though, reproducibility experiments have found worryingly low degrees of similarity with original results. Despite previous appeals from ML researchers on this topic and various initiatives from conference reproducibility tracks to the ACM's new Emerging Interest Group on Reproducibility and Replicability, we contend that the general community continues to take this issue too lightly. Poor reproducibility threatens trust in and integrity of research results. Therefore, in this article, we lay out a new perspective on the key barriers and drivers (both procedural and technical) to increased reproducibility at various levels (methods, code, data, and experiments). We then map the drivers to the barriers to give concrete advice for strategies for researchers to mitigate reproducibility issues in their own work, to lay out key areas where further research is needed in specific areas, and to further ignite discussion on the threat presented by these urgent issues.
