Ai Magazine最新文献

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.

For more than a decade, scientists have been striving to make their datasets available in open repositories, with the goal that they be findable, accessible, interoperable, and reusable (FAIR). Although it is hard for most investigators to remember all the “guiding principles” associated with FAIR data, there is one overarching requirement: The data need to be annotated with “rich,” discipline-specific, standardized metadata that can enable third parties to understand who performed the experiment, who or what the subjects were, what the experimental conditions were, and what the results appear to show. Most areas of science lack standards for such metadata and, when such standards exist, it can be difficult for investigators or data curators to apply them. The Center for Expanded Data Annotation and Retrieval (CEDAR) builds technology that enables scientists to encode descriptive metadata standards as templates that enumerate the attributes of different kinds of experiments and that link those attributes to ontologies or value sets that may supply controlled values for those attributes. These metadata templates capture the preferences of groups of investigators regarding how their data should be described and what a third party needs to know to make sense of their datasets. CEDAR templates describing community metadata preferences have been used to standardize metadata for a variety of scientific consortia. They have been used as the basis for data-annotation systems that acquire metadata through Web forms or through spreadsheets, and they can help correct metadata to ensure adherence to standards. Like the declarative knowledge bases that underpinned intelligent systems decades ago, CEDAR templates capture the knowledge of a community of practice in symbolic form, and they allow that knowledge to be applied in a variety of settings. They provide a mechanism for scientific communities to create shared metadata standards and to encode their preferences for the application of those standards, and for deploying those standards in a range of intelligent systems to promote open science.

Will AI replace social scientists? The real issue concerns reshaping rather than replacement. Confronting the integration of large language models (LLMs) into academic training establishes “prompt engineering” as the core interface for collaboration, defining it as a method to translate sociological thinking into precise instructions. LLMs are becoming essential partners across the research spectrum. They transform qualitative analysis from a solitary craft into a dialogical coding process and assist in theoretical localization and the construction of localized measurement scales. Beyond text analysis, they provide a low-cost virtual testbed for experimental design through “silicon samples” and enable the deduction of complex social interactions via “generative agents.” In the quantitative realm, they act as translators connecting research intentions with statistical code. Ultimately, the core challenge facing researchers is not technical. It lies in proactively cultivating a critical “literacy for human-AI collaboration” to master this paradigm shift.

Generative AI's growing use in higher education research requires strong protocols for responsible use. This need arises from the potential for misuse and the current uncertainty around ethical concerns and intellectual property. The lack of clear rules about openness in AI use, along with the “black box” nature of many AI systems, raises worries about reproducibility and the possibility of biased or fake results. This paper focuses specifically on generative AI tools (e.g., LLMs like ChatGPT, research-specific platforms like Elicit/SciSpace). The paper presents the ETHICAL protocol (i.e., Establish your purpose, Thoroughly explore options, Harness the appropriate tool, Inspect and verify output, Cite and reference accurately, Acknowledge AI usage transparently, and Look over publisher's guidelines), a detailed guide designed to direct researchers in the ethical and responsible inclusion of generative AI in their work. The protocol was created through a multi-step process, including a scientometric review of current trends, a systematic review of researcher experiences, and a policy analysis of 74 documents from various stakeholders (authorities, universities, publishers, and publication manuals). This analysis shaped the creation of a seven-heading, nine-item checklist covering key aspects of responsible AI use, from setting clear research goals to checking outputs and openly acknowledging AI help. The ETHICAL protocol gives practical examples and detailed explanations for each item, highlighting the importance of AI literacy and careful choice of suitable tools. It also stresses the vital need for checking AI-generated content to lessen the risk of errors and made-up information (“hallucinations”). The resulting protocol offers a practical and easy-to-use guide for researchers, encouraging responsible AI practices and upholding academic integrity. The ETHICAL protocol offers a helpful tool for managing the complex area of AI in research, ultimately leading to more open, reliable, and ethically sound scholarly work. Its broad acceptance could greatly improve the responsible use of AI in higher education, building trust and furthering knowledge growth.

Proportional representation is a foundational principle in social choice theory, ensuring that groups influence collective decisions in proportion to their size. While it has traditionally been studied in the context of political elections, recent work in computational social choice has broadened its scope to a variety of voting frameworks. This article showcases how proportional representation can be formalized and applied beyond these frameworks, spotlighting AI domains where it naturally takes shape. In particular, we focus on two such domains: clustering and AI alignment. In clustering, proportionality ensures that sufficiently large and cohesive groups of data points or agents are adequately represented in the selection of cluster centers or group assignments, to both centroid-based and noncentroid-based paradigms. In AI alignment, particularly in reinforcement learning from human feedback (RLHF), proportionality provides a principled framework for aggregating heterogeneous preferences by designing committees of reward functions that reflect annotators' viewpoints in proportion to their prevalence. We also discuss additional promising applications, including client selection in federated learning and forming committees of pre-trained models in meta-learning, and argue that incorporating proportional representation into AI systems provides a mathematically rigorous foundation for aligning algorithmic outcomes with the breadth of human viewpoints.

Artificial Intelligence (AI) systems increasingly shape many aspects of daily life, influencing our jobs, finances, healthcare, and online content. This expansion has led to the rise of human–AI systems, where humans communicate, collaborate, or otherwise interact with AI, such as using AI outputs to make decisions. While these systems have shown potential to enhance human capabilities and improve performance on benchmarks, evidence suggests that they often underperform compared to AI-only or human-only approaches in experiments and real-world applications. Here, we argue that human–AI systems should be developed with a greater emphasis on human-centered factors—such as usability, fairness, trust, and user autonomy—within the algorithmic design and evaluation process. We advocate for integrating human-centered principles into AI development through human-centered algorithmic design and contextual evaluation with real users. Drawing on interdisciplinary research and our tutorial at two major AI conferences, we highlight examples and strategies for AI researchers and practitioners to embed these principles effectively. This work offers a systematic synthesis that integrates technical, practical, and ethical insights into a unified framework. Additionally, we highlight critical ethical considerations, including fairness, labor, privacy, and human agency to ensure that systems meet performance goals while serving broader societal interests. Through this work, we aim to inspire the field to embrace a truly human-centered approach to algorithmic design and deployment.

As data continues to grow in volume and complexity across domains such as finance, manufacturing, and healthcare, effective anomaly detection is essential for identifying irregular patterns that may signal critical issues. Recently, foundation models (FMs) have emerged as a powerful tool for advancing anomaly detection. They have demonstrated unprecedented capabilities in enhancing anomaly identification, generating detailed data descriptions, and providing visual explanations. This survey presents the first comprehensive review of recent advancements in FM-based anomaly detection. We propose a novel taxonomy for classifying FMs in anomaly detection tasks, based on the roles they play within the detection pipeline. Specifically, we categorize FMs as encoders, detectors, or interpreters, reflecting whether they are used for feature extraction, anomaly detection, or result interpretation, respectively. We provide a systematic analysis of state-of-the-art methods and discuss key challenges in leveraging FMs for improved anomaly detection. We also outline future research directions in this rapidly evolving field.

Concerns about extinction risk from AI vary among experts in the field. However, AI encompasses a very broad category of algorithms. Perhaps some algorithms would pose an extinction risk, and others would not. Such an observation might be of great interest to both regulators and innovators. This paper argues that advanced imitation learners would likely not cause human extinction. We first present a simple argument to that effect, and then we rebut six different arguments that have been made to the contrary. A common theme of most of these arguments is a story for how a subroutine within an advanced imitation learner could hijack the imitation learner's behavior toward its own ends. However, we argue that each argument is flawed and each story implausible.

Most recent writing about the directions for AI has focused on the potential risks of widespread use of AI and what we DO NOT want from AI. This has led to many, largely ignored, calls for a pause in research and deployment. This essay takes the view that there are too many factors in play to slow the deployment much and for long. Hence, instead, this paper looks at what we DO want from AI (18 principles or goals) and how to get there.

This paper proposes to understand arithmetic operations in Language Models (LM) by framing them as digit-based reasoning challenges. Our research focuses on arithmetic optimization challenges specific to LLMs, not on solving mathematical word problems. We introduce a metric called the Count of Sequential Intermediate Digits (CSID), which measures the complexity of arithmetic equations by counting the missing steps in digit reasoning. Our empirical findings suggest that increasing the model size does little to improve the handling of equations with high CSID values.

We propose RevOrder, a method that incorporates techniques such as reversing the output order, step-by-step decomposition, and rollback mechanisms to maintain a low CSID, thereby enhancing the solvability of arithmetic equations in LMs. RevOrder also introduces a more compact reasoning process, which reduces the token requirements without affecting the CSID, significantly enhancing token efficiency.

Comprehensive testing shows that RevOrder achieves perfect accuracy in operations such as addition, subtraction, and multiplication, and substantially improves performance in division tasks, especially with large numbers where traditional models falter.