Ai Magazine最新文献

Researchers focusing on how artificial intelligence (AI) methods explain their decisions often discuss controversies and limitations. Some even assert that most publications offer little to no valuable contributions. In this article, we substantiate the claim that explainable AI (XAI) is in trouble by describing and illustrating four problems: the disagreements on the scope of XAI, the lack of definitional cohesion, precision, and adoption, the issues with motivations for XAI research, and limited and inconsistent evaluations. As we delve into their potential underlying sources, our analysis finds these problems seem to originate from AI researchers succumbing to the pitfalls of interdisciplinarity or from insufficient scientific rigor. Analyzing these potential factors, we discuss the literature at times coming across unexplored research questions. Hoping to alleviate existing problems, we make recommendations on precautions against the challenges of interdisciplinarity and propose directions in support of scientific rigor.

The European Union Parliament passed the EU AI Act in 2024, which is an important milestone towards the world's first comprehensive AI law to formally take effect. Although this is a significant achievement, the real work begins with putting these rules into action, a journey filled with challenges and opportunities. This perspective article reviews recent interdisciplinary research aimed at facilitating the implementation of the prohibited AI practices outlined in the EU AI Act. It also explores the necessary future efforts to effectively enforce the banning of those prohibited practices across the EU market and the challenges associated with such enforcement. Addressing these future tasks and challenges calls for the establishment of an interdisciplinary governance framework. This framework may contain a workflow that can identify the necessary expertise and coordinate experts’ collaboration at different stages of AI governance. Additionally, it involves developing and implementing a set of compliance and ethical safeguards to ensure effective management and supervision of AI practices.

The current literature on AI-advised decision making—involving explainable AI systems advising human decision makers—presents a series of inconclusive and confounding results. To synthesize these findings, we propose a simple theory that elucidates the frequent failure of AI explanations to engender appropriate reliance and complementary decision making performance. In contrast to other common desiderata, for example, interpretability or spelling out the AI's reasoning process, we argue that explanations are only useful to the extent that they allow a human decision maker to verify the correctness of the AI's prediction. Prior studies find in many decision making contexts that AI explanations do not facilitate such verification. Moreover, most tasks fundamentally do not allow easy verification, regardless of explanation method, limiting the potential benefit of any type of explanation. We also compare the objective of complementary performance with that of appropriate reliance, decomposing the latter into the notions of outcome-graded and strategy-graded reliance.

Correcting misinformation is a complex task, influenced by various psychological, social, and technical factors. Most research evaluation methods for identifying effective correction approaches tend to rely on either crowdsourcing, questionnaires, lab-based simulations, or hypothetical scenarios. However, the translation of these methods and findings into real-world settings, where individuals willingly and freely disseminate misinformation, remains largely unexplored. Consequently, we lack a comprehensive understanding of how individuals who share misinformation in natural online environments would respond to corrective interventions. In this study, we explore the effectiveness of corrective messaging on 3898 users who shared misinformation on Twitter/X over 2 years. We designed and deployed a bot to automatically identify individuals who share misinformation and subsequently alert them to related fact-checks in various message formats. Our analysis shows that only a small minority of users react positively to the corrective messages, with most users either ignoring them or reacting negatively. Nevertheless, we also found that more active users were proportionally more likely to react positively to corrections and we observed that different message tones made particular user groups more likely to react to the bot.

With their advertising and audience revenues in decline, local news organizations have been experiencing comparatively high degrees of disruption in recent years. Artificial Intelligence (AI) offers opportunities for local news organizations to better cope with the economic challenges they face. However, local news organizations need to carefully prioritize where AI will create the most value. After all, they serve customers in the audience and advertising markets, with external effects on society. At the same time, they are limited by scarce resources, which constrains the implementation of AI. Therefore, based on Porter's value chain, this article pursues two goals. First, drawing on previous research, we provide a systematic overview of activities for which local news organizations see the biggest potential of AI to create value. Moreover, we highlight promising AI use cases based on benchmarking with national news organizations. Second, we discuss local news organizations’ challenges in implementing AI and how they might overcome such obstacles.

The advance of machine learning (ML) systems in real-world scenarios usually expects safe deployment in high-stake applications (e.g., medical diagnosis) for critical decision-making process. To this end, provable robustness of ML is usually required to measure and understand how reliable the deployed ML system is and how trustworthy their predictions can be. Many studies have been done to enhance the robustness in recent years from different angles, such as variance-regularized robust objective functions and conformal prediction (CP) for uncertainty quantification on testing data. Although these tools provably improve the robustness of ML model, there is still an inevitable gap to integrate them into an end-to-end deployment. For example, robust objectives usually require carefully designed optimization algorithms, while CP treats ML models as black boxes. This paper is a brief introduction to our recent research focusing on filling this gap. Specifically, for learning robust objectives, we designed sample-efficient stochastic optimization algorithms that achieves the optimal (or faster compared to existing algorithms) convergence rates. Moreover, for CP-based uncertainty quantification, we established a framework to analyze the expected prediction set size (smaller size means more efficiency) of CP methods in both standard and adversarial settings. This paper elaborates the key challenges and our exploration towards efficient algorithms with details of background methods, notions for robustness measure, concepts of algorithmic efficiency, our proposed algorithms and results. All of them further motivate our future research on risk-aware ML that can be critical for AI–human collaborative systems. The future work mainly targets designing conformal robust objectives and their efficient optimization algorithms.

Personalized news experiences powered by recommender systems permeate our lives and have the potential to influence not only our opinions, but also our decisions. At the same time, the content and viewpoints contained within news recommendations are driven by multiple factors, including both personalization and editorial selection. Explanations could help users gain a better understanding of the factors contributing to the news items selected for them to read. Indeed, recent works show that explanations are essential for users of news recommenders to understand their consumption preferences and set intentions in line with their goals, such as goals for knowledge development and increased diversity of content or viewpoints. We give examples of such works on explanation and interactive interface interventions which have been effective in influencing readers' consumption intentions and behaviors in news recommendations. However, the state-of-the-art in news recommender systems currently fall short in terms of evaluating such interventions in live systems, limiting our ability to measure their true impact on user behavior and opinions. To help understand the true benefit of these interfaces, we therefore call for improving the realism of studies for news.

This article considers the impact of AI on the economy and financing of journalism organizations. AI has structural implications on the news media beyond the practice of journalism and the management of news as a process. AI also shifts the premises of competition, competitive advantage, mergers and acquisitions, and IT capabilities in the news industries. Not least, it fundamentally challenges journalism's traditional business model. Considered hereunder is the two-sided market model, journalism's traditional platform function, its network effects, and its public good characteristics. The aim of the article is, thus, to reconceptualize core economic features of the news industries in the context of AI to provide a vocabulary with which to assess the economic future of journalism in a data-driven platform economy.