arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~

一种从涌现、集体行动问题和外部性视角探讨人工智能系统风险的方法

An approach to systemic risks of AI through the lens of emergence, collective action problems, and externalities

Carsten Orwat, Lucas Staab, Alexandros Gazos

arXiv 2607.18170首次发表:更新:

AI 中文总结

探讨人工智能系统风险,指出尚无统一定义致概念多样,阻碍风险评估与治理。提出概念化方法,将系统风险主要视为复杂外部性和集体行动问题,关注反馈动态等导致风险的现象,以助理解人工智能系统风险。

AI 中文摘要

通用人工智能模型融入下游人工智能系统引发了比传统人工智能风险更具系统性的新风险形式。然而,对于系统风险,尤其是人工智能的系统风险,尚无普遍接受的定义。这些风险的概念在研究和监管中各不相同。将系统风险方法应用于人权或基本权利相对较新,对人工智能在系统性歧视、侵犯隐私等方面贡献的研究也如此。我们认为目前一些概念未充分考虑复杂性和涌现性,这可能阻碍对人工智能系统风险的评估,导致预防和缓解措施不足及治理无效。为增进对人工智能系统风险的理解,我们提出一种概念化方法,将其主要概述为复杂的外部性和集体行动问题,特别关注反馈动态、导致市场集中的过程、信息不对称等导致系统风险的现象。

英文摘要

The integration of general-purpose artificial intelligence models into downstream AI systems, among other developments, has given rise to new forms of risk that are more systemic in nature than conventional AI risks. However, there is no generally accepted definition of systemic risks in general and for AI in particular. Conceptualisations of these risks vary across research and regulation. Especially the application of the systemic risk approach to human rights or fundamental rights, like in the EU AI Act, is relatively new, just as the research on the contribution of AI to systemic forms of discrimination, privacy violations, erosions of democracy, or climate and environmental degradation. We argue that some concepts so far have not sufficiently take complexity and emergence into account. Furthermore, this variety of concepts might hinder responsible actors to adequately assess the systemic risks of AI, leading to inadequate prevention and mitigation measures and ineffective governance. To contribute to the understanding of systemic risks of AI, we propose a conceptualisation of systemic risks of AI that considers complex phenomena that lead to the emergence of harms at the societal or global level. We outline systemic risks mainly as complex externalities and collective action problems. Of particular interest are feedback dynamics, processes that lead to market concentration like network effects, algorithmic monocultures, and integration processes of AI supply chains or 'AI ecosystems' and across societal sectors, which can result in structural dominance, (inter-) dependencies, and cascading risks. Further phenomena contributing to systemic risks are information asymmetries, informational emergence, and deficits of the governance and institutional framework.

Comments42 pages

论文原文

arXiv 摘要页 · PDF 原文 · HTML 原文

↑