AuraTracer智迹闻
中文

EVENT DOSSIER

When Does an Interpretation Count as Established? The Formation, Evaluation, and Responsibility of Interpretation in Generative AI

2026-09-07 12:00 Science 🔥 42.2 heat score
1sources
1days unfolding
42.2heat score
0mentions
SummaryAI generated

Regarding the issue of defining the validity of explanations in generative AI, relevant research suggests that five common requirements must be met to distinguish between local pass and research-based boundaries. Core concepts include: the appearance of explanations (the gap between the finished product and the traceable process), evaluation contracts (the boundary conditions for determining local validity), and alternative standing (the act of transforming inadequate local passes into stronger claims). Research finds that although texts may be recognized, they are insufficient as sufficient evidence if they lack public justification statements,反驳 mechanisms, and withdrawal structures. To address this, the study introduces the practice of “delayed closure,” emphasizing that mere local evaluation, finished product form, and public recognition cannot provide sufficient proof of the formation of explanations. This analysis is conceptual and normative in nature; it does not set a benchmark for model understanding or determine whether it possesses understanding capabilities.

Related eventsRELATED EVENTS

All reports (1)SOURCES

A arXiv cs.AI en 2026-09-07 12:00

When Does an Interpretation Count as Established? The Formation, Evaluation, and Responsibility of Interpretation in Generative AI

本文提出“解释确立”需满足五项公共要求,以区分局部通过与研究能力确立的界限。文章引入三个核心概念:解释外观指成品与可追溯过程间的差距;评估契约界定局部判断有效的边界条件;替代性站立指将局部通过不当转化为更强主张的行为。研究指出,文本可能获得认可却缺乏公开理由陈述、反驳及撤回的结构,人文社科提供了验证标准可变性的测试。为此,本文提出“延迟闭合”实践,强调局部评估、成品形式与公共认可不足以为解释形成提供充分证据。该论证属概念性与规范性分析,不设定模型理解基准或判定其是否具备理解能力。