国家预印本平台

When accessing the learning effect in statistical learning, participants are requested to distinguish target words from partwords or nonwords in a two-alternative forced-choice task. However, this task did not answer how individuals represent target words and foils, and thus may not be sufficient in providing an independent learning effect on the items. The current study examined the explicit learning effect for each word type with a familiarity rating task. Participants were randomly assigned to learn a continuous artificial speech in one of three conditions: a baseline, short-exposure time, or long-exposure time condition. The ratings scores and correlations across three types of words between the baseline condition and the other two learning conditions were examined. Results revealed a significant contrasting learning effect: familiarity ratings for target words were significantly higher than baseline, whereas foils’ ratings were significantly lower, reflecting the explicit knowledge during the exposure phase and the metacognition during the testing process. Furthermore, the distribution of target words’ rating scores tends to be more centralized in the long-exposure time condition, suggesting a new type of SL effect. This study is the first to explore explicit learning effects across word types, and provides insight regarding how to measure SL more exactly.

各类人工智能系统正作为代理广泛嵌入企业、政府与个人的决策流程，对人类决策的道德性和道德判断产生了深刻影响。尽管相关实证和理论研究快速增长，现有文献尚缺乏对人工智能代理区别于其他代理特殊性的系统性分析，也缺乏统一的分析框架来系统刻画人工智能代理对道德决策的影响路径。因此本文构建了“决策者—代理—反馈者”的决策与归责框架，对现有研究进行梳理和重组。本文认为：代理介入决策后，拉长了决策者的决策链条和反馈者(包含受影响者和第三方观察者)的反馈链条，由此削弱了决策者的道德感知和反馈者的归责，进而促进了决策者的不道德行为。而人工智能代理的黑箱性、高遵从、规模化、工具性等特点则在决策链上加剧了不道德指令的执行、增强了决策者的可否认机会、扩大了不道德行为的影响范围；同时，这些特性在反馈链上增加了反馈者对不道德行为的道德容忍、模糊了反馈者对决策者意图的判断与归责，进而促进决策者做出不道德行为。本文指出未来研究有必要继续完善该决策框架内各机制相对作用关系，考察在组织与社会层面道德行为的扩散与放大机制，并探索人机协作情境下的治理工具与制度安排。

本研究以自尊和自悯为靶点，探索大语言模型(LLM)在干预网络欺凌受害者心理韧性中的效果与作用路径。研究1(干预效果检验)采用随机对照设计，比较自尊-自悯对话、心理教育对话和心理教育阅读三组的干预效果。结果显示两组对话式LLM均显著提升了心理韧性。研究2(作用路径分析)进一步采用网络干预分析揭示干预作用路径，发现自尊-自悯对话干预除了可以直接提高心理韧性外，还可以靶向孤立感、正念和社交维度自尊，通过自尊自悯多维度的相互作用，提升干预效果。本研究首次聚焦自尊与自悯路径，基于LLM构建具备共情交互能力的对话干预平台，为网络欺凌受害者提供个性化心理支持，并为智能化心理干预工具的开发提供理论依据与实践路径。

抑郁呈现一个核心悖论。从基因到社会压力，极端异质的触发因素收敛于高度协调的表型组合。这种模式跨越物种、保守数亿年，且找不到一致的结构性损伤证据。这究竟是系统故障，还是被激活的功能性程序？为解决这一悖论，我们构建了程序特征指数PFI作为黑箱推断工具，通过输入异质性、输出协调性、跨物种保守性三个维度量化程序性证据支持强度，并用可退出性进行独立验证。对56种生命现象的计算显示，抑郁落入高PFI组，与睡眠、呕吐等公认的程序并列。如果抑郁是程序，又为何呈现病理状态？本研究提出抑郁是系统能效比持续下降后触发的程序性反应，在演化历史中具有特定功能，但在现代环境中常常错误激活，或陷入难以退出的病理状态。本文是系列研究的第一部分，第二部分将提出能效假说及神经实现，第三部分将论述古代起源和现代失配，系统回答这一问题。

Projection, Irreversible Time, and the limits of process recoverability. This work examines driven, irreversible systems under finite observational projection, focusing on the structural consequences that arise when processes cannot be fully resolved across time. Rather than constructing unified causal explanations, the analysis treats time irreversibility and projection as non-negotiable constraints and examines how system behavior becomes fragmented across distinct visibility conditions. Within this framework, three non-interchangeable observational regimes emerge-instantaneous distinguishability, transitional progression, and deposited structure-linked only through residual formation and constrained coupling. A set of recurring real-world phenomena-including aviation system failures, large-scale auroral redistribution, viral emergence patterns, and organizational-capital dynamics-are examined solely in terms of how they appear under these projection conditions, without cross-scale reconstruction or causal closure. The work does not seek to resolve these phenomena into explanatory chains, but to show that apparent failure, persistence, or control cannot be meaningfully interpreted without respecting irreversible time, finite projection, and the non-compressibility of living systems.