📄 中文摘要
AI 的“幻觉”现象并不是机器的缺陷，而是与人类的认知过程相似。人类的记忆是重建性的，而非简单的回放。认知科学称之为“建构性情节记忆”，这种现象是为了应对信息不完整而产生的自动填补、模式匹配、未来模拟和意义构建等过程。生存需要这种能力，因为等待完整信息可能导致瘫痪。大型语言模型（LLMs）在处理信息时也会表现出类似的特征，通过预测和重组信息来生成输出。
AI 幻觉：机器为何会出错（就像我们一样）
出处: AI Hallucinations: Why Machines Get It Wrong (Like We Do)
发布: 2026年2月26日
📄 English Summary
AI Hallucinations: Why Machines Get It Wrong (Like We Do)
The phenomenon of AI 'hallucinations' is not a flaw of machines but rather similar to human cognitive processes. Human memory is reconstructive rather than merely replaying events, a concept known as 'constructive episodic memory' in cognitive science. This phenomenon arises from automatic gap-filling, pattern-matching, future simulation, and meaning-making, all of which are essential for survival, as waiting for complete information can lead to paralysis. Large language models (LLMs) exhibit similar traits by predicting and recombining information to generate outputs.