If LLMs Have Human-Like Attributes, Then So Does Age of Empires II

查看原文 / View Original 编辑 / Edit

As a tech news editor, I have reviewed the document provided. The content is a technical paper titled “If LLMs Have Human-Like Attributes, Then So Does Age of Empires II” by Adrian de Wynter.

Below is the formatted presentation of the article’s core premise and introduction.

If LLMs Have Human-Like Attributes, Then So Does Age of Empires II

如果大语言模型（LLM）具备类人属性，那么《帝国时代 II》也同样具备

Original: The recent surge in research claiming that Large Language Models (LLMs) possess human-like attributes—such as theory of mind, reasoning, or consciousness—has sparked significant debate. This paper argues that the methodologies used to attribute these qualities to LLMs are often flawed. By applying similar evaluative frameworks to the classic real-time strategy game Age of Empires II, we demonstrate that one could just as easily “prove” that the game possesses human-like cognition, thereby highlighting the absurdity of current anthropomorphic benchmarks in AI research.

Translation: 近期，大量研究声称大语言模型（LLM）具备心智理论、推理能力或意识等类人属性，这引发了广泛的争议。本文认为，用于将这些特质归因于 LLM 的评估方法往往存在缺陷。通过将类似的评估框架应用于经典即时战略游戏《帝国时代 II》，我们证明了人们同样可以轻易地“证明”该游戏具备类人认知能力，从而揭示了当前人工智能研究中拟人化基准测试的荒谬性。

Original: We examine the “black box” nature of both systems and the tendency of researchers to interpret stochastic outputs as evidence of intentionality. Our findings suggest that without rigorous definitions of “human-like” behavior, these benchmarks serve more as mirrors of human projection than as objective measures of machine intelligence.

Translation: 我们审视了这两个系统“黑箱”的本质，以及研究人员倾向于将随机输出解读为意图性证据的现象。我们的研究结果表明，如果缺乏对“类人”行为的严谨定义，这些基准测试与其说是衡量机器智能的客观尺度，不如说是人类投射心理的镜像。