关键词: 问答－生成, 认知殖民, 控制社会, 平庸之恶, 伦理风险, 治理建议, 生成式搜索引擎

Abstract: Generative search engines shift the way humans access information from the traditional retrieval-filtering approach to a“Q&A-generation”model. This will not only bring about a transformation in Internet infrastructure and the human cognitive paradigm but also cause users to slide along the cognitive track laid down by algorithms, potentially succumbing to cognitive colonization. This type of cognitive“outsourcing” is likely to lead to“thinking alienation”or even“cognitive alienation.”Generative search engines utilize Large Language Models (LLMs) as mediators to generate answers. However, the inherent defects of LLMs—such as “poisoned”training data and“model hallucinations”—will aggravate information pollution, homogenize information sources, proliferate misinformation, damage the information ecology, and lead to fragmented infringement. Consequently, answers generated by large models are not always objective or trustworthy, and knowledge colonization can, in turn, exacerbate security and ethical risks. To mitigate these risks, beyond establishing higher ethical requirements for enterprises and users, it is essential to build an integrated governance model encompassing four dimensions: technology, institutions, the market, and human values. This will ensure that generative search engines become the Promethean fire of human cognition, rather than a new form of cave captivity.

Key words: Q&A-generation, cognitive colonization, society of control, the evil of mediocrity, ethical risks, governance recommendations, generative search engines