相较于传统的数字技术，生成式人工智能呈现出全新的技术样貌，并催生出具有非理性特征的“自反式信息茧房”。传统信息茧房依循“回音室—信息茧房—群体极化”的生成路径，而自反式信息茧房遵循“技术放大器—沉浸逆向—长尾效应”的构成逻辑，后者同时针对用户与生成式人工智能模型自身，是传统信息茧房的强化形态。它基于“沉浸性”“非理性”等特征而蕴含更大的潜在风险，突出表现为个体过度沉浸、群体高度分化、社会空间失序。对此，应分别以主流文化引领用户端的理性化发展、以市场规制修正模型端的资本化逻辑、以规范体系保障“人机交互”的秩序化图景，从而有效应对“自反式信息茧房”在个体、群体、社会等层面潜藏的挑战。

生成式人工智能, 自反式信息茧房, 虚拟空间治理, 互联网, 人机交互

Compared with traditional digital technology, generative AI exhibits a novel technological form, giving rise to the “reflexive information cocoon” characterized by irrationality. The traditional information cocoon follows the generation path of “echo chamber - information cocoon - group polarization”, whereas the reflexive information cocoon adheres to the logic of “technical amplifier - immersive reversal-long-tail effect”. It represents an intensified version of the traditional information cocoon, impacting both users and generative AI models. Rooted in the characteristics of “immersion” and “irrationality”, it poses greater potential risks, notably individual over-immersion, heightened group differentiation, and social spatial disorder. To address these issues, it is imperative to guide the rational development of users with mainstream culture, rectify the capitalization logic of models through market regulations, and safeguard the orderly landscape of “human-computer interaction” with a normative system. This approach will effectively tackle the risks and challenges inherent in the “reflexive information cocoon” at the individual, group, and social levels.

generative artificial intelligence, reflexive information cocoon, virtual space governance, internet,  human-computer interaction