从语言出发:我们离超级智能还有多远OACHSSCD
Starting from Language:How Far Are We from Superintelligence?
超级智能的定义本身随技术进步而动态演进,可能永远无法完全实现.语言作为人类思想的主要载体,极大地推动了智能化进程,使大语言模型成为当前最接近超级智能的技术路径.基于语言的人工智能虽能趋近人类的语言智能,却难以突破其能力边界.语言智能自身存在三方面局限:有限性,即无法完整表达连续、细粒度的感知经验;无现象性,即缺乏主观体验与感受质;无自主性,即不具备内在动机与价值判断.通过进一步分析感知建模、具身交互、价值驱动等语言外技术路径的优势与不足,指出未来超级智能需突破语言框架,构建多层次的认知体系,从而为相关研究提供参考.
The definition of superintelligence itself evolves dynamically with technological progress and may never be fully realized.As the primary vehicle for human thought,language has significantly propelled the advancement of intelligence,making large language models the current technological pathway closest to superintelligence.While language-based artificial intelligence can approximate human linguistic intelligence,it struggles to transcend its inherent limitations.Language intelligence itself faces three inherent limitations:finiteness—the inability to fully express continuous,fine-grained perceptual experiences;non-phenomenalism—the absence of subjective experience and qualia;and non-autonomy—the lack of intrinsic motivation and value judgment.By further analyzing the strengths and weaknesses of non-linguistic technical pathways such as perceptual modeling,embodied interaction,and value-driven approaches,this paper argues that future superintelligence must transcend the linguistic framework to construct a multi-layered cognitive system,thereby providing reference for related research.
胡雨韦;龚新奇
中国人民大学高瓴人工智能学院中国人民大学||中国人民大学国家治理大数据和人工智能创新平台||中国人民大学国家发展与战略研究院||中国人民大学交叉科学研究院
超级智能语言智能大语言模型分布语义学现象学
《阅江学刊》 2026 (2)
90-95,6
评论