人工智能大模型表征方式的一致性及其哲学启示  

The Consistency of Representation and Its Philosophical Enlightenment of Artificial Intelligence Large Models

作　　者：涂良川 Tu Liangchuan

机构地区：[1]华南师范大学马克思主义学院

出　　处：《学术前沿》2024年第14期45-54,共10页Frontiers

基　　金：国家社会科学一般项目“马克思主义哲学视域中的人工智能奇点论研究”的阶段性成果,项目编号:21BZX002。

摘　　要：以大数据、强算力、多模态和高维度等训练出来的人工智能大模型愈发智能,体现出类人的“聪明”。基于系统稳定性、功能有效性和优化可能性要求,大模型将注意力机制嵌入系统之中,使基于不同数据训练出来的不同大模型在处理数据时体现出表征收敛的趋向。大模型的表征收敛,一方面,显示出基于神经网络的深度学习具有实现通用人工智能的技术潜质;另一方面,也印证了大数据挖掘、大模型超越、强算力迭代和高维度透视所形成的智能具有类人性。因此,虽然人工智能大模型的表征收敛是人工智能智能性的技术体现,本质上却是以人类本质力量对象化的方式考问智能本质的哲学追问。与其说是大模型试图表示现实模型的稳定性推动了系统的表征收敛,倒不如说是大模型以“挖掘即认知”“学习获智能”“高维达简洁”对观测的经验升华构成了表征收敛的智能动因。Large models of AI trained with big data,strong computing power,multimodality and high dimensionality are becoming more and more intelligent,reflecting human-like"smartness".Based on the requirements of system stability,functional effectiveness and optimisation possibilities,large models embed the attention mechanism into the system,so that different large models trained based on diferent data reflect the tendency of representational convergence when processing data.Representational convergence of the large models,on the one hand,shows that the deep learning based on neural networks has the technical potential to realise general artificial intelligence,and on the other hand,it also confirms that the intelligence formed by big data mining,large models transcendence,strong computing power iteration and high-dimensional perspectives has a human-like nature.Thus,while representational convergence of large models of Al is a technical embodiment of artificial intelligence,it is essentially a philosophical inquiry that quizzes the nature of intelligence in the form of an objectification of the essential power of humanity.It is not so much the stability of the large model that attempts to represent reality that drives the representational convergence of the system,Rather,it is the experiential sublimation of observation by large models with"mining as cognition","learning to gain intelligence"and"high-dimensional simplicity",which constitutes the intelligent motivation for representational convergence.

关 键 词：表征收敛 哲学叙事 大模型 人工智能 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]