从人机互动到“人际”互动?GPT带出的主体性问题  

From Human-Computer Interaction to Interpersonal Interaction?The Problem of Subjectivity Raised by GPT

作　　者：于海[1] YU Hai(School of Social Development and Public Policy,Fudan University,Shanghai 200433)

机构地区：[1]复旦大学社会发展与公共政策学院,上海200433

出　　处：《文化艺术研究》2023年第3期1-6,111,共7页Studies in Culture and Art

摘　　要：人机对话概念为诺伯特·维纳所发明,只是,到了智能机器能用自然语言跟人类顺畅交流时,维纳的理想才算真正实现。机器是在专家的训练和常人的陪练中学会人类语言的,训练机器同训练青少年一样困难和缓慢。这意味着日益人性化的ChatGPT近似一个类社会化产物。反之,人的社会化或许也是一个生成性预训练过程。人的语言、行为、价值和规范模式不正是接受人类和机构导师无数预先训练而获得,并在不同场景中举一反三才生成合宜的判断和举止?GPT的类社会化性质和社会化的似生成性预训练过程,令人机对话内含人际对话的想象。智能机器或终将无所不知、无所不能,恐会发生反向的社会化,即人被人工智能化。常人会被各种强大的人工智能所降服,这样演化出的“人际”互动势必侵蚀人的主体性。人若不想被机器所淘汰或降维,就别奢想“人际”对话,人机对话足矣!The concept of human-machine dialog was invented by Norbert Wiener.Yet it is only realized when machines could communicate fluently with human beings in natural language.Machines learn human language through training and interactions with the public,a process as difficult and slow as training adolescents.The increasingly human-like ChatGPT approximates a pseudo-social product.Human socialization is also a generative pre-training process.Human speech,behavior,values,and patterns are acquired through the pre-training by human mentors and institutions,enabling appropriate judgments and conduct in different situations.The pseudo-social nature of GPT and the generative pre-training process of human socialization evoke the imagination of interpersonal dialog within human-machine dialog.Intelligent machines may eventually become all-knowing and may lead to a reversed socialization process where humans are influenced by artificial intelligence.AI's impact on human beings will result in an erosion of human subjectivity.If humans do not want to be eliminated or reduced by machines,they should not harbor the illusion of"interpersonal"dialog.Human-machine dialog is sufficient.

关 键 词：人机互动 主体性 类社会化 

分 类 号：G201[文化科学—传播学]