优化:人工智能时代的文论问题  

Optimization:Issues of Literary Theories in the Era of Artificial Intelligence

作　　者：曾军[1] ZENG Jun(School of Literature,Shanghai University,Shanghai 200444)

机构地区：[1]上海大学文学院,上海200444

出　　处：《文化艺术研究》2024年第2期1-13,112,共14页Studies in Culture and Art

基　　金：国家社会科学基金重点项目“当代中国对话主义文学理论的话语建构研究”(项目编号:22AZW003);上海市教委科研创新计划(人文社科重大项目)“新媒体艺术理论基本问题研究”(项目编号:2023SKZD15)的阶段性成果。

摘　　要：日常用语中的“优化”指采取一定措施使事物变得优异。作为计算科学和运筹学中的重要内容,优化是指在一定限制条件下,选取某种方案以获得“最优化”的目标。将优化引入文论,可以与美化相关联,彼此之间有同有异。生成式AI中,优化有模型训练、创意提示、调试迭代三种方式。文艺作品之所以能够被优化,与其作为物质基础的艺术媒介及其作品的“可修正性”有关。在优化的技术演进中,先后经历了手工加工时代的优化、机械复制时代的优化、数字创生时代的优化三个阶段。手工加工时代出现了诸如对“手稿”的“润色、修改”,对“乐器”的“调音”,对“原作”的“临摹、誊写甚至重写”,对“文物、遗迹”的“修复”等不同的“类优化”类型;机械复制时代出现了“参数优化”“剪辑优化”“表演优化”等类型;数字创生时代出现了以算法优化为核心的智能优化。将优化引入文论,需要警惕“最优化”的结果不是“奇异性”,而是“普通性”的算法悖论、数据和变量中的偏差不易被纠偏;相反,易强化成偏见的偏差悖论以及随机并不意味着不同,而只是相似,是风格的重复的生成悖论。The term"optimization"denotes the process of enhancing things through specific measures.In the realms of computer science and operations research,optimization involves selecting the most effective method within defined constraints to achieve the desired outcome.Introducing optimization into literary theory can be compared to"beautification",with both similarities and distinctions between the two concepts.Within generative AI,optimization manifests in three forms:training,prompts,and iterations.The optimization of literary and artistic works hinges on their material mediums and their potential for refinement.The evolution of optimizing artworks has progressed through three stages:manual optimization,mechanical reproduction optimization,and AI optimization.During the era of manual optimization,various forms of"pseudo-optimization"emerged,including polishing and revising manuscripts,fine-tuning instruments,duplicating,transcribing,or even rewriting original works,as well as restoring cultural artifacts and archaeological sites.The era of mechanical reproduction optimization introduced categories such as"parameter optimization","editing optimization",and"performance optimization".In the digital age,AI optimization driven by algorithms has become prevalent.When incorporating optimization into literary theories,it's important to be cautious of several paradoxes.Firstly,while aiming for"optimized"results,it's essential to recognize that these outcomes may not be unique but rather commonplace.Secondly,biases present in data and variables may prove difficult to rectify and,paradoxically,can exacerbate biases,leading to prejudices.Thirdly,randomness doesn't always signify divergence but rather similarity and the recurrence of stylistic elements.

关 键 词：优化 可修正性 人工智能时代 媒介艺术史 文论问题 

分 类 号：J0-05[艺术—艺术理论] I0[文学—文学理论]