人工智能深度学习平台如何实现开源式创新

How Can Artificial Intelligence Deep Learning Platform Achieve Open SourceInnovation

开源式创新正成为人工智能产业发展的主流趋势,尤其在开源起步较晚的新兴经济体情境中,人工智能深度学习平台如何实现开源式创新值得深入探索。本文聚焦中国人工智能深度学习平台实现开源式创新的过程,通过对中国开源式创新的领先者一一百度飞桨深度学习平台开展纵向单案例研究,构建了“驱动逻辑一→行动模式一→开源机制一→开源结果”的开源式创新过程的整合性逻辑框架。研究发现:人工智能深度学习平台实现开源式创新是一个由积势蓄能期、混合流变期和整合成熟期三阶段构成的演进过程,分别经历了“竞争优势驱动”“高效选代驱动”和“价值最优驱动”三种驱动逻辑的动态更送。受不同驱动逻辑引导,开源式创新行动者依次呈现出三种差异化的行动模式:“工具型行动模式”“螺旋型行动模式”和“生态型行动模式”。上述过程中的开源式创新机制,既包括位于阶段内的“惯例变革机制”“自适应聚变机制”和“聚合共生机制”三种联动机制,又包括位于阶段间的“开源社区化演变机制”和“开源商业化演变机制”两种演变机制,由此分别形成了“开源流程协同”“开源范式转换”和“开源系统共创”三种开源结果,实现了由低阶到高阶的开源式创新的动态演进。综上,基于中国为代表的新兴经济体的后发开源情境,本文系统构建了人工智能深度学习平台实现开源式创新的过程理论模型,挖掘了适用于该情境的预设性开源式创新的过程、模式和机制,进而比较该种预设性开源式创新与非预设性开源式创新的异同点,从而丰富了开源式创新的理论探索,同时提供了人工智能深度学习平台开源建设的实践启示。

Open source innovation is becoming the mainstream trend for artificial intelligence(Al)deep learning platforms,particularly in emerging economies where open source practices have started relatively later.It is worth exploring how AI deep learning platforms can achieve open source innovation under such circumstances.Existing literature on open source innovation has formed three main schools:process-based view,product-based view,and structure-based view.However,these studies have not focused on constructing process theories specific to open source innovation in AI deep learning platforms.Moreover,the implicit assumptions in current research do not apply to the Chinese context.The reason is that existing research,primarily based on developed economies,assumes that open source innovation is a spontaneous,uncontrolled,and non-preset activity driven by developers.These underlying assumptions fail to explain the success of another type of open source innovation in China's practices,namely,the“preset open source innovation”initiated by AI enterprises.This study focuses on achieving open source innovation in AI deep learning platforms in China.Based on a longitudinal single case study of the open source leaderBaidu PaddlePaddle deep learning platform,this study follows the process research paradigm and constructs an integrated logical framework of“driving logic-action mode-open source mechanism-open source output”.The findings are as follows.First,the open source innovation process of AI deep learning platforms is seen as an evolutionary process consisting of three stages:the momentum accumulation stage,the intense change stage,and the mature integration stage.They undergo a dynamic change of three driving logics:“competitive advantage”,“efficient iteration”and“value optimization”.Second,driven by different driving logics,the actors involved in open source innovation exhibit three distinct action patterns sequentially:“instrumental mode”,“spiral mode”and“ecological mode”.Third,the open source innovation mechanisms in the above process include both the mechanisms within the stage(“linkage mechanisms”)and between the stages(“evolution mechanisms”).The linkage mechanisms include the“routine transformation mechanism”,“self-adaptation fusion mechanism”and“polymerization symbiosis mechanism”,while the evolution mechanisms include the“communitization evolution mechanism”and“commercialization evolution mechanism”.Fourth,the above process results in three separate open source results:“open source process collaboration”,“open source paradigm shift”and“open source systems co-creation”,realizing the evolution of open source capabilities from low to high levels.Then,this study compares the differences between preset open source innovation and non-preset opensource innovation.The theoretical contributions of this study mainly include the following two aspects.First,it constructs a process theory model for open source innovation in Al deep learning platforms from a dynamic perspective.It focuses on the progressive evolution of open source innovation in Al deep learning platforms over an extended time span,revealing the evolutionary patterns and underlying mechanisms of open source innovation based on a process-oriented view.Second,this study uncovers an alternative form of open source innovation,preset open source innovation,under the latecomer open source context of emerging economies.This broadens the fundamental assumptions and theoretical boundaries of open source innovation and offers practical insights for the open source development of AI deep learning platforms.