| 杨超,高文岭,常帅文.生成式人工智能技术如何影响企业全要素生产率——基于技术专利分类系统的创新测度[J].数量经济技术经济研究,2026,(6):110-136 | | 生成式人工智能技术如何影响企业全要素生产率——基于技术专利分类系统的创新测度 | | How do Generative Artificial Intelligence Technologies Affect Firms’ Total Factor Productivity： An Innovative Measure of the Technology Patent Classification System | | | | DOI： | | 中文关键词: 生成式人工智能 全要素生产率 技术专利分类 智能经济 生产力革命 | | 英文关键词: Generative Artificial Intelligence Total Factor Productivity Technical Patent Classification Intelligent Economy Productivity Revolution | | 基金项目: | | | 中文摘要: | | 在国家“人工智能+”行动战略持续推进的背景下，生成式人工智能技术持续迭代升级。本文基于《关键数字技术专利分类体系（2023）》构建了识别判别式和生成式人工智能的技术分类系统，基于2010~2023年国家知识产权局公布的上市公司专利明细数据，研究了生成式人工智能技术对企业全要素生产率的影响及作用机制。研究表明，生成式人工智能技术能够显著提升企业全要素生产率。机制分析结果表明，生成式人工智能技术通过劳动要素强化机制、成本费用优化机制和经营质量提升机制促进企业全要素生产率提升。拓展性分析发现，企业应优先保证生成式人工智能技术应用的投入，重点支持模型层和应用层技术的发展，并通过充分利用城市数字金融水平、行业基础资源能力和企业内部支撑要素，增强技术经济效应的释放。本文为优化生成式人工智能技术的经济效应并建构政策支撑体系提供了重要支持。 | | 英文摘要: | | Against the backdrop of China’s continued advancement of the “AI+” strategic initiative, generative artificial intelligence (GAI) technologies have undergone rapid iterative upgrading and are increasingly transitioning from laboratory experimentation to large scale commercial application. Compared to traditional artificial intelligence (TAI), also referred to as discriminative artificial intelligence, which primarily focuses on recognition and prediction tasks, GAI technologies, characterized by large-scale pre-trained models, generative architectures, and multimodal alignment, demonstrate stronger capabilities in content creation, reasoning, and decision support. For these reasons, they are becoming deeply embedded in firms’ innovation chains and value chains, reshaping production processes, organizational structures, and competitive dynamics. Drawing on a comprehensive panel dataset of Chinese A-share listed firms’ patent records from 2010 to 2023, this study develops a systematic technology classification framework that distinguishes between discriminative and generative AI technologies. This framework was built based on the Patent Classification System of Key Digital Technologies (2023) and further operationalized through a layered structure encompassing the infrastructure layer, the model layer, and the application layer. By leveraging patent classification codes as objective indicators of technological innovation, this study provides a more precise and replicable measurement of firms’ GAI technological capabilities and examines their impact on firms’ total factor productivity, along with the underlying transmission mechanisms.
This study has several key empirical findings. First, GAI technologies significantly enhance firms’ total factor productivity, and their positive effects are consistently stronger than those of TAI technologies, a result that holds even after addressing potential endogeneity concerns and conducting a series of robustness tests. This finding highlights the transformative nature of GAI as a general purpose technology, which extends beyond efficiency improvement to enable production paradigm shifts. Second, when comparing the three-layer structure of GAI and TAI technologies, the results show that the strongest impact on total factor productivity is exerted by the model layer, followed by the application layer and the infrastructure layer. Notably, GAI technologies outperform TAI technologies, particularly in the application and infrastructure layers, while their effects are comparable at the model layer; This suggests that the advantages of GAI are more pronounced in downstream integration and real world deployment. Third, the mechanism analysis demonstrates that GAI technologies promote total factor productivity through three primary channels, namely labor factor enhancement, cost efficiency optimization, and operational quality improvement. Specifically, GAI reduces routine labor time and enhances workforce productivity by automating standardized tasks and augmenting human decision making; Moreover, it lowers operational and coordination costs by optimizing resource allocation and reducing information asymmetry, and improves firms’ operational quality by strengthening strategic decision making, innovation output, and market responsiveness. Furthermore, the results show that both external and internal enabling conditions significantly reinforce these effects. In particular, higher levels of urban digital finance facilitate access to capital and reduce financing constraints, while stronger industry resource endowments and firm-level internal support capabilities expand the scope of GAI applications and accelerate the realization of its economic value.
This study makes several important contributions to existing literature. First, it develops a novel patent-based classification system for GAI technologies, distinguishing them from TAI technologies within the context of China’s Patent Classification system. Compared to existing approaches that rely on survey data or textual analysis of annual reports, this method provides a more objective, fine-grained, and dynamically traceable measurement of firms’ technological innovation and AI adoption. Second, by explicitly differentiating between GAI and TAI technologies and constructing a layered analytical framework, this study extends current research on the economic effects of AI, offering new insights into the heterogeneous impacts of different AI paradigms on firm productivity. Third, this study systematically identifies and empirically validates three key mechanisms, including labor factor enhancement, cost efficiency optimization, and operational quality improvement, through which GAI technologies influence total factor productivity, while also uncovering the moderating roles of digital finance development, industry resource capacity, and firm-level internal support systems. Finally, the findings of this study provide actionable policy implications, suggesting that firms should prioritize investments in GAI technologies, particularly in the model layer and the application layer, and that policymakers should promote the coordinated development of digital infrastructure, financial environment, and industrial ecosystems to fully unlock the potential of GAI to enhance productivity and foster the emergence of new forms of the intelligent economy. | | 查看全文 相关附件： 下载数据代码附录 |
| | |
