人工智能技术应用赋能企业突破性创新影响机制研究

谢卫红^{a, b}，关千浩^a，李忠顺^a,b

（广东工业大学 a.经济学院；b.数字经济与数据治理重点实验室，广东广州 510520）

摘要：人工智能作为引领新一轮科技革命和产业变革的战略性技术，已成为驱动企业突破性创新增长的核心引擎。本文基于2015—2024年中国沪深A股上市制造业企业数据，运用神经网络模型等机器学习方法，实证检验人工智能技术应用对企业突破性创新的影响。研究结果表明：无论是文本型、交互型、功能型、分析型还是视觉型人工智能技术应用，均能促进企业突破性创新，其中，文本型人工智能技术应用对制造业企业突破性创新作用效果更明显，且在经过一系列内生性处理与稳健性检验后结论依然成立。人工智能技术应用通过优化企业内部劳动力技能结构、增加知识多样性以及提高管理效率促进企业突破性创新。进一步研究发现，高管风险治理能力在人工智能技术应用赋能企业突破性创新的过程中发挥正向调节作用。同时，相较于环境规制风险治理能力，隐私规制风险治理能力对人工智能技术赋能企业突破性创新激励效果更明显。异质性分析结果表明，在非国有企业、算力基础设施越强的地区内企业、供应链韧性水平越强的企业中，人工智能技术赋能企业突破性创新产出效果越明显。本文的研究为探索人工智能技术应用赋能企业突破性创新作用机制以及高管风险治理能力在其中发挥的重要作用提供了新的经验证据。

基金项目：国家社会科学基金重大项目“人工智能对制造业转型升级的影响与治理体系研究”(23&ZD090)

关键词：人工智能技术应用；突破性创新；高管风险治理能力；知识多样性；劳动力技能结构

作者简介：谢卫红，广东工业大学经济学院教授、经济与数据治理重点实验室执行主任；关千浩，广东工业大学经济学院硕士研究生，通信作者；李忠顺，广东工业大学经济学院讲师、数字经济与数据治理重点实验室助理研究员。

引用格式：谢卫红,关千浩,李忠顺.人工智能技术应用赋能企业突破性创新影响机制研究[J].首都经济贸易大学学报,2026,28(2):127-144.

The Impact Mechanism of AI Application Empowering Enterprise Breakthrough Innovation

XIE Weihong, GUAN Qianhao, LI Zhongshun

(Guangdong University of Technology, Guangzhou 510520)

Abstract: This paper examines how artificial intelligence (AI) applications foster breakthrough innovation in Chinese manufacturing firms in the context of China's "AI+" initiative and the policy emphasis on moving AI from "1 to N" in large-scale industrial deployment. Although prior studies link AI to productivity, labor reallocation, and supply-chain outcomes, three issues limit cumulative knowledge on breakthrough innovation. First, many empirical proxies (AI patents or industrial-robot penetration) conflate AI "inputs" with firm-level AI applications, and cannot distinguish heterogeneity across application types. Second, breakthrough innovation is often measured using invention patent counts or citations, which primarily reflect innovation volume and may misclassify incremental inventions as breakthroughs. Third, the moderating role of executive risk governance capability remains inconclusive, partly because it is difficult to operationalize and quantify. Clarifying these issues matters for understanding how manufacturing firms transition from innovation "quantity accumulation" to "quality leapfrogging", while maintaining compliance and managing privacy, ethical, and regulatory risks.

To address these challenges, this paper assembles an unbalanced panel of China's A-share listed manufacturing firms from 2015 to 2024 and develops two key measures. This paper quantifies AI applications via annual-report text analysis rather than patents, robots, or surveys. After standard text preprocessing, this paper expands an AI keyword dictionary using Word2Vec with reference to Stanford HAI's The 2025 AI Index Report, then classifies AI applications into five types—textual, interactive, functional, analytical, and visual AI—and measures each as log (keyword frequency + 1). This paper measures breakthrough innovation more strictly using a patent-citation network approach.

This paper finds that AI applications significantly increase breakthrough innovation, and the effect is positive across all five AI types, with textual AI exhibiting the largest marginal impact. Mechanism evidence indicates that AI promotes breakthroughs by optimizing labor skill structure, enhancing knowledge diversity, and improving managerial efficiency. Executive risk governance capability positively moderates the AI application-innovation relationship, and privacy regulatory risk governance has a stronger moderating effect than environmental regulatory risk governance. The positive effects are stronger for non-state-owned enterprises, firms in regions with stronger computing infrastructure, firms with higher supply-chain resilience, and capital-intensive firms.

Based on these findings, this paper recommends differentiated "AI+" guidance by application type, parallel investments in workforce upskilling and human-AI collaboration, platforms for cross-domain knowledge sharing to sustain diversity and recombination, and deeper AI-enabled process reengineering to raise managerial efficiency. This paper suggests institutionalizing executive risk governance, especially privacy and data governance, to provide the compliance and data foundations necessary for continuous AI iteration and sustained breakthrough innovation.

Keywords: AI application; breakthrough innovation; executive risk governance capability; knowledge diversity; labor skill structure

下载全文