How does AI perform in industry chain? A patent claims analysis approach

Abstract

The development trajectory of AI within the industry chain can offer valuable insights for managers and policymakers. Because the industry chain includes multiple complex nodes, it becomes difficult to showcase the subtle changes in AI at each node. Since patent claims are authoritative legal documents describing technology, we first theoretically demonstrate that integrating them with deep learning can effectively reveal the development of AI within complex nodes. And then, based on claim types and dependencies, we construct a more robust AI Recognition Multiple Attention Mechanism (A&C-Mechanism). Finally, using the battery industry chain (BIC) as a case study, the A&C-Mechanism reveals differences in AI development within the industry chain: (1) The A&C-mechanism can calculate the adjustment weights of patent claims based on variations in claim types and dependencies. Therefore, integrating the A&C-mechanism into NLP models can enhance the models' robustness and sensitivity to the nuanced variations of AI within patent claims; (2) Based on the A&C mechanism, our analysis indicates that AI indeed drives technological upgrades within four BIC nodes of mineral resource extraction (MRE), raw material processing (RMP), finished product manufacturing (FPM), usage, and recycling (UR). However, there is a phenomenon of non-uniform AI development emerging across these nodes; (3) Analyzing the patent application volume and growth rates across the four nodes, we identify that AI development progresses through distinct stages within the industrial chain: early, mid-term, and improvement. With the establishing two coefficients, the AI claim dependency variation coefficient and the AI-NE variation coefficient, we demonstrate that each stage exhibits unique characteristics. AI is used directly in the early stages. As the mid-term stage approaches, AI starts to be optimized and enhanced. During the improvement stage, AI structures, procedures, etc., are adaptively adjusted to better serve each company's goals; (4) Constructing an interaction network of AI with the four nodes based on high-frequency AI named entities within patent claims, we discover that AI development within the industrial chain exhibits iteration and continuity. Moreover, Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) remain the cornerstone, serving as the foundation upon which many cutting-edge technologies are built. Digital image processing and machine learning enhance problem-solving across multiple nodes. We discuss our findings and derive implications for research, managers and policymakers.