Do Societal Promises Influence Patent Value? An Analysis of Inventions in Artificial Intelligence

Abstract

This paper investigates the relationship between narratives on the societal impact of artificial intelligence (AI) and patent value, in the context of the US patenting system. We advance the construct of public value expressions (PVEs), defined as statements that describe societal problems, or solutions to these problems, as offered by the invention being patented. Our dataset comprises 154,934 USPTO AI patent documents, which we analyze using a generative language model for label generation and a discriminative language model for training and large-scale classification, complemented by 12 interviews with patent attorneys, examiners, inventors and a linguist. Implementing industry, year and assignee-level fixed-effects models and count data regression approaches, we find a significant positive relationship between the occurrence and density of PVEs and well-established indicators of patent value, including forward citations, family size, claims count, and the number of technology classes, holding key confounders constant. The study makes two contributions to the STI policy and patent value literatures. First, we demonstrate that PVEs in AI patents are associated with indicators of higher technological and commercial value. This suggests that articulations of societal relevance may play a role in enhancing a patent’s private value and reveal a paradox in which narratives of public benefit are mobilized to support processes that primarily target private interests and rewards. Second, to explain this paradox, we introduce the concept of “societal value embedding”, a patent drafting practice through which inventors and attorneys signal the potential societal benefits of their inventions, both to align with institutional expectations (normative function) and to support patentability discussions (instrumental function).