Software Patents as a Barrier to Scientific Transparency: An Unexpected Consequence of Bayh-Dole

Abstract

Computing has taken a central role in the practice of science. From the ability to capture data, methods, create simulations, and provide dissemination mechanisms, science has gone digital. In this paper we examine the interaction between the digitization of science and Intellectual Property Law, specifically the incentives created by the Bayh-Dole Act to patent inventions associated with university-based research. We show that the number of software patents granted to faculty researchers has more than doubled over the last ten years among top patenting universities and colleges. In a traditional scientific setting methods are usually openly shared in the methods section of a scientific publication, but due to increased levels of complexity and detail deep intellectual contributions to science are now being captured only in the software and codes that generate published computational results. In computational science reproducibility of results can typically only be effected with knowledge of the underlying code and data – the traditional methods section is insufficient for computational science. Because of the ability to patent software, incentives to patent academic code are potentially at odds with scientific norms of transparency and reproducibility.We collect a novel dataset to understand the rise in academic software patenting and its potential for occluding reproducible research – where the code and data underlying published results is openly shared. We implement a predictive model and find preliminary evidence that the increasing number of academic software patents is associated with increasing levels of industry collaboration (defined as at least one industry-based patent author) in patentable research, but there a weakly negative association with published articles (defined as those cited in the patent application). This extends a result in the literature to software patents: that patents arise more frequently from collaborations with industry, but does not confirm a second result, that positive reinforcement effect on publication by faculty who seek patents. From the perspective of reproducible computational research, this indicates that prior publications are either not be made if there is an intent to patent, or not be listed on the patent application. These findings indicate that the question of whether code needed to reproducible computational results is being patented and licensed is still an open one.