Essays on Open Science and Open Software in Firm Innovation

Abstract

Why do firms publish science and release open-source software for free? The classical view in economics argues that firms will underproduce innovation relative to the social optimum because they cannot capture the social value available in cumulative ideas. And yet, firms do freely share their innovative knowledge, do so frequently, and do so with dramatic effect (with examples ranging in time from the transistor to the transformer architecture). Why? An influential perspective in the literature is that firms do capture value from this strategy, but they do so elsewhere – controlling the complement, whether through vertical innovation, co-invention, or the presence of unique complementary skills. Yet, while this perspective has impressive explanatory power across a wide range of industries, it fails to explain recent developments in artificial intelligence (AI) technologies. This dissertation therefore empirically examines this innovation strategy in the context of firm involvement in AI research, in search of new explanations. The first chapter of this dissertation observationally studies the impact of firm involvement on foundational scientific research, and, using a special set of “dual-affiliated” researchers located on the boundary of universities and firms, shows that firms can surprisingly have a positive effect on the scientific impact of a researcher’s papers. The second chapter uses a field experiment to estimate the impact of open source software libraries on the productivity of data scientists and illustrate the types of skills necessary for realizing those productivity gains. The final chapter uses the unexpected shift in governance of PyTorch (a leading machine learning framework) from Meta to the Linux Foundation in order to illustrate the importance of technology control rights in understanding why firms get involved in open-source software development. A key theme that emerges from these studies is that open firm innovation at scale (due to digitization) can lead to new possibilities for value capture that can align with value creation.