Electronic Thesis and Dissertation Repository

Thesis Format

Integrated Article

Degree

Doctor of Philosophy

Program

Business

Supervisor

Vergne, Jean-Philippe

Affiliation

University College London

Abstract

Corporate research and development (R&D) plays an important role in firm-level innovation strategies. To maintain competitive advantage, firms tend to disclose their internal research strategically. Essay 1 of this dissertation examines what motivates firms to publish more papers in Artificial Intelligence (AI). Combining two disparate literatures— R&D disclosure strategy and strategic human capital literature— I argue that scientists have a preference to publish research and when scientists have higher bargaining power, firms tend to disclose more internal research to recruit talent. To test my propositions, I use a comprehensive dataset of 200 million US job postings (from Burning Glass Technologies) and 1 million firm-level peer-reviewed publications from AI firms. Using rich qualitative data, I document that, in AI, there is a shortage of talent which increases scientists’ bargaining power. Next, I demonstrate that firms’ AI job posts lead to AI publications, which supports my proposition. This relationship is even more salient when job posts require PhD degrees, indicating that bargaining power is a key driver of increased R&D disclosure.

Interestingly, scholars have documented a secular decline of firm-level publications across many different knowledge-intensive industries. Contrary to this, in Essay 2, using a rich dataset of 171,394 papers from 57 prestigious computer science conferences, I document that in AI, firms have increased publications relative to non-AI research fields. The unexpected resurgence of AI is due to deep learning, a sub-field of AI, which requires significant computing power and large datasets. Building on the resource-based view and economics of innovation literature, I hypothesize and find that access to key resources provides competitive advantages to large technology firms and elite universities. On the other hand, the rise of deep learning creates entry barriers for non-elite universities, which are struggling to publish in top-tier conferences. Taken together, the results suggest that the rise of deep learning has resulted in a “de-democratization” of the AI research field.

Taken together, these two essays advance our understanding of corporate R&D, the role of resources in knowledge production, and firms’ publications strategy. Further, both essays have significant policy implications on increasing societal welfare.

Summary for Lay Audience

Companies need to conduct internal research to stay ahead of the competition. Sometimes companies disclose their research by filing patents or publishing peer-reviewed papers. However, peer-reviewed publication is costly and risky for firms since, unlike patents, they do not offer any intellectual rights. Therefore, scholars have provided varied and different explanations as to what motivates companies to publish papers.

In Essay 1, I give one particular motivation for why firms publish papers: to recruit talent. Using strategic human capital literature and innovation literature, I hypothesize that when scientists have higher bargaining power, firms use publications to recruit talent. I use a large dataset and advanced statistical methods to show that firms publish more AI papers if they need to hire AI talents. I also find that firms publish even more papers if their job posts have PhD requirements, indicating that PhD-holders can increase firm-level publications due to their higher bargaining power.

Scholars have documented that, in general, companies have published fewer peer-reviewed papers over the last two decades, which they call the secular decline of corporate science. In contrast to this decline, in Essay 2, I document that firms have increased publications in AI research. I also argue that large technology companies have an advantage due to their access to computing power and proprietary datasets. Similarly, elite universities have trained scientists who can do research in AI. On the other hand, non-elite universities lack access to resources like data, computing power, and trained human capital. Overall, I document that the AI research field has become a less accessible or less democratic field than before.

Taken together, these two essays advance our understanding of corporate R&D and firms’ publications strategy and have significant implications for policymakers.

Download

Share

COinS